Renamed libs

Cleaner and more flexible.

64 files changed, 14133 insertions, 0 deletions

diff --git a/libs/libc/Makefile b/libs/libc/Makefile
new file mode 100644
index 0000000..1c66f43
--- /dev/null
+++ b/libs/libc/Makefile
@@ -0,0 +1,42 @@
+# MIT License, Copyright (c) 2020 Marvin Borner
+
+# TODO: Remove cpu from libc?
+COBJS = sanitize.o \
+	errno.o \
+	str.o \
+	alloc.o \
+	mem.o \
+	math.o \
+	crypto.o \
+	conv.o \
+	print.o \
+	cpu.o \
+	sys.o \
+	list.o \
+	stack.o \
+	random.o
+CC = ccache ../../cross/opt/bin/i686-elf-gcc
+LD = ccache ../../cross/opt/bin/i686-elf-ld
+AR = ccache ../../cross/opt/bin/i686-elf-ar
+AS = ccache nasm
+
+CFLAGS = $(CFLAGS_DEFAULT) -Iinc/
+
+ASFLAGS = -f elf32
+
+%.o: %.c
+	@$(CC) -c $(CFLAGS) $< -o $@
+
+libc: CFLAGS += -Duserspace
+libc: $(COBJS)
+	@$(AS) $(ASFLAGS) crt/crt0.asm -o crt0.o
+	@mkdir -p ../../build/
+	@$(AR) rcs ../../build/libc.a crt0.o $+
+
+libk: CFLAGS += -Dkernel -ffreestanding -I../../kernel/inc/ $(CFLAGS_EXTRA)
+libk: $(COBJS)
+	@mkdir -p ../../build/
+	@$(AR) rcs ../../build/libk.a $+
+
+clean:
+	@find . -name "*.o" -type f -delete
diff --git a/libs/libc/alloc.c b/libs/libc/alloc.c
new file mode 100644
index 0000000..b8139d0
--- /dev/null
+++ b/libs/libc/alloc.c
@@ -0,0 +1,425 @@
+// MIT License, Copyright (c) 2021 Marvin Borner
+// Mostly by Durand Miller, released into public domain
+
+#include <assert.h>
+#include <cpu.h>
+#include <mem.h>
+
+#ifdef kernel
+
+#include <mm.h>
+
+static void *liballoc_alloc(u32 p)
+{
+	return memory_alloc(virtual_kernel_dir(), p, MEMORY_CLEAR);
+}
+
+static int liballoc_free(void *ptr, u32 p)
+{
+	memory_free(virtual_kernel_dir(), memory_range((u32)ptr, (u32)p));
+	return 0;
+}
+
+#else
+
+#include <sys.h>
+
+static void *liballoc_alloc(u32 p)
+{
+	u32 addr;
+	assert(sys_alloc(p, &addr) == EOK);
+	return (void *)addr;
+}
+
+static int liballoc_free(void *ptr, u32 p)
+{
+	UNUSED(p);
+	assert(sys_free(ptr) == EOK);
+	return 0;
+}
+
+#endif
+
+static int locked = 0;
+
+static int liballoc_lock(void)
+{
+	spinlock(&locked);
+	return 0;
+}
+
+static int liballoc_unlock(void)
+{
+	locked = 0;
+	return 0;
+}
+
+#define ALIGNMENT 16
+#define ALIGN_UP(__addr, __align) (((__addr) + (__align)-1) & ~((__align)-1))
+#define ALIGN_DOWN(__addr, __align) ((__addr) & ~((__align)-1))
+
+#define USE_CASE1
+#define USE_CASE2
+#define USE_CASE3
+#define USE_CASE4
+#define USE_CASE5
+#define LIBALLOC_MAGIC 0x900df00d
+#define LIBALLOC_DEAD 0xbaadf00d
+
+struct liballoc_major {
+	struct liballoc_major *prev;
+	struct liballoc_major *next;
+	u32 pages;
+	u32 size;
+	u32 usage;
+	struct liballoc_minor *first;
+};
+
+struct liballoc_minor {
+	struct liballoc_minor *prev;
+	struct liballoc_minor *next;
+	struct liballoc_major *block;
+	u32 magic;
+	u32 size;
+	u32 req_size;
+};
+
+#define MAJOR_SIZE (ALIGN_UP(sizeof(struct liballoc_major), 16))
+#define MINOR_SIZE (ALIGN_UP(sizeof(struct liballoc_minor), 16))
+
+static struct liballoc_major *l_mem_root = NULL;
+static struct liballoc_major *l_best_bet = NULL;
+
+static u32 l_page_size = 4096;
+static u32 l_page_count = 16;
+
+static struct liballoc_major *allocate_new_page(u32 size)
+{
+	u32 st = size + MAJOR_SIZE + MINOR_SIZE;
+
+	if ((st % l_page_size) == 0)
+		st = st / (l_page_size);
+	else
+		st = st / (l_page_size) + 1;
+
+	st = MAX(st, l_page_count);
+
+	struct liballoc_major *maj = (struct liballoc_major *)liballoc_alloc(st * l_page_size);
+
+	if (maj == NULL)
+		return NULL;
+
+	maj->prev = NULL;
+	maj->next = NULL;
+	maj->pages = st;
+	maj->size = st * l_page_size;
+	maj->usage = MAJOR_SIZE;
+	maj->first = NULL;
+
+	return maj;
+}
+
+static void *_malloc(u32 req_size)
+{
+	req_size = ALIGN_UP(req_size, 16);
+
+	u32 best_size = 0;
+	u32 size = req_size;
+
+	liballoc_lock();
+
+	if (size == 0) {
+		liballoc_unlock();
+		return malloc(1);
+	}
+
+	if (l_mem_root == NULL) {
+		l_mem_root = allocate_new_page(size);
+		if (l_mem_root == NULL) {
+			liballoc_unlock();
+			panic("Malloc failed!\n");
+		}
+	}
+
+	struct liballoc_major *maj = l_mem_root;
+	u8 started_bet = 0;
+
+	if (l_best_bet != NULL) {
+		best_size = l_best_bet->size - l_best_bet->usage;
+
+		if (best_size > (size + MINOR_SIZE)) {
+			maj = l_best_bet;
+			started_bet = 1;
+		}
+	}
+
+	while (maj != NULL) {
+		u32 diff = maj->size - maj->usage;
+		if (best_size < diff) {
+			l_best_bet = maj;
+			best_size = diff;
+		}
+
+#ifdef USE_CASE1
+		if (diff < (size + MINOR_SIZE)) {
+			if (maj->next != NULL) {
+				maj = maj->next;
+				continue;
+			}
+
+			if (started_bet == 1) {
+				maj = l_mem_root;
+				started_bet = 0;
+				continue;
+			}
+
+			maj->next = allocate_new_page(size);
+			if (maj->next == NULL)
+				break;
+			maj->next->prev = maj;
+			maj = maj->next;
+		}
+#endif
+
+#ifdef USE_CASE2
+		if (maj->first == NULL) {
+			maj->first = (struct liballoc_minor *)((u32)maj + MAJOR_SIZE);
+
+			maj->first->magic = LIBALLOC_MAGIC;
+			maj->first->prev = NULL;
+			maj->first->next = NULL;
+			maj->first->block = maj;
+			maj->first->size = size;
+			maj->first->req_size = req_size;
+			maj->usage += size + MINOR_SIZE;
+			void *p = (void *)((u32)(maj->first) + MINOR_SIZE);
+			liballoc_unlock();
+			return p;
+		}
+#endif
+
+#ifdef USE_CASE3
+		diff = (u32)(maj->first);
+		diff -= (u32)maj;
+		diff -= MAJOR_SIZE;
+
+		if (diff >= (size + MINOR_SIZE)) {
+			maj->first->prev = (struct liballoc_minor *)((u32)maj + MAJOR_SIZE);
+			maj->first->prev->next = maj->first;
+			maj->first = maj->first->prev;
+			maj->first->magic = LIBALLOC_MAGIC;
+			maj->first->prev = NULL;
+			maj->first->block = maj;
+			maj->first->size = size;
+			maj->first->req_size = req_size;
+			maj->usage += size + MINOR_SIZE;
+			void *p = (void *)((u32)(maj->first) + MINOR_SIZE);
+			liballoc_unlock();
+			return p;
+		}
+#endif
+
+#ifdef USE_CASE4
+		struct liballoc_minor *min = maj->first;
+
+		while (min != NULL) {
+			if (min->next == NULL) {
+				diff = (u32)(maj) + maj->size;
+				diff -= (u32)min;
+				diff -= MINOR_SIZE;
+				diff -= min->size;
+				if (diff >= (size + MINOR_SIZE)) {
+					min->next =
+						(struct liballoc_minor *)((u32)min + MINOR_SIZE +
+									  min->size);
+					min->next->prev = min;
+					min = min->next;
+					min->next = NULL;
+					min->magic = LIBALLOC_MAGIC;
+					min->block = maj;
+					min->size = size;
+					min->req_size = req_size;
+					maj->usage += size + MINOR_SIZE;
+					void *p = (void *)((u32)min + MINOR_SIZE);
+					liballoc_unlock();
+					return p;
+				}
+			}
+
+			if (min->next != NULL) {
+				diff = (u32)(min->next);
+				diff -= (u32)min;
+				diff -= MINOR_SIZE;
+				diff -= min->size;
+
+				if (diff >= (size + MINOR_SIZE)) {
+					struct liballoc_minor *new_min =
+						(struct liballoc_minor *)((u32)min + MINOR_SIZE +
+									  min->size);
+					new_min->magic = LIBALLOC_MAGIC;
+					new_min->next = min->next;
+					new_min->prev = min;
+					new_min->size = size;
+					new_min->req_size = req_size;
+					new_min->block = maj;
+					min->next->prev = new_min;
+					min->next = new_min;
+					maj->usage += size + MINOR_SIZE;
+					void *p = (void *)((u32)new_min + MINOR_SIZE);
+					liballoc_unlock();
+					return p;
+				}
+			}
+
+			min = min->next;
+		}
+#endif
+
+#ifdef USE_CASE5
+		if (maj->next == NULL) {
+			if (started_bet == 1) {
+				maj = l_mem_root;
+				started_bet = 0;
+				continue;
+			}
+			maj->next = allocate_new_page(size);
+			if (maj->next == NULL)
+				break;
+			maj->next->prev = maj;
+		}
+#endif
+		maj = maj->next;
+	}
+
+	liballoc_unlock();
+
+	panic("Malloc failed!\n");
+	return NULL;
+}
+
+static void _free(void *ptr)
+{
+	if (ptr == NULL) {
+		return;
+	}
+
+	liballoc_lock();
+
+	struct liballoc_minor *min = (struct liballoc_minor *)((u32)ptr - MINOR_SIZE);
+
+	if (min->magic != LIBALLOC_MAGIC) {
+		liballoc_unlock();
+		return;
+	}
+
+	struct liballoc_major *maj = min->block;
+	maj->usage -= (min->size + MINOR_SIZE);
+	min->magic = LIBALLOC_DEAD;
+
+	if (min->next != NULL)
+		min->next->prev = min->prev;
+	if (min->prev != NULL)
+		min->prev->next = min->next;
+	if (min->prev == NULL)
+		maj->first = min->next;
+	if (maj->first == NULL) {
+		if (l_mem_root == maj)
+			l_mem_root = maj->next;
+		if (l_best_bet == maj)
+			l_best_bet = NULL;
+		if (maj->prev != NULL)
+			maj->prev->next = maj->next;
+		if (maj->next != NULL)
+			maj->next->prev = maj->prev;
+		liballoc_free(maj, maj->pages * l_page_size);
+	} else {
+		if (l_best_bet != NULL) {
+			int best_size = l_best_bet->size - l_best_bet->usage;
+			int maj_size = maj->size - maj->usage;
+			if (maj_size > best_size)
+				l_best_bet = maj;
+		}
+	}
+	liballoc_unlock();
+}
+
+static void *_realloc(void *ptr, u32 size)
+{
+	size = ALIGN_UP(size, 16);
+
+	if (size == 0) {
+		free(ptr);
+		return NULL;
+	}
+
+	if (ptr == NULL)
+		return malloc(size);
+
+	liballoc_lock();
+	struct liballoc_minor *min = (struct liballoc_minor *)((u32)ptr - MINOR_SIZE);
+
+	if (min->magic != LIBALLOC_MAGIC) {
+		liballoc_unlock();
+		panic("Malloc failed!\n");
+		return NULL;
+	}
+
+	if (min->size >= size) {
+		min->req_size = size;
+		liballoc_unlock();
+		return ptr;
+	}
+
+	liballoc_unlock();
+
+	void *new_ptr = malloc(size);
+	memcpy(new_ptr, ptr, min->req_size);
+	free(ptr);
+
+	return new_ptr;
+}
+
+#ifdef kernel
+#define PREFIX "K"
+#define FUNC printf
+#else
+#define PREFIX "U"
+#define FUNC log
+#endif
+
+void *zalloc(u32 size)
+{
+	void *ret = malloc(size);
+	memset(ret, 0, size);
+	return ret;
+}
+
+void *realloc(void *ptr, u32 size)
+{
+	return _realloc(ptr, size);
+}
+
+void *malloc_debug(u32 size, const char *file, int line, const char *func, const char *inp)
+{
+	assert(size < (100 << 20)); // Don't brag with memory pls
+	void *ret = _malloc(size);
+
+	(void)file;
+	(void)line;
+	(void)func;
+	(void)inp;
+	/* FUNC(PREFIX "MALLOC\t%s:%d: %s: 0x%x %dB (%s)\n", file, line, func, ret, size, inp); */
+	return ret;
+}
+
+void free_debug(void *ptr, const char *file, int line, const char *func, const char *inp)
+{
+	if (ptr)
+		_free(ptr);
+
+	(void)file;
+	(void)line;
+	(void)func;
+	(void)inp;
+	/* FUNC(PREFIX "FREE\t%s:%d: %s: 0x%x (%s)\n", file, line, func, ptr, inp); */
+}
diff --git a/libs/libc/conv.c b/libs/libc/conv.c
new file mode 100644
index 0000000..670fdb3
--- /dev/null
+++ b/libs/libc/conv.c
@@ -0,0 +1,145 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#include <conv.h>
+#include <def.h>
+#include <math.h>
+#include <mem.h>
+#include <str.h>
+
+static const char HTOA_TABLE[] = "0123456789ABCDEF";
+static const char ITOA_TABLE[] = "0123456789";
+
+int atoi(const char *str)
+{
+	u32 s_str = strlen(str);
+	if (!s_str)
+		return 0;
+
+	u8 negative = 0;
+	if (str[0] == '-')
+		negative = 1;
+
+	u32 i = 0;
+	if (negative)
+		i++;
+
+	int ret = 0;
+	for (; i < s_str; i++) {
+		ret += (str[i] - '0') * pow(10, (int)((s_str - i) - 1));
+	}
+
+	if (negative)
+		ret *= -1;
+	return ret;
+}
+
+char *htoa(u32 n)
+{
+	char *ret = (char *)malloc(10);
+
+	int i = 0;
+	while (n) {
+		ret[i++] = HTOA_TABLE[n & 0xF];
+		n >>= 4;
+	}
+
+	if (!i) {
+		ret[0] = '0';
+		i++;
+	}
+
+	for (; i <= 9; i++)
+		ret[i] = 0;
+
+	char *aux = strdup(ret);
+	free(ret);
+	ret = aux;
+
+	strinv(ret);
+	return ret;
+}
+
+int htoi(const char *str)
+{
+	u32 s_str = strlen(str);
+
+	u32 i = 0;
+	int ret = 0;
+	for (; i < s_str; i++) {
+		char c = str[i];
+		int aux = 0;
+		if (c >= '0' && c <= '9')
+			aux = c - '0';
+		else if (c >= 'A' && c <= 'F')
+			aux = (c - 'A') + 10;
+
+		ret += aux * pow(16, (int)((s_str - i) - 1));
+	}
+
+	return ret;
+}
+
+char *itoa(int n)
+{
+	if (!n) {
+		char *ret = (char *)malloc(2);
+		ret[0] = '0';
+		ret[1] = 0;
+		return ret;
+	}
+	u8 negative = (u8)(n < 0);
+	if (negative)
+		n *= -1;
+
+	int sz;
+	for (sz = 0; n % pow(10, sz) != n; sz++) {
+	}
+
+	char *ret = (char *)malloc((u32)(sz + 1));
+
+	for (int i = 0; i < sz; i++) {
+		int digit = (n % pow(10, i + 1)) / pow(10, i);
+		ret[i] = ITOA_TABLE[digit];
+	}
+	ret[sz] = 0;
+
+	if (negative) {
+		char *aux = (char *)malloc((u32)(sz + 2));
+		strcpy(aux, ret);
+		aux[sz] = '-';
+		aux[sz + 1] = 0;
+		free(ret);
+		ret = aux;
+	}
+
+	strinv(ret);
+	return ret;
+}
+
+char *conv_base(int value, char *result, int base, int is_signed)
+{
+	if (base < 2 || base > 36) {
+		*result = '\0';
+		return result;
+	}
+
+	char *ptr = result, *ptr1 = result, tmp_char;
+	int tmp_value;
+
+	do {
+		tmp_value = value;
+		value /= base;
+		*ptr++ = "zyxwvutsrqponmlkjihgfedcba9876543210123456789abcdefghijklmnopqrstuvwxyz"
+			[35 + (tmp_value - value * base)];
+	} while (value);
+
+	if (is_signed && tmp_value < 0)
+		*ptr++ = '-';
+	*ptr-- = '\0';
+	while (ptr1 < ptr) {
+		tmp_char = *ptr;
+		*ptr-- = *ptr1;
+		*ptr1++ = tmp_char;
+	}
+	return result;
+}
diff --git a/libs/libc/cpu.c b/libs/libc/cpu.c
new file mode 100644
index 0000000..8ca4d27
--- /dev/null
+++ b/libs/libc/cpu.c
@@ -0,0 +1,182 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+// This file is a wrapper around some CPU asm calls
+
+#include <cpu.h>
+#include <def.h>
+#include <print.h>
+
+u8 inb(u16 port)
+{
+	u8 value;
+	__asm__ volatile("inb %1, %0" : "=a"(value) : "Nd"(port));
+	return value;
+}
+
+u16 inw(u16 port)
+{
+	u16 value;
+	__asm__ volatile("inw %1, %0" : "=a"(value) : "Nd"(port));
+	return value;
+}
+
+u32 inl(u16 port)
+{
+	u32 value;
+	__asm__ volatile("inl %1, %0" : "=a"(value) : "Nd"(port));
+	return value;
+}
+
+void insl(u16 port, void *addr, int n)
+{
+	__asm__ volatile("rep insl" ::"c"(n), // Count
+			 "d"(port), // Port #
+			 "D"(addr)); // Buffer
+}
+
+void outb(u16 port, u8 data)
+{
+	__asm__ volatile("outb %0, %1" ::"a"(data), "Nd"(port));
+}
+
+void outw(u16 port, u16 data)
+{
+	__asm__ volatile("outw %0, %1" ::"a"(data), "Nd"(port));
+}
+
+void outl(u16 port, u32 data)
+{
+	__asm__ volatile("outl %0, %1" ::"a"(data), "Nd"(port));
+}
+
+#ifdef kernel
+
+static void cpuid(int code, u32 *a, u32 *b, u32 *c, u32 *d)
+{
+	__asm__ volatile("cpuid" : "=a"(*a), "=b"(*b), "=c"(*c), "=d"(*d) : "a"(code));
+}
+
+static char *cpu_string(char buf[16])
+{
+	// wtf
+	cpuid(CPUID_VENDOR_STRING, (u32 *)(buf + 12), (u32 *)(buf), (u32 *)(buf + 8),
+	      (u32 *)(buf + 4));
+
+	return buf;
+}
+
+void cpu_print(void)
+{
+	char buf[16] = { 0 };
+	printf("CPU vendor: %s\n", cpu_string(buf));
+}
+
+u32 cr0_get(void)
+{
+	u32 cr0;
+	__asm__ volatile("movl %%cr0, %%eax" : "=a"(cr0));
+	return cr0;
+}
+
+void cr0_set(u32 cr0)
+{
+	__asm__ volatile("movl %%eax, %%cr0" ::"a"(cr0));
+}
+
+u32 cr3_get(void)
+{
+	u32 cr3;
+	__asm__ volatile("movl %%cr0, %%eax" : "=a"(cr3));
+	return cr3;
+}
+
+void cr3_set(u32 cr3)
+{
+	__asm__ volatile("movl %%eax, %%cr3" ::"a"(cr3));
+}
+
+u32 cr4_get(void)
+{
+	u32 cr4;
+	__asm__ volatile("movl %%cr4, %%eax" : "=a"(cr4));
+	return cr4;
+}
+
+void cr4_set(u32 cr4)
+{
+	__asm__ volatile("movl %%eax, %%cr4" ::"a"(cr4));
+}
+
+static u32 cpu_cfeatures = 0;
+u8 cpu_has_cfeature(enum cpuid_features feature)
+{
+	return (cpu_cfeatures & feature) != 0;
+}
+
+static u32 cpu_dfeatures = 0;
+u8 cpu_has_dfeature(enum cpuid_features feature)
+{
+	return (cpu_dfeatures & feature) != 0;
+}
+
+static void fpu_handler(struct regs *r)
+{
+	UNUSED(r);
+	__asm__ volatile("clts");
+}
+
+static u8 fpu_state[512] __attribute__((aligned(16)));
+void fpu_restore(void)
+{
+	__asm__ volatile("fxrstor (%0)" ::"r"(fpu_state));
+}
+
+void cpu_enable_features(void)
+{
+	u32 a, b, c, d;
+	cpuid(CPUID_FEATURES, &a, &b, &c, &d);
+	cpu_cfeatures = c;
+	cpu_dfeatures = d;
+	if (cpu_has_dfeature(CPUID_FEAT_EDX_SSE)) {
+		cr0_set(cr0_get() & ~(1 << 2));
+		cr0_set(cr0_get() | (1 << 1));
+		cr4_set(cr4_get() | (3 << 9));
+	} else {
+		panic("No SSE support!\n");
+	}
+
+	if (cpu_has_dfeature(CPUID_FEAT_EDX_FPU)) {
+		__asm__ volatile("fninit");
+		__asm__ volatile("fxsave %0" : "=m"(fpu_state));
+		irq_install_handler(7, fpu_handler);
+	} else {
+		panic("No FPU support!\n");
+	}
+}
+
+void cli(void)
+{
+	__asm__ volatile("cli");
+}
+
+void sti(void)
+{
+	__asm__ volatile("sti");
+}
+
+void hlt(void)
+{
+	__asm__ volatile("hlt");
+}
+
+void idle(void)
+{
+	while (1)
+		hlt();
+}
+
+void loop(void)
+{
+	cli();
+	idle();
+}
+#endif
diff --git a/libs/libc/crt/crt0.asm b/libs/libc/crt/crt0.asm
new file mode 100644
index 0000000..e002952
--- /dev/null
+++ b/libs/libc/crt/crt0.asm
@@ -0,0 +1,15 @@
+; MIT License, Copyright (c) 2020 Marvin Borner
+
+section .text
+
+extern main
+extern exit
+extern atexit_trigger
+
+global _start
+_start:
+	call main
+
+	push eax
+	call exit
+	jmp $
diff --git a/libs/libc/crypto.c b/libs/libc/crypto.c
new file mode 100644
index 0000000..310931d
--- /dev/null
+++ b/libs/libc/crypto.c
@@ -0,0 +1,166 @@
+// MIT License, Copyright (c) 2021 Marvin Borner
+
+#include <crypto.h>
+#include <def.h>
+#include <mem.h>
+
+/**
+ * MD5
+ */
+
+#define MD5_LEFTROTATE(x, c) (((x) << (c)) | ((x) >> (32 - (c))))
+
+static const u32 md5_tab[64] = {
+	0xd76aa478, 0xe8c7b756, 0x242070db, 0xc1bdceee, 0xf57c0faf, 0x4787c62a, 0xa8304613,
+	0xfd469501, 0x698098d8, 0x8b44f7af, 0xffff5bb1, 0x895cd7be, 0x6b901122, 0xfd987193,
+	0xa679438e, 0x49b40821, 0xf61e2562, 0xc040b340, 0x265e5a51, 0xe9b6c7aa, 0xd62f105d,
+	0x02441453, 0xd8a1e681, 0xe7d3fbc8, 0x21e1cde6, 0xc33707d6, 0xf4d50d87, 0x455a14ed,
+	0xa9e3e905, 0xfcefa3f8, 0x676f02d9, 0x8d2a4c8a, 0xfffa3942, 0x8771f681, 0x6d9d6122,
+	0xfde5380c, 0xa4beea44, 0x4bdecfa9, 0xf6bb4b60, 0xbebfbc70, 0x289b7ec6, 0xeaa127fa,
+	0xd4ef3085, 0x04881d05, 0xd9d4d039, 0xe6db99e5, 0x1fa27cf8, 0xc4ac5665, 0xf4292244,
+	0x432aff97, 0xab9423a7, 0xfc93a039, 0x655b59c3, 0x8f0ccc92, 0xffeff47d, 0x85845dd1,
+	0x6fa87e4f, 0xfe2ce6e0, 0xa3014314, 0x4e0811a1, 0xf7537e82, 0xbd3af235, 0x2ad7d2bb,
+	0xeb86d391,
+};
+
+static const u32 md5_rounds[] = {
+	7,  12, 17, 22, 7,  12, 17, 22, 7,  12, 17, 22, 7,  12, 17, 22, 5,  9,	14, 20, 5,  9,
+	14, 20, 5,  9,	14, 20, 5,  9,	14, 20, 4,  11, 16, 23, 4,  11, 16, 23, 4,  11, 16, 23,
+	4,  11, 16, 23, 6,  10, 15, 21, 6,  10, 15, 21, 6,  10, 15, 21, 6,  10, 15, 21,
+};
+
+static void md5_u32_to_bytes(u32 val, u8 *bytes)
+{
+	bytes[0] = (u8)val;
+	bytes[1] = (u8)(val >> 8);
+	bytes[2] = (u8)(val >> 16);
+	bytes[3] = (u8)(val >> 24);
+}
+
+static u32 md5_bytes_to_u32(u8 *bytes)
+{
+	return (u32)bytes[0] | ((u32)bytes[1] << 8) | ((u32)bytes[2] << 16) | ((u32)bytes[3] << 24);
+}
+
+void md5(const void *initial_msg, u32 initial_len, u8 digest[16])
+{
+	u32 w[16] = { 0 };
+
+	u32 h0 = 0x67452301;
+	u32 h1 = 0xefcdab89;
+	u32 h2 = 0x98badcfe;
+	u32 h3 = 0x10325476;
+
+	u32 new_len;
+	for (new_len = initial_len + 1; new_len % (512 / 8) != 448 / 8; new_len++)
+		;
+
+	u8 *msg = malloc(new_len + 8);
+	memcpy(msg, initial_msg, initial_len);
+	msg[initial_len] = 0x80;
+	for (u32 offset = initial_len + 1; offset < new_len; offset++)
+		msg[offset] = 0;
+
+	md5_u32_to_bytes(initial_len * 8, msg + new_len);
+	md5_u32_to_bytes(initial_len >> 29, msg + new_len + 4);
+
+	for (u32 offset = 0; offset < new_len; offset += (512 / 8)) {
+		for (u32 i = 0; i < 16; i++)
+			w[i] = md5_bytes_to_u32(msg + offset + i * 4);
+
+		u32 a = h0;
+		u32 b = h1;
+		u32 c = h2;
+		u32 d = h3;
+
+		u32 f, g;
+		for (u32 i = 0; i < 64; i++) {
+			if (i < 16) {
+				f = (b & c) | ((~b) & d);
+				g = i;
+			} else if (i < 32) {
+				f = (d & b) | ((~d) & c);
+				g = (5 * i + 1) % 16;
+			} else if (i < 48) {
+				f = b ^ c ^ d;
+				g = (3 * i + 5) % 16;
+			} else {
+				f = c ^ (b | (~d));
+				g = (7 * i) % 16;
+			}
+
+			u32 temp = d;
+			d = c;
+			c = b;
+			b = b + MD5_LEFTROTATE((a + f + md5_tab[i] + w[g]), md5_rounds[i]);
+			a = temp;
+		}
+
+		h0 += a;
+		h1 += b;
+		h2 += c;
+		h3 += d;
+	}
+
+	free(msg);
+
+	md5_u32_to_bytes(h0, digest);
+	md5_u32_to_bytes(h1, digest + 4);
+	md5_u32_to_bytes(h2, digest + 8);
+	md5_u32_to_bytes(h3, digest + 12);
+}
+
+/**
+ * CRC32
+ */
+
+static const u32 crc32_tab[] = {
+	0x00000000, 0x77073096, 0xee0e612c, 0x990951ba, 0x076dc419, 0x706af48f, 0xe963a535,
+	0x9e6495a3, 0x0edb8832, 0x79dcb8a4, 0xe0d5e91e, 0x97d2d988, 0x09b64c2b, 0x7eb17cbd,
+	0xe7b82d07, 0x90bf1d91, 0x1db71064, 0x6ab020f2, 0xf3b97148, 0x84be41de, 0x1adad47d,
+	0x6ddde4eb, 0xf4d4b551, 0x83d385c7, 0x136c9856, 0x646ba8c0, 0xfd62f97a, 0x8a65c9ec,
+	0x14015c4f, 0x63066cd9, 0xfa0f3d63, 0x8d080df5, 0x3b6e20c8, 0x4c69105e, 0xd56041e4,
+	0xa2677172, 0x3c03e4d1, 0x4b04d447, 0xd20d85fd, 0xa50ab56b, 0x35b5a8fa, 0x42b2986c,
+	0xdbbbc9d6, 0xacbcf940, 0x32d86ce3, 0x45df5c75, 0xdcd60dcf, 0xabd13d59, 0x26d930ac,
+	0x51de003a, 0xc8d75180, 0xbfd06116, 0x21b4f4b5, 0x56b3c423, 0xcfba9599, 0xb8bda50f,
+	0x2802b89e, 0x5f058808, 0xc60cd9b2, 0xb10be924, 0x2f6f7c87, 0x58684c11, 0xc1611dab,
+	0xb6662d3d, 0x76dc4190, 0x01db7106, 0x98d220bc, 0xefd5102a, 0x71b18589, 0x06b6b51f,
+	0x9fbfe4a5, 0xe8b8d433, 0x7807c9a2, 0x0f00f934, 0x9609a88e, 0xe10e9818, 0x7f6a0dbb,
+	0x086d3d2d, 0x91646c97, 0xe6635c01, 0x6b6b51f4, 0x1c6c6162, 0x856530d8, 0xf262004e,
+	0x6c0695ed, 0x1b01a57b, 0x8208f4c1, 0xf50fc457, 0x65b0d9c6, 0x12b7e950, 0x8bbeb8ea,
+	0xfcb9887c, 0x62dd1ddf, 0x15da2d49, 0x8cd37cf3, 0xfbd44c65, 0x4db26158, 0x3ab551ce,
+	0xa3bc0074, 0xd4bb30e2, 0x4adfa541, 0x3dd895d7, 0xa4d1c46d, 0xd3d6f4fb, 0x4369e96a,
+	0x346ed9fc, 0xad678846, 0xda60b8d0, 0x44042d73, 0x33031de5, 0xaa0a4c5f, 0xdd0d7cc9,
+	0x5005713c, 0x270241aa, 0xbe0b1010, 0xc90c2086, 0x5768b525, 0x206f85b3, 0xb966d409,
+	0xce61e49f, 0x5edef90e, 0x29d9c998, 0xb0d09822, 0xc7d7a8b4, 0x59b33d17, 0x2eb40d81,
+	0xb7bd5c3b, 0xc0ba6cad, 0xedb88320, 0x9abfb3b6, 0x03b6e20c, 0x74b1d29a, 0xead54739,
+	0x9dd277af, 0x04db2615, 0x73dc1683, 0xe3630b12, 0x94643b84, 0x0d6d6a3e, 0x7a6a5aa8,
+	0xe40ecf0b, 0x9309ff9d, 0x0a00ae27, 0x7d079eb1, 0xf00f9344, 0x8708a3d2, 0x1e01f268,
+	0x6906c2fe, 0xf762575d, 0x806567cb, 0x196c3671, 0x6e6b06e7, 0xfed41b76, 0x89d32be0,
+	0x10da7a5a, 0x67dd4acc, 0xf9b9df6f, 0x8ebeeff9, 0x17b7be43, 0x60b08ed5, 0xd6d6a3e8,
+	0xa1d1937e, 0x38d8c2c4, 0x4fdff252, 0xd1bb67f1, 0xa6bc5767, 0x3fb506dd, 0x48b2364b,
+	0xd80d2bda, 0xaf0a1b4c, 0x36034af6, 0x41047a60, 0xdf60efc3, 0xa867df55, 0x316e8eef,
+	0x4669be79, 0xcb61b38c, 0xbc66831a, 0x256fd2a0, 0x5268e236, 0xcc0c7795, 0xbb0b4703,
+	0x220216b9, 0x5505262f, 0xc5ba3bbe, 0xb2bd0b28, 0x2bb45a92, 0x5cb36a04, 0xc2d7ffa7,
+	0xb5d0cf31, 0x2cd99e8b, 0x5bdeae1d, 0x9b64c2b0, 0xec63f226, 0x756aa39c, 0x026d930a,
+	0x9c0906a9, 0xeb0e363f, 0x72076785, 0x05005713, 0x95bf4a82, 0xe2b87a14, 0x7bb12bae,
+	0x0cb61b38, 0x92d28e9b, 0xe5d5be0d, 0x7cdcefb7, 0x0bdbdf21, 0x86d3d2d4, 0xf1d4e242,
+	0x68ddb3f8, 0x1fda836e, 0x81be16cd, 0xf6b9265b, 0x6fb077e1, 0x18b74777, 0x88085ae6,
+	0xff0f6a70, 0x66063bca, 0x11010b5c, 0x8f659eff, 0xf862ae69, 0x616bffd3, 0x166ccf45,
+	0xa00ae278, 0xd70dd2ee, 0x4e048354, 0x3903b3c2, 0xa7672661, 0xd06016f7, 0x4969474d,
+	0x3e6e77db, 0xaed16a4a, 0xd9d65adc, 0x40df0b66, 0x37d83bf0, 0xa9bcae53, 0xdebb9ec5,
+	0x47b2cf7f, 0x30b5ffe9, 0xbdbdf21c, 0xcabac28a, 0x53b39330, 0x24b4a3a6, 0xbad03605,
+	0xcdd70693, 0x54de5729, 0x23d967bf, 0xb3667a2e, 0xc4614ab8, 0x5d681b02, 0x2a6f2b94,
+	0xb40bbe37, 0xc30c8ea1, 0x5a05df1b, 0x2d02ef8d,
+};
+
+u32 crc32(u32 crc, const void *buf, u32 size)
+{
+	const u8 *p = buf;
+	crc = crc ^ ~0UL;
+
+	while (size--)
+		crc = crc32_tab[(crc ^ *p++) & 0xFF] ^ (crc >> 8);
+
+	return crc ^ ~0UL;
+}
diff --git a/libs/libc/errno.c b/libs/libc/errno.c
new file mode 100644
index 0000000..7204f5b
--- /dev/null
+++ b/libs/libc/errno.c
@@ -0,0 +1,14 @@
+// MIT License, Copyright (c) 2021 Marvin Borner
+
+#include <def.h>
+#include <errno.h>
+#ifdef userspace
+
+static u32 error = 0;
+
+u32 *__errno(void)
+{
+	return &error;
+}
+
+#endif
diff --git a/libs/libc/inc/arg.h b/libs/libc/inc/arg.h
new file mode 100644
index 0000000..73e592d
--- /dev/null
+++ b/libs/libc/inc/arg.h
@@ -0,0 +1,11 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef ARG_H
+#define ARG_H
+
+typedef __builtin_va_list va_list;
+#define va_start __builtin_va_start
+#define va_end __builtin_va_end
+#define va_arg __builtin_va_arg
+
+#endif
diff --git a/libs/libc/inc/assert.h b/libs/libc/inc/assert.h
new file mode 100644
index 0000000..3656c33
--- /dev/null
+++ b/libs/libc/inc/assert.h
@@ -0,0 +1,28 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef ASSERT_H
+#define ASSERT_H
+
+#include <print.h>
+
+#ifdef kernel
+#include <proc.h>
+#define assert(exp)                                                                                \
+	if (!(exp)) {                                                                              \
+		printf("%s:%d: %s: Kernel assertion '%s' failed\n", __FILE__, __LINE__, __func__,  \
+		       #exp);                                                                      \
+		struct proc *assert_proc = proc_current();                                         \
+		if (assert_proc)                                                                   \
+			proc_exit(assert_proc, 1);                                                 \
+		else                                                                               \
+			__asm__ volatile("cli\nhlt");                                              \
+	}
+#elif defined(userspace)
+#define assert(exp)                                                                                \
+	if (!(exp))                                                                                \
+		err(1, "%s:%d: %s: Assertion '%s' failed\n", __FILE__, __LINE__, __func__, #exp);
+#else
+#error "No lib target specified. Please use -Dkernel or -Duserspace"
+#endif
+
+#endif
diff --git a/libs/libc/inc/conv.h b/libs/libc/inc/conv.h
new file mode 100644
index 0000000..adf9003
--- /dev/null
+++ b/libs/libc/inc/conv.h
@@ -0,0 +1,15 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef CONV_H
+#define CONV_H
+
+#include <def.h>
+
+int atoi(const char *str);
+char *htoa(u32 n);
+int htoi(const char *str);
+char *itoa(int n);
+
+char *conv_base(int value, char *result, int base, int is_signed);
+
+#endif
diff --git a/libs/libc/inc/cpu.h b/libs/libc/inc/cpu.h
new file mode 100644
index 0000000..d709d86
--- /dev/null
+++ b/libs/libc/inc/cpu.h
@@ -0,0 +1,110 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef CPU_H
+#define CPU_H
+
+#include <def.h>
+
+u8 inb(u16 port);
+u16 inw(u16 port);
+u32 inl(u16 port);
+void insl(u16 port, void *addr, int n);
+
+void outb(u16 port, u8 data);
+void outw(u16 port, u16 data);
+void outl(u16 port, u32 data);
+
+static inline void spinlock(int *ptr)
+{
+	int prev;
+	do
+		__asm__ volatile("lock xchgl %0,%1" : "=a"(prev) : "m"(*ptr), "a"(1));
+	while (prev);
+}
+
+#ifdef kernel
+void cpu_print(void);
+void cpu_enable_features(void);
+void fpu_restore(void);
+
+u32 cr0_get(void);
+void cr0_set(u32 cr0);
+u32 cr3_get(void);
+void cr3_set(u32 cr3);
+u32 cr4_get(void);
+void cr4_set(u32 cr4);
+
+void cli(void);
+void sti(void);
+void hlt(void);
+void idle(void);
+void loop(void);
+
+enum cpuid_requests { CPUID_VENDOR_STRING, CPUID_FEATURES, CPUID_TLB, CPUID_SERIAL };
+enum cpuid_features {
+	CPUID_FEAT_ECX_SSE3 = 1u << 0,
+	CPUID_FEAT_ECX_PCLMUL = 1u << 1,
+	CPUID_FEAT_ECX_DTES64 = 1u << 2,
+	CPUID_FEAT_ECX_MONITOR = 1u << 3,
+	CPUID_FEAT_ECX_DS_CPL = 1u << 4,
+	CPUID_FEAT_ECX_VMX = 1u << 5,
+	CPUID_FEAT_ECX_SMX = 1u << 6,
+	CPUID_FEAT_ECX_EST = 1u << 7,
+	CPUID_FEAT_ECX_TM2 = 1u << 8,
+	CPUID_FEAT_ECX_SSSE3 = 1u << 9,
+	CPUID_FEAT_ECX_CID = 1u << 10,
+	CPUID_FEAT_ECX_FMA = 1u << 12,
+	CPUID_FEAT_ECX_CX16 = 1u << 13,
+	CPUID_FEAT_ECX_ETPRD = 1u << 14,
+	CPUID_FEAT_ECX_PDCM = 1u << 15,
+	CPUID_FEAT_ECX_PCIDE = 1u << 17,
+	CPUID_FEAT_ECX_DCA = 1u << 18,
+	CPUID_FEAT_ECX_SSE4_1 = 1u << 19,
+	CPUID_FEAT_ECX_SSE4_2 = 1u << 20,
+	CPUID_FEAT_ECX_x2APIC = 1u << 21,
+	CPUID_FEAT_ECX_MOVBE = 1u << 22,
+	CPUID_FEAT_ECX_POPCNT = 1u << 23,
+	CPUID_FEAT_ECX_AES = 1u << 25,
+	CPUID_FEAT_ECX_XSAVE = 1u << 26,
+	CPUID_FEAT_ECX_OSXSAVE = 1u << 27,
+	CPUID_FEAT_ECX_AVX = 1u << 28,
+	CPUID_FEAT_ECX_F16C = 1u << 29,
+	CPUID_FEAT_ECX_RDRND = 1u << 30,
+
+	CPUID_FEAT_EDX_FPU = 1u << 0,
+	CPUID_FEAT_EDX_VME = 1u << 1,
+	CPUID_FEAT_EDX_DE = 1u << 2,
+	CPUID_FEAT_EDX_PSE = 1u << 3,
+	CPUID_FEAT_EDX_TSC = 1u << 4,
+	CPUID_FEAT_EDX_MSR = 1u << 5,
+	CPUID_FEAT_EDX_PAE = 1u << 6,
+	CPUID_FEAT_EDX_MCE = 1u << 7,
+	CPUID_FEAT_EDX_CX8 = 1u << 8,
+	CPUID_FEAT_EDX_APIC = 1u << 9,
+	CPUID_FEAT_EDX_SEP = 1u << 11,
+	CPUID_FEAT_EDX_MTRR = 1u << 12,
+	CPUID_FEAT_EDX_PGE = 1u << 13,
+	CPUID_FEAT_EDX_MCA = 1u << 14,
+	CPUID_FEAT_EDX_CMOV = 1u << 15,
+	CPUID_FEAT_EDX_PAT = 1u << 16,
+	CPUID_FEAT_EDX_PSE36 = 1u << 17,
+	CPUID_FEAT_EDX_PSN = 1u << 18,
+	CPUID_FEAT_EDX_CLF = 1u << 19,
+	CPUID_FEAT_EDX_DTES = 1u << 21,
+	CPUID_FEAT_EDX_ACPI = 1u << 22,
+	CPUID_FEAT_EDX_MMX = 1u << 23,
+	CPUID_FEAT_EDX_FXSR = 1u << 24,
+	CPUID_FEAT_EDX_SSE = 1u << 25,
+	CPUID_FEAT_EDX_SSE2 = 1u << 26,
+	CPUID_FEAT_EDX_SS = 1u << 27,
+	CPUID_FEAT_EDX_HTT = 1u << 28,
+	CPUID_FEAT_EDX_TM1 = 1u << 29,
+	CPUID_FEAT_EDX_IA64 = 1u << 30,
+};
+
+u8 cpu_has_cfeature(enum cpuid_features feature);
+u8 cpu_has_dfeature(enum cpuid_features feature);
+
+#endif
+
+#endif
diff --git a/libs/libc/inc/crypto.h b/libs/libc/inc/crypto.h
new file mode 100644
index 0000000..bbe8d7e
--- /dev/null
+++ b/libs/libc/inc/crypto.h
@@ -0,0 +1,11 @@
+// MIT License, Copyright (c) 2021 Marvin Borner
+
+#ifndef CRYPTO_H
+#define CRYPTO_H
+
+#include <def.h>
+
+void md5(const void *initial_msg, u32 initial_len, u8 digest[16]);
+u32 crc32(u32 crc, const void *buf, u32 size);
+
+#endif
diff --git a/libs/libc/inc/def.h b/libs/libc/inc/def.h
new file mode 100644
index 0000000..708ffee
--- /dev/null
+++ b/libs/libc/inc/def.h
@@ -0,0 +1,59 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef DEF_H
+#define DEF_H
+
+/**
+ * Types
+ */
+
+typedef signed char s8;
+typedef unsigned char u8;
+
+typedef signed short s16;
+typedef unsigned short u16;
+
+typedef signed long s32;
+typedef unsigned long u32;
+
+typedef signed long long s64;
+typedef unsigned long long u64;
+
+/**
+ * Macros
+ */
+
+#define UNUSED(a) ((void)(a))
+
+#define MIN(a, b) (((a) < (b)) ? (a) : (b))
+#define MAX(a, b) (((a) > (b)) ? (a) : (b))
+
+#define ABS(a) (((a) < 0) ? (-a) : (a))
+
+#define NORETURN __attribute__((noreturn))
+#define NO_SANITIZE __attribute__((no_sanitize("undefined")))
+#define PACKED __attribute__((packed))
+#define ALIGNED(align) __attribute__((aligned(align)))
+
+#define EOF (-1)
+#define NULL ((void *)0)
+
+#define U8_MAX 255
+#define S8_MAX 127
+#define S8_MIN -128
+#define U16_MAX 65535
+#define S16_MAX 32767
+#define S16_MIN -32768
+#define U32_MAX 4294967295
+#define S32_MAX 2147483647
+#define S32_MIN -2147483648
+
+#define LONG_MAX S32_MAX
+#define LONG_MIN S32_MIN
+
+#define MILLION 1000000
+#define BILLION 1000000000
+#define TRILLION 1000000000000
+#define QUADRILLION 1000000000000000
+
+#endif
diff --git a/libs/libc/inc/errno.h b/libs/libc/inc/errno.h
new file mode 100644
index 0000000..e226aba
--- /dev/null
+++ b/libs/libc/inc/errno.h
@@ -0,0 +1,52 @@
+// MIT License, Copyright (c) 2021 Marvin Borner
+
+#ifndef ERRNO_H
+#define ERRNO_H
+
+#include <def.h>
+
+typedef s32 res;
+
+#define EOK 0 /* Success */
+#define EPERM 1 /* Operation not permitted */
+#define ENOENT 2 /* No such file or directory */
+#define ESRCH 3 /* No such process */
+#define EINTR 4 /* Interrupted system call */
+#define EIO 5 /* I/O error */
+#define ENXIO 6 /* No such device or address */
+#define E2BIG 7 /* Argument list too long */
+#define ENOEXEC 8 /* Exec format error */
+#define EBADF 9 /* Bad file number */
+#define ECHILD 10 /* No child processes */
+#define EAGAIN 11 /* Try again */
+#define ENOMEM 12 /* Out of memory */
+#define EACCES 13 /* Permission denied */
+#define EFAULT 14 /* Bad address */
+#define ENOTBLK 15 /* Block device required */
+#define EBUSY 16 /* Device or resource busy */
+#define EEXIST 17 /* File exists */
+#define EXDEV 18 /* Cross-device link */
+#define ENODEV 19 /* No such device */
+#define ENOTDIR 20 /* Not a directory */
+#define EISDIR 21 /* Is a directory */
+#define EINVAL 22 /* Invalid argument */
+#define ENFILE 23 /* File table overflow */
+#define EMFILE 24 /* Too many open files */
+#define ENOTTY 25 /* Not a typewriter */
+#define ETXTBSY 26 /* Text file busy */
+#define EFBIG 27 /* File too large */
+#define ENOSPC 28 /* No space left on device */
+#define ESPIPE 29 /* Illegal seek */
+#define EROFS 30 /* Read-only file system */
+#define EMLINK 31 /* Too many links */
+#define EPIPE 32 /* Broken pipe */
+#define EDOM 33 /* Math argument out of domain of func */
+#define ERANGE 34 /* Math result not representable */
+#define EMAX 35 /* Max errno */
+
+#ifdef userspace
+#define errno (*__errno())
+extern u32 *__errno(void);
+#endif
+
+#endif
diff --git a/libs/libc/inc/input.h b/libs/libc/inc/input.h
new file mode 100644
index 0000000..5fc7ed1
--- /dev/null
+++ b/libs/libc/inc/input.h
@@ -0,0 +1,279 @@
+// GPL-2.0 WITH Linux-syscall-note License
+// Copyright (c) 1999-2002 Vojtech Pavlik
+// Copyright (c) 2015 Hans de Goede <hdegoede@redhat.com>
+// This file is from the linux source and was modified by Marvin Borner
+
+#ifndef SCANCODE_H
+#define SCANCODE_H
+
+#define KEY_RESERVED 0
+#define KEY_ESC 1
+#define KEY_1 2
+#define KEY_2 3
+#define KEY_3 4
+#define KEY_4 5
+#define KEY_5 6
+#define KEY_6 7
+#define KEY_7 8
+#define KEY_8 9
+#define KEY_9 10
+#define KEY_0 11
+#define KEY_MINUS 12
+#define KEY_EQUAL 13
+#define KEY_BACKSPACE 14
+#define KEY_TAB 15
+#define KEY_Q 16
+#define KEY_W 17
+#define KEY_E 18
+#define KEY_R 19
+#define KEY_T 20
+#define KEY_Y 21
+#define KEY_U 22
+#define KEY_I 23
+#define KEY_O 24
+#define KEY_P 25
+#define KEY_LEFTBRACE 26
+#define KEY_RIGHTBRACE 27
+#define KEY_ENTER 28
+#define KEY_LEFTCTRL 29
+#define KEY_A 30
+#define KEY_S 31
+#define KEY_D 32
+#define KEY_F 33
+#define KEY_G 34
+#define KEY_H 35
+#define KEY_J 36
+#define KEY_K 37
+#define KEY_L 38
+#define KEY_SEMICOLON 39
+#define KEY_APOSTROPHE 40
+#define KEY_GRAVE 41
+#define KEY_LEFTSHIFT 42
+#define KEY_BACKSLASH 43
+#define KEY_Z 44
+#define KEY_X 45
+#define KEY_C 46
+#define KEY_V 47
+#define KEY_B 48
+#define KEY_N 49
+#define KEY_M 50
+#define KEY_COMMA 51
+#define KEY_DOT 52
+#define KEY_SLASH 53
+#define KEY_RIGHTSHIFT 54
+#define KEY_KPASTERISK 55
+#define KEY_LEFTALT 56
+#define KEY_SPACE 57
+#define KEY_CAPSLOCK 58
+#define KEY_F1 59
+#define KEY_F2 60
+#define KEY_F3 61
+#define KEY_F4 62
+#define KEY_F5 63
+#define KEY_F6 64
+#define KEY_F7 65
+#define KEY_F8 66
+#define KEY_F9 67
+#define KEY_F10 68
+#define KEY_NUMLOCK 69
+#define KEY_SCROLLLOCK 70
+#define KEY_KP7 71
+#define KEY_KP8 72
+#define KEY_KP9 73
+#define KEY_KPMINUS 74
+#define KEY_KP4 75
+#define KEY_KP5 76
+#define KEY_KP6 77
+#define KEY_KPPLUS 78
+#define KEY_KP1 79
+#define KEY_KP2 80
+#define KEY_KP3 81
+#define KEY_KP0 82
+#define KEY_KPDOT 83
+
+#define KEY_ZENKAKUHANKAKU 85
+#define KEY_102ND 86
+#define KEY_F11 87
+#define KEY_F12 88
+#define KEY_RO 89
+#define KEY_KATAKANA 90
+#define KEY_HIRAGANA 91
+#define KEY_HENKAN 92
+#define KEY_KATAKANAHIRAGANA 93
+#define KEY_MUHENKAN 94
+#define KEY_KPJPCOMMA 95
+#define KEY_KPENTER 96
+#define KEY_RIGHTCTRL 97
+#define KEY_KPSLASH 98
+#define KEY_SYSRQ 99
+#define KEY_RIGHTALT 100
+#define KEY_LINEFEED 101
+#define KEY_HOME 102
+#define KEY_UP 103
+#define KEY_PAGEUP 104
+#define KEY_LEFT 105
+#define KEY_RIGHT 106
+#define KEY_END 107
+#define KEY_DOWN 108
+#define KEY_PAGEDOWN 109
+#define KEY_INSERT 110
+#define KEY_DELETE 111
+#define KEY_MACRO 112
+#define KEY_MUTE 113
+#define KEY_VOLUMEDOWN 114
+#define KEY_VOLUMEUP 115
+#define KEY_POWER 116
+#define KEY_KPEQUAL 117
+#define KEY_KPPLUSMINUS 118
+#define KEY_PAUSE 119
+#define KEY_SCALE 120
+
+#define KEY_KPCOMMA 121
+#define KEY_HANGEUL 122
+#define KEY_HANGUEL KEY_HANGEUL
+#define KEY_HANJA 123
+#define KEY_YEN 124
+#define KEY_LEFTMETA 125
+#define KEY_RIGHTMETA 126
+#define KEY_COMPOSE 127
+
+#define KEY_STOP 128
+#define KEY_AGAIN 129
+#define KEY_PROPS 130
+#define KEY_UNDO 131
+#define KEY_FRONT 132
+#define KEY_COPY 133
+#define KEY_OPEN 134
+#define KEY_PASTE 135
+#define KEY_FIND 136
+#define KEY_CUT 137
+#define KEY_HELP 138
+#define KEY_MENU 139
+#define KEY_CALC 140
+#define KEY_SETUP 141
+#define KEY_SLEEP 142
+#define KEY_WAKEUP 143
+#define KEY_FILE 144
+#define KEY_SENDFILE 145
+#define KEY_DELETEFILE 146
+#define KEY_XFER 147
+#define KEY_PROG1 148
+#define KEY_PROG2 149
+#define KEY_WWW 150
+#define KEY_MSDOS 151
+#define KEY_COFFEE 152
+#define KEY_SCREENLOCK KEY_COFFEE
+#define KEY_ROTATE_DISPLAY 153
+#define KEY_DIRECTION KEY_ROTATE_DISPLAY
+#define KEY_CYCLEWINDOWS 154
+#define KEY_MAIL 155
+#define KEY_BOOKMARKS 156
+#define KEY_COMPUTER 157
+#define KEY_BACK 158
+#define KEY_FORWARD 159
+#define KEY_CLOSECD 160
+#define KEY_EJECTCD 161
+#define KEY_EJECTCLOSECD 162
+#define KEY_NEXTSONG 163
+#define KEY_PLAYPAUSE 164
+#define KEY_PREVIOUSSONG 165
+#define KEY_STOPCD 166
+#define KEY_RECORD 167
+#define KEY_REWIND 168
+#define KEY_PHONE 169
+#define KEY_ISO 170
+#define KEY_CONFIG 171
+#define KEY_HOMEPAGE 172
+#define KEY_REFRESH 173
+#define KEY_EXIT 174
+#define KEY_MOVE 175
+#define KEY_EDIT 176
+#define KEY_SCROLLUP 177
+#define KEY_SCROLLDOWN 178
+#define KEY_KPLEFTPAREN 179
+#define KEY_KPRIGHTPAREN 180
+#define KEY_NEW 181
+#define KEY_REDO 182
+
+#define KEY_F13 183
+#define KEY_F14 184
+#define KEY_F15 185
+#define KEY_F16 186
+#define KEY_F17 187
+#define KEY_F18 188
+#define KEY_F19 189
+#define KEY_F20 190
+#define KEY_F21 191
+#define KEY_F22 192
+#define KEY_F23 193
+#define KEY_F24 194
+
+#define KEY_PLAYCD 200
+#define KEY_PAUSECD 201
+#define KEY_PROG3 202
+#define KEY_PROG4 203
+#define KEY_DASHBOARD 204
+#define KEY_SUSPEND 205
+#define KEY_CLOSE 206
+#define KEY_PLAY 207
+#define KEY_FASTFORWARD 208
+#define KEY_BASSBOOST 209
+#define KEY_PRINT 210
+#define KEY_HP 211
+#define KEY_CAMERA 212
+#define KEY_SOUND 213
+#define KEY_QUESTION 214
+#define KEY_EMAIL 215
+#define KEY_CHAT 216
+#define KEY_SEARCH 217
+#define KEY_CONNECT 218
+#define KEY_FINANCE 219
+#define KEY_SPORT 220
+#define KEY_SHOP 221
+#define KEY_ALTERASE 222
+#define KEY_CANCEL 223
+#define KEY_BRIGHTNESSDOWN 224
+#define KEY_BRIGHTNESSUP 225
+#define KEY_MEDIA 226
+
+#define KEY_SWITCHVIDEOMODE 227
+#define KEY_KBDILLUMTOGGLE 228
+#define KEY_KBDILLUMDOWN 229
+#define KEY_KBDILLUMUP 230
+
+#define KEY_SEND 231
+#define KEY_REPLY 232
+#define KEY_FORWARDMAIL 233
+#define KEY_SAVE 234
+#define KEY_DOCUMENTS 235
+
+#define KEY_BATTERY 236
+
+#define KEY_BLUETOOTH 237
+#define KEY_WLAN 238
+#define KEY_UWB 239
+
+#define KEY_UNKNOWN 240
+
+#define KEY_VIDEO_NEXT 241
+#define KEY_VIDEO_PREV 242
+#define KEY_BRIGHTNESS_CYCLE 243
+#define KEY_BRIGHTNESS_AUTO 244
+#define KEY_BRIGHTNESS_ZERO KEY_BRIGHTNESS_AUTO
+#define KEY_DISPLAY_OFF 245
+
+#define KEY_WWAN 246
+#define KEY_WIMAX KEY_WWAN
+#define KEY_RFKILL 247
+
+#define KEY_MICMUTE 248
+
+#define KEY_NUMERIC(code) ((code) >= KEY_1 && (code) <= KEY_0)
+#define KEY_ALPHABETIC(code)                                                                       \
+	(((code) >= KEY_Q && (code) <= KEY_P) || ((code) >= KEY_A && (code) <= KEY_L) ||           \
+	 ((code) >= KEY_Z && (code) <= KEY_M))
+#define KEY_ALPHANUMERIC(code) (KEY_NUMERIC((code)) || KEY_ALPHABETIC((code)))
+
+typedef int fortytwo;
+
+#endif
diff --git a/libs/libc/inc/ioctl.h b/libs/libc/inc/ioctl.h
new file mode 100644
index 0000000..c3eec56
--- /dev/null
+++ b/libs/libc/inc/ioctl.h
@@ -0,0 +1,11 @@
+// MIT License, Copyright (c) 2021 Marvin Borner
+
+#ifndef IOCTL
+#define IOCTL
+
+// FB interface
+#define IO_FB_GET 0
+
+int ioctl_is_awesome; // GCC is not
+
+#endif
diff --git a/libs/libc/inc/list.h b/libs/libc/inc/list.h
new file mode 100644
index 0000000..0b82b48
--- /dev/null
+++ b/libs/libc/inc/list.h
@@ -0,0 +1,29 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef LIST_H
+#define LIST_H
+
+#include <def.h>
+
+struct list {
+	struct node *head;
+};
+
+struct node {
+	void *data;
+	int nonce;
+	struct node *next;
+	struct node *prev;
+};
+
+struct list *list_new(void);
+void list_destroy(struct list *list);
+/* struct node *list_new_node(); */ // TODO: Make node-specific things static/private?
+/* void list_add_node(struct list *list, struct node *node); */
+struct node *list_add(struct list *list, void *data);
+struct list *list_remove(struct list *list, struct node *node);
+struct node *list_last(struct list *list);
+struct list *list_swap(struct list *list, struct node *a, struct node *b);
+struct node *list_first_data(struct list *list, void *data);
+
+#endif
diff --git a/libs/libc/inc/math.h b/libs/libc/inc/math.h
new file mode 100644
index 0000000..82f431f
--- /dev/null
+++ b/libs/libc/inc/math.h
@@ -0,0 +1,8 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef MATH_H
+#define MATH_H
+
+int pow(int base, int exp);
+
+#endif
diff --git a/libs/libc/inc/mem.h b/libs/libc/inc/mem.h
new file mode 100644
index 0000000..ec00628
--- /dev/null
+++ b/libs/libc/inc/mem.h
@@ -0,0 +1,29 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef MEM_H
+#define MEM_H
+
+#include <def.h>
+
+void *malloc_debug(u32 size, const char *file, int line, const char *func, const char *inp);
+void free_debug(void *ptr, const char *file, int line, const char *func, const char *inp);
+#define malloc(size) malloc_debug((u32)(size), __FILE__, __LINE__, __func__, #size)
+#define free(ptr) free_debug((void *)(ptr), __FILE__, __LINE__, __func__, #ptr)
+void *realloc(void *ptr, u32 size);
+void *zalloc(u32 size);
+
+#ifdef kernel
+#define STACK_START 0x00500000 // Defined it bootloader
+#define STACK_SIZE 0x1000 // idk
+#elif defined(userspace)
+#else
+#error "No lib target specified. Please use -Dkernel or -Duserspace"
+#endif
+
+void *memcpy(void *dest, const void *src, u32 n);
+void *memset(void *dest, int val, u32 n);
+void *memchr(void *src, int c, u32 n);
+int memcmp(const void *s1, const void *s2, u32 n);
+int mememp(const u8 *buf, u32 n);
+
+#endif
diff --git a/libs/libc/inc/print.h b/libs/libc/inc/print.h
new file mode 100644
index 0000000..58b5dc6
--- /dev/null
+++ b/libs/libc/inc/print.h
@@ -0,0 +1,27 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef PRINT_H
+#define PRINT_H
+
+#include "arg.h"
+#include <def.h>
+
+int printf(const char *format, ...);
+int vprintf(const char *format, va_list ap);
+int sprintf(char *str, const char *format, ...);
+int vsprintf(char *str, const char *format, va_list ap);
+int print(const char *str);
+NORETURN void panic(const char *format, ...);
+
+#ifdef userspace
+int vfprintf(const char *path, const char *format, va_list ap);
+int fprintf(const char *path, const char *format, ...);
+int log(const char *format, ...);
+int err(int code, const char *format, ...);
+#else
+#include <proc.h>
+int print_app(enum stream_defaults id, const char *proc_name, const char *str);
+void print_trace(u32 count);
+#endif
+
+#endif
diff --git a/libs/libc/inc/random.h b/libs/libc/inc/random.h
new file mode 100644
index 0000000..a82524c
--- /dev/null
+++ b/libs/libc/inc/random.h
@@ -0,0 +1,14 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef RANDOM_H
+#define RANDOM_H
+
+#include <def.h>
+
+void srand(u32 seed);
+u32 rdrand(void);
+u32 rdseed(void);
+u32 rand(void);
+char *randstr(u32 size);
+
+#endif
diff --git a/libs/libc/inc/socket.h b/libs/libc/inc/socket.h
new file mode 100644
index 0000000..18bc316
--- /dev/null
+++ b/libs/libc/inc/socket.h
@@ -0,0 +1,38 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef SOCKET_H
+#define SOCKET_H
+
+#include <def.h>
+#include <list.h>
+
+// TODO: Use actual socket types (stream etc)
+enum socket_type { S_TCP, S_UDP };
+enum socket_state { S_CONNECTING, S_CONNECTED, S_OPEN, S_CLOSING, S_CLOSED, S_FAILED };
+
+struct tcp_socket {
+	u32 seq_no;
+	u32 ack_no;
+	u32 state;
+};
+
+struct socket_data {
+	u8 *data;
+	u32 length;
+};
+
+struct socket {
+	u32 pid;
+	u32 ip_addr;
+	u32 dst_port;
+	u32 src_port;
+	enum socket_state state;
+	enum socket_type type;
+	struct list *packets;
+	union {
+		struct tcp_socket tcp;
+		/* struct udp_socket udp; */
+	} prot;
+};
+
+#endif
diff --git a/libs/libc/inc/stack.h b/libs/libc/inc/stack.h
new file mode 100644
index 0000000..f5ad52b
--- /dev/null
+++ b/libs/libc/inc/stack.h
@@ -0,0 +1,28 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef STACK_H
+#define STACK_H
+
+#include <def.h>
+
+struct stack_node {
+	void *data;
+	int nonce;
+	struct stack_node *next;
+	struct stack_node *prev;
+};
+
+struct stack {
+	struct stack_node *tail;
+};
+
+struct stack *stack_new(void);
+void stack_destroy(struct stack *stack);
+u32 stack_empty(struct stack *stack);
+u32 stack_push_bot(struct stack *stack, void *data);
+u32 stack_push(struct stack *stack, void *data);
+void *stack_pop(struct stack *stack);
+void *stack_peek(struct stack *stack);
+void stack_clear(struct stack *stack);
+
+#endif
diff --git a/libs/libc/inc/str.h b/libs/libc/inc/str.h
new file mode 100644
index 0000000..d0a521f
--- /dev/null
+++ b/libs/libc/inc/str.h
@@ -0,0 +1,22 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef STR_H
+#define STR_H
+
+#include <def.h>
+
+u32 strlen(const char *s);
+char *strcpy(char *dst, const char *src);
+char *strncpy(char *dst, const char *src, u32 n);
+char *strchr(char *s, int c);
+char *strrchr(char *s, int c);
+char *strcat(char *dst, const char *src);
+char *strncat(char *dst, const char *src, u32 n);
+int strcmp(const char *s1, const char *s2);
+int strncmp(const char *s1, const char *s2, u32 n);
+char *strinv(char *s);
+char *strdup(const char *s);
+
+const char *strerror(u32 err);
+
+#endif
diff --git a/libs/libc/inc/sys.h b/libs/libc/inc/sys.h
new file mode 100644
index 0000000..19fb3ee
--- /dev/null
+++ b/libs/libc/inc/sys.h
@@ -0,0 +1,130 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+// Syscall implementation
+
+#ifndef SYS_H
+#define SYS_H
+
+#include <def.h>
+#include <errno.h>
+
+#define KEYBOARD_MAGIC 0x555555
+#define MOUSE_MAGIC 0xaaaaaa
+
+#define SYS_BOOT_MAGIC 0x18122002
+#define SYS_BOOT_REBOOT 0xeeb007
+#define SYS_BOOT_SHUTDOWN 0xdead
+
+enum sys {
+	SYS_LOOP, // To infinity and beyond (debug)!
+	SYS_ALLOC, // Allocate memory
+	SYS_SHACCESS, // Access shared memory
+	SYS_FREE, // Free memory
+	SYS_STAT, // Get file information
+	SYS_READ, // Read file
+	SYS_WRITE, // Write to file
+	SYS_IOCTL, // Interact with a file/device
+	SYS_POLL, // Wait for multiple files
+	SYS_EXEC, // Execute path
+	SYS_EXIT, // Exit current process
+	SYS_BOOT, // Boot functions (e.g. reboot/shutdown)
+	SYS_YIELD, // Switch to next process
+	SYS_TIME, // Get kernel time
+	/* SYS_NET_OPEN, // Open network socket */
+	/* SYS_NET_CLOSE, // Close network socket */
+	/* SYS_NET_CONNECT, // Connect to destination */
+	/* SYS_NET_SEND, // Send to socket */
+	/* SYS_NET_RECEIVE, // Receive data from socket */
+};
+
+struct event_keyboard {
+	u32 magic;
+	u32 scancode;
+	u8 press;
+};
+
+struct event_mouse {
+	u32 magic;
+	s32 diff_x;
+	s32 diff_y;
+	u8 but1;
+	u8 but2;
+	u8 but3;
+};
+
+struct stat {
+	u32 dev_id;
+	u32 mode;
+	u32 uid;
+	u32 gid;
+	u32 size;
+};
+
+#if defined(userspace)
+
+/**
+ * Syscall wrappers
+ */
+
+void loop(void);
+void exit(s32 status);
+res read(const char *path, void *buf, u32 offset, u32 count);
+res write(const char *path, const void *buf, u32 offset, u32 count);
+res ioctl(const char *path, ...);
+res stat(const char *path, struct stat *buf);
+res poll(const char **files);
+res exec(const char *path, ...);
+res yield(void);
+res boot(u32 cmd);
+u32 time(void);
+
+res sys_alloc(u32 size, u32 *addr);
+res sys_free(void *ptr);
+res shalloc(u32 size, u32 *addr, u32 *id);
+res shaccess(u32 id, u32 *addr, u32 *size);
+
+static inline u32 getpid(void)
+{
+	static u32 buf = 0;
+	if (buf)
+		return buf;
+	read("/proc/self/pid", &buf, 0, sizeof(buf));
+	return buf;
+}
+
+#include <print.h>
+#include <str.h>
+static inline u32 pidof(const char *name)
+{
+	u32 curr = 1;
+	char buf[32] = { 0 }, path[32] = { 0 };
+	while (curr < 1000) { // Max pid??
+		if (sprintf(path, "/proc/%d/name", curr) > 0 && read(path, buf, 0, 32) > 0)
+			if (!strcmp(name, buf))
+				return curr;
+
+		curr++;
+	}
+
+	return -1;
+}
+
+// Simple read wrapper
+#include <mem.h>
+static inline void *sread(const char *path)
+{
+	struct stat s = { 0 };
+	if (stat(path, &s) != 0 || !s.size)
+		return NULL;
+	void *buf = malloc(s.size);
+	read(path, buf, 0, s.size);
+	return buf;
+}
+
+/**
+ * At exit
+ */
+
+void atexit(void (*func)(void));
+
+#endif
+#endif
diff --git a/libs/libc/inc/vec.h b/libs/libc/inc/vec.h
new file mode 100644
index 0000000..a036931
--- /dev/null
+++ b/libs/libc/inc/vec.h
@@ -0,0 +1,36 @@
+// MIT License, Copyright (c) 2021 Marvin Borner
+
+#ifndef VEC_H
+#define VEC_H
+
+#include <def.h>
+
+typedef struct vec2 {
+	u32 x, y;
+} vec2;
+
+typedef struct vec3 {
+	u32 x, y, z;
+} vec3;
+
+#define vec2(x, y) ((vec2){ (x), (y) })
+#define vec2to3(a, z) ((vec3){ a.x, a.y, (z) })
+#define vec2_add(a, b) ((vec2){ a.x + b.x, a.y + b.y })
+#define vec2_sub(a, b) ((vec2){ a.x - b.x, a.y - b.y })
+#define vec2_mul(a, b) ((vec2){ a.x * (b), a.y * (b) })
+#define vec2_div(a, b) ((vec2){ a.x / (b), a.y / (b) })
+#define vec2_dot(a, b) ((u32)(a.x * b.x + a.y * b.y))
+#define vec2_eq(a, b) (a.x == b.x && a.y == b.y)
+
+#define vec3(x, y, z) ((vec3){ (x), (y), (z) })
+#define vec3to2(a) ((vec2){ a.x, a.y })
+#define vec3_add(a, b) ((vec3){ a.x + b.x, a.y + b.y, a.z + b.z })
+#define vec3_sub(a, b) ((vec3){ a.x - b.x, a.y - b.y, a.z - b.z })
+#define vec3_mul(a, b) ((vec3){ a.x * (b), a.y * (b), a.z * (b) })
+#define vec3_div(a, b) ((vec3){ a.x / (b), a.y / (b), a.z / (b) })
+#define vec3_dot(a, b) ((u32)(a.x * b.x + a.y * b.y + a.z * b.z))
+#define vec3_eq(a, b) (a.x == b.x && a.y == b.y && a.z == c.z)
+#define vec3_cross(a, b)                                                                           \
+	((vec3){ a.y * b.z - a.z * b.y, a.z * b.x - a.x * b.z, a.x * b.y - a.y * b.x })
+
+#endif
diff --git a/libs/libc/list.c b/libs/libc/list.c
new file mode 100644
index 0000000..c86b23d
--- /dev/null
+++ b/libs/libc/list.c
@@ -0,0 +1,138 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#include <def.h>
+#include <list.h>
+#include <mem.h>
+
+static int nonce = 0;
+
+struct list *list_new(void)
+{
+	struct list *list = malloc(sizeof(*list));
+	list->head = NULL;
+	return list;
+}
+
+void list_destroy(struct list *list)
+{
+	if (!list)
+		return;
+	struct node *iterator = list->head;
+	while (iterator != NULL) {
+		if (iterator->next == NULL) {
+			free(iterator);
+			break;
+		}
+		iterator = iterator->next;
+		free(iterator->prev);
+	}
+	list->head = NULL;
+	free(list);
+	list = NULL;
+}
+
+static struct node *list_new_node(void)
+{
+	struct node *node = malloc(sizeof(*node));
+	node->data = NULL;
+	node->prev = NULL;
+	node->next = NULL;
+	node->nonce = nonce++;
+	return node;
+}
+
+static struct node *list_add_node(struct list *list, struct node *node)
+{
+	if (!list || !node)
+		return NULL;
+
+	if (list->head == NULL) {
+		list->head = node;
+		return list->head;
+	}
+
+	struct node *iterator = list->head;
+	while (iterator != NULL) {
+		if (iterator->next == NULL) {
+			iterator->next = node;
+			node->prev = iterator;
+			break;
+		}
+		iterator = iterator->next;
+	}
+	return node;
+}
+
+struct node *list_last(struct list *list)
+{
+	if (!list || !list->head)
+		return NULL;
+
+	struct node *iterator = list->head;
+	while (iterator != NULL) {
+		if (iterator->next == NULL)
+			return iterator;
+		iterator = iterator->next;
+	}
+
+	return NULL;
+}
+
+struct node *list_first_data(struct list *list, void *data)
+{
+	if (!list || !list->head || !data)
+		return NULL;
+
+	struct node *iterator = list->head;
+	while (iterator != NULL) {
+		if (iterator->data == data)
+			return iterator;
+		iterator = iterator->next;
+	}
+
+	return NULL;
+}
+
+// TODO: Actually swap the nodes, not the data
+struct list *list_swap(struct list *list, struct node *a, struct node *b)
+{
+	if (!list || !list->head || !a || !b)
+		return NULL;
+
+	void *tmp = a->data;
+	a->data = b->data;
+	b->data = tmp;
+
+	return list;
+}
+
+struct node *list_add(struct list *list, void *data)
+{
+	struct node *node = list_new_node();
+	node->data = data;
+	return list_add_node(list, node);
+}
+
+// Maybe list_remove_node?
+struct list *list_remove(struct list *list, struct node *node)
+{
+	if (!list || !list->head || !node)
+		return NULL;
+
+	if (list->head == node) {
+		list->head = list->head->next;
+		return list;
+	}
+
+	struct node *iterator = list->head->next;
+	while (iterator != node) {
+		iterator = iterator->next;
+		if (iterator == NULL)
+			return NULL;
+	}
+
+	iterator->prev->next = iterator->next;
+	if (iterator->next)
+		iterator->next->prev = iterator->prev;
+	return list;
+}
diff --git a/libs/libc/math.c b/libs/libc/math.c
new file mode 100644
index 0000000..c8142b5
--- /dev/null
+++ b/libs/libc/math.c
@@ -0,0 +1,17 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#include <math.h>
+
+int pow(int base, int exp)
+{
+	if (exp < 0)
+		return 0;
+
+	if (!exp)
+		return 1;
+
+	int ret = base;
+	for (int i = 1; i < exp; i++)
+		ret *= base;
+	return ret;
+}
diff --git a/libs/libc/mem.c b/libs/libc/mem.c
new file mode 100644
index 0000000..95242e4
--- /dev/null
+++ b/libs/libc/mem.c
@@ -0,0 +1,121 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#include <assert.h>
+#include <def.h>
+#include <mem.h>
+#include <sys.h>
+
+void *memcpy(void *dest, const void *src, u32 n)
+{
+#ifdef userspace
+	// Inspired by Jeko at osdev
+	u8 *dest_byte = dest;
+	const u8 *src_byte = src;
+	for (u32 i = 0; i < n / 16; i++) {
+		__asm__ volatile("movups (%0), %%xmm0\n"
+				 "movntdq %%xmm0, (%1)\n" ::"r"(src_byte),
+				 "r"(dest_byte)
+				 : "memory");
+
+		src_byte += 16;
+		dest_byte += 16;
+	}
+
+	if (n & 7) {
+		n = n & 7;
+
+		int d0, d1, d2;
+		__asm__ volatile("rep ; movsl\n\t"
+				 "testb $2,%b4\n\t"
+				 "je 1f\n\t"
+				 "movsw\n"
+				 "1:\ttestb $1,%b4\n\t"
+				 "je 2f\n\t"
+				 "movsb\n"
+				 "2:"
+				 : "=&c"(d0), "=&D"(d1), "=&S"(d2)
+				 : "0"(n / 4), "q"(n), "1"((long)dest_byte), "2"((long)src_byte)
+				 : "memory");
+	}
+	return dest_byte;
+#else
+	// Inspired by jgraef at osdev
+	u32 num_dwords = n / 4;
+	u32 num_bytes = n % 4;
+	u32 *dest32 = (u32 *)dest;
+	const u32 *src32 = (const u32 *)src;
+	u8 *dest8 = ((u8 *)dest) + num_dwords * 4;
+	const u8 *src8 = ((const u8 *)src) + num_dwords * 4;
+
+	// TODO: What's faster?
+	__asm__ volatile("rep movsl\n"
+			 : "=S"(src32), "=D"(dest32), "=c"(num_dwords)
+			 : "S"(src32), "D"(dest32), "c"(num_dwords)
+			 : "memory");
+
+	/* for (u32 i = 0; i < num_dwords; i++) { */
+	/* 	dest32[i] = src32[i]; */
+	/* } */
+
+	for (u32 i = 0; i < num_bytes; i++) {
+		dest8[i] = src8[i];
+	}
+	return dest;
+#endif
+}
+
+void *memset(void *dest, int val, u32 n)
+{
+	u32 uval = val;
+	u32 num_dwords = n / 4;
+	u32 num_bytes = n % 4;
+	u32 *dest32 = (u32 *)dest;
+	u8 *dest8 = ((u8 *)dest) + num_dwords * 4;
+	u8 val8 = (u8)val;
+	u32 val32 = uval | (uval << 8) | (uval << 16) | (uval << 24);
+
+	// TODO: What's faster?
+	__asm__ volatile("rep stosl\n"
+			 : "=D"(dest32), "=c"(num_dwords)
+			 : "D"(dest32), "c"(num_dwords), "a"(val32)
+			 : "memory");
+
+	/* for (u32 i = 0; i < num_dwords; i++) { */
+	/* 	dest32[i] = val32; */
+	/* } */
+
+	for (u32 i = 0; i < num_bytes; i++) {
+		dest8[i] = val8;
+	}
+	return dest;
+}
+
+void *memchr(void *src, int c, u32 n)
+{
+	u8 *s = (u8 *)src;
+
+	while (n-- > 0) {
+		if (*s == c)
+			return s;
+		s++;
+	}
+	return NULL;
+}
+
+int memcmp(const void *s1, const void *s2, u32 n)
+{
+	const u8 *a = (const u8 *)s1;
+	const u8 *b = (const u8 *)s2;
+	for (u32 i = 0; i < n; i++) {
+		if (a[i] < b[i])
+			return -1;
+		else if (b[i] < a[i])
+			return 1;
+	}
+	return 0;
+}
+
+int mememp(const u8 *buf, u32 n)
+{
+	return buf[0] == 0 && !memcmp(buf, buf + 1, n - 1);
+}
diff --git a/libs/libc/print.c b/libs/libc/print.c
new file mode 100644
index 0000000..2422fed
--- /dev/null
+++ b/libs/libc/print.c
@@ -0,0 +1,257 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#include <arg.h>
+#include <assert.h>
+#include <conv.h>
+#include <cpu.h>
+#include <def.h>
+#include <mem.h>
+#include <str.h>
+
+static void append(char *dest, char *src, int index)
+{
+	for (u32 i = index; i < strlen(src) + index; i++)
+		dest[i] = src[i - index];
+	dest[index + strlen(src)] = 0;
+}
+
+int vsprintf(char *str, const char *format, va_list ap)
+{
+	u8 ready_to_format = 0;
+
+	int i = 0;
+	char buf = 0;
+	char format_buffer[20] = { '\0' };
+
+	for (; *format; format++) {
+		if (ready_to_format) {
+			ready_to_format = 0;
+
+			if (*format == '%') {
+				str[i] = '%';
+				continue;
+			}
+
+			buf = *format;
+
+			// TODO: Improve this repetitive code
+			if (buf == 's') {
+				char *string = va_arg(ap, char *);
+				assert(string);
+				append(str, string, i);
+				i = strlen(str);
+			} else if (buf == 'x') {
+				conv_base(va_arg(ap, u32), format_buffer, 16, 0);
+				append(str, format_buffer, i);
+				i = strlen(str);
+			} else if (buf == 'd' || buf == 'i') {
+				conv_base(va_arg(ap, s32), format_buffer, 10, 1);
+				append(str, format_buffer, i);
+				i = strlen(str);
+			} else if (buf == 'u') {
+				conv_base(va_arg(ap, u32), format_buffer, 10, 0);
+				append(str, format_buffer, i);
+				i = strlen(str);
+			} else if (buf == 'o') {
+				conv_base(va_arg(ap, u32), format_buffer, 8, 0);
+				append(str, format_buffer, i);
+				i = strlen(str);
+			} else if (buf == 'b') {
+				conv_base(va_arg(ap, u32), format_buffer, 2, 0);
+				append(str, format_buffer, i);
+				i = strlen(str);
+			} else if (buf == 'c') {
+				str[i] = (char)va_arg(ap, int);
+				i++;
+			} else {
+				assert(0);
+			}
+		} else {
+			if (*format == '%')
+				ready_to_format = 1;
+			else {
+				str[i] = *format;
+				i++;
+			}
+		}
+
+		format_buffer[0] = '\0';
+	}
+
+	return strlen(str);
+}
+
+int sprintf(char *str, const char *format, ...)
+{
+	va_list ap;
+	va_start(ap, format);
+	int len = vsprintf(str, format, ap);
+	va_end(ap);
+
+	return len;
+}
+
+#ifdef userspace
+
+#include <sys.h>
+#define PATH_OUT "/proc/self/io/out"
+#define PATH_LOG "/proc/self/io/log"
+#define PATH_ERR "/proc/self/io/err"
+
+int vprintf(const char *format, va_list ap)
+{
+	return vfprintf(PATH_OUT, format, ap);
+}
+
+int vfprintf(const char *path, const char *format, va_list ap)
+{
+	char buf[1024] = { 0 };
+	int len = vsprintf(buf, format, ap);
+	return write(path, buf, 0, len);
+}
+
+int fprintf(const char *path, const char *format, ...)
+{
+	va_list ap;
+	va_start(ap, format);
+	int len = vfprintf(path, format, ap);
+	va_end(ap);
+
+	return len;
+}
+
+int printf(const char *format, ...)
+{
+	va_list ap;
+	va_start(ap, format);
+	int len = vprintf(format, ap);
+	va_end(ap);
+
+	return len;
+}
+
+int log(const char *format, ...)
+{
+	va_list ap;
+	va_start(ap, format);
+	int len = vfprintf(PATH_LOG, format, ap);
+	va_end(ap);
+
+	return len;
+}
+
+int err(int code, const char *format, ...)
+{
+	if (errno != EOK)
+		log("ERRNO: %d (%s)\n", errno, strerror(errno));
+	va_list ap;
+	va_start(ap, format);
+	vfprintf(PATH_ERR, format, ap);
+	va_end(ap);
+	exit(code);
+	return -1;
+}
+
+int print(const char *str)
+{
+	return write(PATH_OUT, str, 0, strlen(str));
+}
+
+#else
+
+// The kernel prints everything into the serial console
+
+#include <mm.h>
+#include <proc.h>
+#include <serial.h>
+
+#define RED "\x1B[1;31m"
+#define GRN "\x1B[1;32m"
+#define YEL "\x1B[1;33m"
+#define BLU "\x1B[1;34m"
+#define MAG "\x1B[1;35m"
+#define CYN "\x1B[1;36m"
+#define WHT "\x1B[1;37m"
+#define RES "\x1B[0m"
+
+static void print_kernel(const char *str)
+{
+	serial_print(RED);
+	serial_print("[KER] ");
+	serial_print(str);
+	serial_print(RES);
+}
+
+int vprintf(const char *format, va_list ap)
+{
+	char buf[1024] = { 0 };
+	int len = vsprintf(buf, format, ap);
+	print_kernel(buf);
+	return len;
+}
+
+int printf(const char *format, ...)
+{
+	va_list ap;
+	va_start(ap, format);
+	int len = vprintf(format, ap);
+	va_end(ap);
+
+	return len;
+}
+
+int print_app(enum stream_defaults id, const char *proc_name, const char *str)
+{
+	if (id == STREAM_LOG)
+		serial_print(CYN "[LOG] to ");
+	else if (id == STREAM_ERR)
+		serial_print(YEL "[ERR] to ");
+	serial_print(proc_name);
+	serial_print(": ");
+	serial_print(str);
+	serial_print(RES);
+	return 1;
+}
+
+int print(const char *str)
+{
+	print_kernel(str);
+	return strlen(str);
+}
+
+void print_trace(u32 count)
+{
+	struct frame {
+		struct frame *ebp;
+		u32 eip;
+	} * stk;
+	__asm__ volatile("movl %%ebp, %0;" : "=r"(stk));
+	print("EBP\tEIP\n");
+	for (u32 i = 0; stk && i < count; i++) {
+		/* u32 eip = memory_valid((void *)stk->eip) ? stk->eip : stk->eip + 64; */
+		/* printf("0x%x\t0x%x\n", stk->ebp, eip); */
+		printf("0x%x\t0x%x\n", stk->ebp, stk->eip);
+		stk = stk->ebp;
+	}
+}
+
+#endif
+
+NORETURN void panic(const char *format, ...)
+{
+	char buf[1024] = { 0 };
+	va_list ap;
+	va_start(ap, format);
+	vsprintf(buf, format, ap);
+	va_end(ap);
+#ifdef kernel
+	print("--- DON'T PANIC! ---\n");
+	print(buf);
+	print_trace(5);
+	loop();
+#else
+	err(1, buf);
+#endif
+	while (1)
+		;
+}
diff --git a/libs/libc/random.c b/libs/libc/random.c
new file mode 100644
index 0000000..2801029
--- /dev/null
+++ b/libs/libc/random.c
@@ -0,0 +1,71 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#include <cpu.h>
+#include <def.h>
+#include <mem.h>
+#include <random.h>
+
+static u32 g_seed = 1;
+
+void srand(u32 seed)
+{
+	g_seed = seed;
+}
+
+u32 rdrand(void)
+{
+#ifdef kernel
+	if (!cpu_has_cfeature(CPUID_FEAT_ECX_RDRND))
+		return rand();
+
+	u32 rd;
+	__asm__ volatile("1:\n"
+			 "rdrand %0\n"
+			 "jnc 1b\n"
+			 : "=r"(rd));
+	return rd;
+#else
+	return rand();
+#endif
+}
+
+u32 rdseed(void)
+{
+#ifdef kernel
+	if (!cpu_has_cfeature(CPUID_FEAT_ECX_RDRND))
+		return rand();
+
+	u32 rd;
+	__asm__ volatile("1:\n"
+			 "rdseed %0\n"
+			 "jnc 1b\n"
+			 : "=r"(rd));
+	return rd;
+#else
+	return rand();
+#endif
+}
+
+u32 rand(void)
+{
+	g_seed = g_seed * 1103515245 + 12345;
+	return (g_seed >> 16) & 0x7FFF;
+}
+
+char *randstr(u32 size)
+{
+	if (!size)
+		return NULL;
+
+	char *buf = malloc(size + 1);
+	const char charset[] = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789";
+
+	size--;
+	for (u32 i = 0; i < size; i++) {
+		int key = rand() % (sizeof(charset) - 1);
+		buf[i] = charset[key];
+	}
+	buf[size] = '\0';
+
+	return buf;
+}
diff --git a/libs/libc/sanitize.c b/libs/libc/sanitize.c
new file mode 100644
index 0000000..02eef84
--- /dev/null
+++ b/libs/libc/sanitize.c
@@ -0,0 +1,228 @@
+// MIT License, Copyright (c) 2021 Marvin Borner
+// Detect stack overflows and other bugs
+
+#include <def.h>
+#include <print.h>
+
+/**
+ * Stack protector
+ */
+
+#define STACK_CHK_GUARD 0xdeadbeef
+
+u32 __stack_chk_guard = STACK_CHK_GUARD;
+
+void __stack_chk_fail(void);
+NORETURN void __stack_chk_fail(void)
+{
+	panic("FATAL: Stack smashing detected\n");
+}
+
+void __stack_chk_fail_local(void);
+NORETURN void __stack_chk_fail_local(void)
+{
+	panic("FATAL: Local stack smashing detected\n");
+}
+
+/**
+ * UBSan
+ * TODO: Fix san-paths for userspace (maybe due to -fPIE?)
+ */
+
+#define is_aligned(value, alignment) !(value & (alignment - 1))
+
+struct source_location {
+	const char *file;
+	u32 line;
+	u32 column;
+};
+
+struct type_descriptor {
+	u16 kind;
+	u16 info;
+	char name[];
+};
+
+struct type_mismatch {
+	struct source_location location;
+	struct type_descriptor *type;
+	u8 alignment;
+	u8 type_check_kind;
+};
+
+struct overflow {
+	struct source_location location;
+	struct type_descriptor *type;
+};
+
+struct out_of_bounds {
+	struct source_location location;
+	struct type_descriptor *left_type;
+	struct type_descriptor *right_type;
+};
+
+void __ubsan_handle_load_invalid_value(void);
+void __ubsan_handle_load_invalid_value(void)
+{
+	panic("UBSAN: load-invalid-value\n");
+}
+
+void __ubsan_handle_nonnull_arg(void);
+void __ubsan_handle_nonnull_arg(void)
+{
+	panic("UBSAN: nonnull-arg\n");
+}
+
+void __ubsan_handle_nullability_arg(void);
+void __ubsan_handle_nullability_arg(void)
+{
+	panic("UBSAN: nullability-arg\n");
+}
+
+void __ubsan_handle_nonnull_return_v1(void);
+void __ubsan_handle_nonnull_return_v1(void)
+{
+	panic("UBSAN: nonnull-return-v1\n");
+}
+
+void __ubsan_handle_nullability_return_v1(void);
+void __ubsan_handle_nullability_return_v1(void)
+{
+	panic("UBSAN: nullability-return-v1\n");
+}
+
+void __ubsan_handle_vla_bound_not_positive(void);
+void __ubsan_handle_vla_bound_not_positive(void)
+{
+	panic("UBSAN: vla-bound-not-positive\n");
+}
+
+void __ubsan_handle_add_overflow(struct overflow *data, void *left, void *right);
+void __ubsan_handle_add_overflow(struct overflow *data, void *left, void *right)
+{
+	UNUSED(left);
+	UNUSED(right);
+	struct source_location *loc = &data->location;
+	panic("%s:%d: UBSAN: add-overflow [type: %s]\n", loc->file, loc->line, data->type->name);
+}
+
+void __ubsan_handle_sub_overflow(struct overflow *data, void *left, void *right);
+void __ubsan_handle_sub_overflow(struct overflow *data, void *left, void *right)
+{
+	UNUSED(left);
+	UNUSED(right);
+	struct source_location *loc = &data->location;
+	panic("%s:%d: UBSAN: sub-overflow [type: %s]\n", loc->file, loc->line, data->type->name);
+}
+
+void __ubsan_handle_negate_overflow(struct overflow *data, void *left, void *right);
+void __ubsan_handle_negate_overflow(struct overflow *data, void *left, void *right)
+{
+	UNUSED(left);
+	UNUSED(right);
+	struct source_location *loc = &data->location;
+	panic("%s:%d: UBSAN: negate-overflow [type: %s]\n", loc->file, loc->line, data->type->name);
+}
+
+void __ubsan_handle_mul_overflow(struct overflow *data, void *left, void *right);
+void __ubsan_handle_mul_overflow(struct overflow *data, void *left, void *right)
+{
+	UNUSED(left);
+	UNUSED(right);
+	struct source_location *loc = &data->location;
+	panic("%s:%d: UBSAN: mul-overflow [type: %s]\n", loc->file, loc->line, data->type->name);
+}
+
+void __ubsan_handle_shift_out_of_bounds(struct overflow *data, void *left, void *right);
+void __ubsan_handle_shift_out_of_bounds(struct overflow *data, void *left, void *right)
+{
+	UNUSED(left);
+	UNUSED(right);
+	struct source_location *loc = &data->location;
+	panic("%s:%d: UBSAN: shift-out-of-bounds [type: %s]\n", loc->file, loc->line,
+	      data->type->name);
+}
+
+void __ubsan_handle_divrem_overflow(struct overflow *data, void *left, void *right);
+void __ubsan_handle_divrem_overflow(struct overflow *data, void *left, void *right)
+{
+	UNUSED(left);
+	UNUSED(right);
+	struct source_location *loc = &data->location;
+	panic("%s:%d: UBSAN: divrem-overflow (probably div-by-zero) [type: %s]\n", loc->file,
+	      loc->line, data->type->name);
+}
+
+void __ubsan_handle_out_of_bounds(struct out_of_bounds *data, void *value);
+void __ubsan_handle_out_of_bounds(struct out_of_bounds *data, void *value)
+{
+	UNUSED(value);
+	struct source_location *loc = &data->location;
+	panic("%s:%d: UBSAN: out-of-bounds\n", loc->file, loc->line);
+}
+
+void __ubsan_handle_type_mismatch_v1(struct type_mismatch *data, u32 ptr);
+void __ubsan_handle_type_mismatch_v1(struct type_mismatch *data, u32 ptr)
+{
+	static const char *kinds[] = {
+		"Load of",
+		"Store to",
+		"Reference binding to",
+		"Member access within",
+		"Member call on",
+		"Constructor call on",
+		"Downcast of",
+		"Downcast of",
+		"Upcast of",
+		"Cast to virtual base of",
+		"Nonnull binding to",
+		"Dynamic operation on",
+	};
+
+	struct source_location *loc = &data->location;
+	const char *msg = "";
+	if (ptr == 0) {
+		msg = "null pointer";
+	} else if (data->alignment != 0 && is_aligned(ptr, data->alignment))
+		msg = "misaligned memory address";
+	else
+		msg = "address with insufficient space";
+	panic("%s:%d: UBSAN: %s %s [type: %s; addr: 0x%x; align: %d]\n", loc->file, loc->line,
+	      kinds[data->type_check_kind], msg, data->type->name, ptr, data->alignment);
+}
+
+void __ubsan_handle_alignment_assumption(void);
+void __ubsan_handle_alignment_assumption(void)
+{
+	panic("UBSAN: alignment-assumption\n");
+}
+
+void __ubsan_handle_builtin_unreachable(void);
+void __ubsan_handle_builtin_unreachable(void)
+{
+	panic("UBSAN: builtin-unreachable\n");
+}
+
+void __ubsan_handle_missing_return(void);
+void __ubsan_handle_missing_return(void)
+{
+	panic("UBSAN: missing-return\n");
+}
+
+void __ubsan_handle_implicit_conversion(void);
+void __ubsan_handle_implicit_conversion(void)
+{
+	panic("UBSAN: implicit-conversion\n");
+}
+
+void __ubsan_handle_invalid_builtin(void);
+void __ubsan_handle_invalid_builtin(void)
+{
+	panic("UBSAN: invalid-builtin\n");
+}
+
+void __ubsan_handle_pointer_overflow(void);
+void __ubsan_handle_pointer_overflow(void)
+{
+	panic("UBSAN: pointer-overflow\n");
+}
diff --git a/libs/libc/stack.c b/libs/libc/stack.c
new file mode 100644
index 0000000..0cbb69d
--- /dev/null
+++ b/libs/libc/stack.c
@@ -0,0 +1,126 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#include <def.h>
+#include <mem.h>
+#include <stack.h>
+
+static int nonce = 0;
+
+struct stack *stack_new(void)
+{
+	struct stack *stack = malloc(sizeof(*stack));
+	stack->tail = NULL;
+	return stack;
+}
+
+void stack_destroy(struct stack *stack)
+{
+	struct stack_node *iterator = stack->tail;
+	while (iterator) {
+		if (!iterator->prev) {
+			free(iterator);
+			break;
+		}
+		iterator = iterator->prev;
+		free(iterator->next);
+	}
+	stack->tail = NULL;
+	free(stack);
+	stack = NULL;
+}
+
+static struct stack_node *stack_new_node(void)
+{
+	struct stack_node *node = malloc(sizeof(*node));
+	node->data = NULL;
+	node->prev = NULL;
+	node->next = NULL;
+	node->nonce = nonce++;
+	return node;
+}
+
+static u32 stack_push_bot_node(struct stack *stack, struct stack_node *node)
+{
+	if (!stack || !node)
+		return 0;
+
+	if (stack->tail) {
+		struct stack_node *iterator = stack->tail;
+		while (iterator) {
+			if (!iterator->prev)
+				break;
+			iterator = iterator->prev;
+		}
+		iterator->prev = node;
+		node->next = iterator;
+	} else {
+		stack->tail = node;
+	}
+
+	return 1;
+}
+
+static u32 stack_push_node(struct stack *stack, struct stack_node *node)
+{
+	if (!stack || !node)
+		return 0;
+
+	if (stack->tail) {
+		stack->tail->next = node;
+		node->prev = stack->tail;
+		stack->tail = node;
+	} else {
+		stack->tail = node;
+	}
+
+	return 1;
+}
+
+u32 stack_empty(struct stack *stack)
+{
+	return !stack->tail;
+}
+
+u32 stack_push_bot(struct stack *stack, void *data)
+{
+	struct stack_node *node = stack_new_node();
+	node->data = data;
+	return stack_push_bot_node(stack, node);
+}
+
+u32 stack_push(struct stack *stack, void *data)
+{
+	struct stack_node *node = stack_new_node();
+	node->data = data;
+	return stack_push_node(stack, node);
+}
+
+void *stack_pop(struct stack *stack)
+{
+	if (!stack || !stack->tail)
+		return NULL;
+
+	struct stack_node *prev = stack->tail;
+
+	if (stack->tail->prev)
+		stack->tail->prev->next = NULL;
+	stack->tail = stack->tail->prev;
+
+	void *data = prev->data;
+	free(prev);
+	return data;
+}
+
+void *stack_peek(struct stack *stack)
+{
+	if (!stack || !stack->tail)
+		return NULL;
+
+	return stack->tail->data;
+}
+
+void stack_clear(struct stack *stack)
+{
+	while (stack_pop(stack))
+		;
+}
diff --git a/libs/libc/str.c b/libs/libc/str.c
new file mode 100644
index 0000000..ba16920
--- /dev/null
+++ b/libs/libc/str.c
@@ -0,0 +1,207 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#include <def.h>
+#include <errno.h>
+#include <mem.h>
+#include <str.h>
+
+u32 strlen(const char *s)
+{
+	const char *ss = s;
+	while (*ss)
+		ss++;
+	return ss - s;
+}
+
+char *strcpy(char *dst, const char *src)
+{
+	char *q = dst;
+	const char *p = src;
+	char ch;
+
+	do {
+		*q++ = ch = *p++;
+	} while (ch);
+
+	return dst;
+}
+
+char *strncpy(char *dst, const char *src, u32 n)
+{
+	char *q = dst;
+
+	while (n-- && (*dst++ = *src++))
+		;
+
+	return q;
+}
+
+int strcmp(const char *s1, const char *s2)
+{
+	const u8 *c1 = (const u8 *)s1;
+	const u8 *c2 = (const u8 *)s2;
+	u8 ch;
+	int d = 0;
+
+	while (1) {
+		d = (int)(ch = *c1++) - (int)*c2++;
+		if (d || !ch)
+			break;
+	}
+
+	return d;
+}
+
+int strncmp(const char *s1, const char *s2, u32 n)
+{
+	const u8 *c1 = (const u8 *)s1;
+	const u8 *c2 = (const u8 *)s2;
+	u8 ch;
+	int d = 0;
+
+	while (n--) {
+		d = (int)(ch = *c1++) - (int)*c2++;
+		if (d || !ch)
+			break;
+	}
+
+	return d;
+}
+
+char *strchr(char *s, int c)
+{
+	while (*s != (char)c) {
+		if (!*s)
+			return NULL;
+		s++;
+	}
+
+	return s;
+}
+
+char *strrchr(char *s, int c)
+{
+	char *ret = 0;
+
+	do {
+		if (*s == c)
+			ret = s;
+	} while (*s++);
+
+	return ret;
+}
+
+char *strcat(char *dst, const char *src)
+{
+	strcpy(strchr(dst, '\0'), src);
+	return dst;
+}
+
+char *strncat(char *dst, const char *src, u32 n)
+{
+	strncpy(strchr(dst, '\0'), src, n);
+	return dst;
+}
+
+char *strinv(char *s)
+{
+	u32 s_str = strlen(s);
+
+	int iterations = (int)s_str / 2;
+	for (int i = 0; i < iterations; i++) {
+		char aux = s[i];
+		s[i] = s[(s_str - i) - 1];
+		s[(s_str - i) - 1] = aux;
+	}
+	return s;
+}
+
+char *strdup(const char *s)
+{
+	int l = strlen(s) + 1;
+	char *d = malloc(l);
+
+	memcpy(d, s, l);
+
+	return d;
+}
+
+const char *strerror(u32 error)
+{
+	switch (error) {
+	case 0:
+		return "Success";
+	case EPERM:
+		return "Operation not permitted";
+	case ENOENT:
+		return "No such file or directory";
+	case ESRCH:
+		return "No such process";
+	case EINTR:
+		return "Interrupted system call";
+	case EIO:
+		return "I/O error";
+	case ENXIO:
+		return "No such device or address";
+	case E2BIG:
+		return "Argument list too long";
+	case ENOEXEC:
+		return "Exec format error";
+	case EBADF:
+		return "Bad file number";
+	case ECHILD:
+		return "No child processes";
+	case EAGAIN:
+		return "Try again";
+	case ENOMEM:
+		return "Out of memory";
+	case EACCES:
+		return "Permission denied";
+	case EFAULT:
+		return "Bad address";
+	case ENOTBLK:
+		return "Block device required";
+	case EBUSY:
+		return "Device or resource busy";
+	case EEXIST:
+		return "File exists";
+	case EXDEV:
+		return "Cross-device link";
+	case ENODEV:
+		return "No such device";
+	case ENOTDIR:
+		return "Not a directory";
+	case EISDIR:
+		return "Is a directory";
+	case EINVAL:
+		return "Invalid argument";
+	case ENFILE:
+		return "File table overflow";
+	case EMFILE:
+		return "Too many open files";
+	case ENOTTY:
+		return "Not a typewriter";
+	case ETXTBSY:
+		return "Text file busy";
+	case EFBIG:
+		return "File too large";
+	case ENOSPC:
+		return "No space left on device";
+	case ESPIPE:
+		return "Illegal seek";
+	case EROFS:
+		return "Read-only file system";
+	case EMLINK:
+		return "Too many links";
+	case EPIPE:
+		return "Broken pipe";
+	case EDOM:
+		return "Math argument out of domain of func";
+	case ERANGE:
+		return "Math result not representable";
+	case EMAX:
+		return "Max errno";
+	default:
+		return "Unknown error";
+	}
+}
diff --git a/libs/libc/sys.c b/libs/libc/sys.c
new file mode 100644
index 0000000..491c37c
--- /dev/null
+++ b/libs/libc/sys.c
@@ -0,0 +1,203 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+// Syscall implementation
+
+#include <arg.h>
+#include <assert.h>
+#include <errno.h>
+#include <sys.h>
+
+#if defined(userspace)
+
+/**
+ * Definitions
+ */
+
+#define ERRIFY(ret)                                                                                \
+	if (ret < 0) {                                                                             \
+		errno = -ret;                                                                      \
+		return -1;                                                                         \
+	}                                                                                          \
+	errno = 0;                                                                                 \
+	return ret
+
+res sys0(enum sys num);
+res sys0(enum sys num)
+{
+	int a;
+	__asm__ volatile("int $0x80" : "=a"(a) : "0"(num));
+	ERRIFY(a);
+}
+
+res sys1(enum sys num, int d1);
+res sys1(enum sys num, int d1)
+{
+	int a;
+	__asm__ volatile("int $0x80" : "=a"(a) : "0"(num), "b"((int)d1));
+	ERRIFY(a);
+}
+
+res sys2(enum sys num, int d1, int d2);
+res sys2(enum sys num, int d1, int d2)
+{
+	int a;
+	__asm__ volatile("int $0x80" : "=a"(a) : "0"(num), "b"((int)d1), "c"((int)d2));
+	ERRIFY(a);
+}
+
+res sys3(enum sys num, int d1, int d2, int d3);
+res sys3(enum sys num, int d1, int d2, int d3)
+{
+	int a;
+	__asm__ volatile("int $0x80"
+			 : "=a"(a)
+			 : "0"(num), "b"((int)d1), "c"((int)d2), "d"((int)d3));
+	ERRIFY(a);
+}
+
+res sys4(enum sys num, int d1, int d2, int d3, int d4);
+res sys4(enum sys num, int d1, int d2, int d3, int d4)
+{
+	int a;
+	__asm__ volatile("int $0x80"
+			 : "=a"(a)
+			 : "0"(num), "b"((int)d1), "c"((int)d2), "d"((int)d3), "S"((int)d4));
+	ERRIFY(a);
+}
+
+res sys5(enum sys num, int d1, int d2, int d3, int d4, int d5);
+res sys5(enum sys num, int d1, int d2, int d3, int d4, int d5)
+{
+	int a;
+	__asm__ volatile("int $0x80"
+			 : "=a"(a)
+			 : "0"(num), "b"((int)d1), "c"((int)d2), "d"((int)d3), "S"((int)d4),
+			   "D"((int)d5));
+	ERRIFY(a);
+}
+
+/**
+ * Syscalls
+ */
+
+res sys_alloc(u32 size, u32 *addr)
+{
+	u32 id = 0;
+	return sys4(SYS_ALLOC, (int)size, (int)addr, (int)&id, 0);
+}
+
+res sys_free(void *ptr)
+{
+	return sys1(SYS_FREE, (int)ptr);
+}
+
+res shalloc(u32 size, u32 *addr, u32 *id)
+{
+	return sys4(SYS_ALLOC, (int)size, (int)addr, (int)id, 1);
+}
+
+res shaccess(u32 id, u32 *addr, u32 *size)
+{
+	return sys3(SYS_SHACCESS, (int)id, (int)addr, (int)size);
+}
+
+void loop(void)
+{
+	sys0(SYS_LOOP);
+}
+
+res read(const char *path, void *buf, u32 offset, u32 count)
+{
+	return sys4(SYS_READ, (int)path, (int)buf, (int)offset, (int)count);
+}
+
+res write(const char *path, const void *buf, u32 offset, u32 count)
+{
+	return sys4(SYS_WRITE, (int)path, (int)buf, (int)offset, (int)count);
+}
+
+res ioctl(const char *path, ...)
+{
+	va_list ap;
+	int args[4] = { 0 };
+
+	va_start(ap, path);
+	for (int i = 0; i < 4; i++)
+		args[i] = va_arg(ap, int);
+	va_end(ap);
+
+	return sys5(SYS_IOCTL, (int)path, args[0], args[1], args[2], args[3]);
+}
+
+res stat(const char *path, struct stat *buf)
+{
+	return sys2(SYS_STAT, (int)path, (int)buf);
+}
+
+res poll(const char **files)
+{
+	return sys1(SYS_POLL, (int)files);
+}
+
+res exec(const char *path, ...)
+{
+	va_list ap;
+	int args[4] = { 0 };
+
+	va_start(ap, path);
+	for (int i = 0; i < 4; i++)
+		args[i] = va_arg(ap, int);
+	va_end(ap);
+
+	return sys5(SYS_EXEC, (int)path, args[0], args[1], args[2], args[3]);
+}
+
+res yield(void)
+{
+	return sys0(SYS_YIELD);
+}
+
+static void atexit_trigger(void);
+void exit(s32 status)
+{
+	atexit_trigger();
+	sys1(SYS_EXIT, (int)status);
+	while (1)
+		yield();
+}
+
+res boot(u32 cmd)
+{
+	return sys2(SYS_BOOT, SYS_BOOT_MAGIC, cmd);
+}
+
+u32 time(void)
+{
+	return (u32)sys0(SYS_TIME);
+}
+
+/**
+ * At exit
+ */
+
+#define ATEXIT_MAX 32
+
+static u32 slot = 0;
+static void (*funcs[ATEXIT_MAX])(void) = { 0 };
+
+static void atexit_trigger(void)
+{
+	while (slot-- > 0) {
+		if (funcs[slot]) {
+			funcs[slot]();
+			funcs[slot] = NULL;
+		}
+	}
+}
+
+void atexit(void (*func)(void))
+{
+	if (slot < ATEXIT_MAX)
+		funcs[slot++] = func;
+}
+
+#endif
diff --git a/libs/libgui/Makefile b/libs/libgui/Makefile
new file mode 100644
index 0000000..3e2b0b2
--- /dev/null
+++ b/libs/libgui/Makefile
@@ -0,0 +1,25 @@
+# MIT License, Copyright (c) 2020 Marvin Borner
+
+COBJS = psf.o \
+	bmp.o \
+	png.o \
+	gfx.o \
+	gui.o \
+	msg.o
+CC = ccache ../../cross/opt/bin/i686-elf-gcc
+LD = ccache ../../cross/opt/bin/i686-elf-ld
+AR = ccache ../../cross/opt/bin/i686-elf-ar
+
+CFLAGS = $(CFLAGS_DEFAULT) -I../ -I../libc/inc/ -Duserspace
+
+all: libgui
+
+%.o: %.c
+	@$(CC) -c $(CFLAGS) $< -o $@
+
+libgui: $(COBJS)
+	@mkdir -p ../../build/
+	@$(AR) rcs ../../build/libgui.a $+
+
+clean:
+	@find . -name "*.o" -type f -delete
diff --git a/libs/libgui/bmp.c b/libs/libgui/bmp.c
new file mode 100644
index 0000000..5576d16
--- /dev/null
+++ b/libs/libgui/bmp.c
@@ -0,0 +1,29 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#include <def.h>
+#include <libgui/bmp.h>
+#include <mem.h>
+#include <print.h>
+#include <sys.h>
+
+struct bmp *bmp_load(const char *path)
+{
+	void *buf = sread(path);
+	if (!buf)
+		return NULL;
+
+	struct bmp_header *h = buf;
+	if (h->signature[0] != 'B' || h->signature[1] != 'M')
+		return NULL;
+
+	// TODO: Support padding with odd widths
+	struct bmp_info *info = (struct bmp_info *)((u32)buf + sizeof(*h));
+	struct bmp *bmp = malloc(sizeof(*bmp));
+	bmp->size.x = info->width;
+	bmp->size.y = info->height;
+	bmp->data = (u8 *)((u32)buf + h->offset);
+	bmp->bpp = info->bpp;
+	bmp->pitch = bmp->size.x * (bmp->bpp >> 3);
+
+	return bmp;
+}
diff --git a/libs/libgui/bmp.h b/libs/libgui/bmp.h
new file mode 100644
index 0000000..ff8360b
--- /dev/null
+++ b/libs/libgui/bmp.h
@@ -0,0 +1,39 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef BMP_H
+#define BMP_H
+
+#include <def.h>
+#include <vec.h>
+
+struct bmp_header {
+	u8 signature[2];
+	u32 size;
+	u32 reserved;
+	u32 offset;
+} __attribute__((packed));
+
+struct bmp_info {
+	u32 size;
+	u32 width;
+	u32 height;
+	u16 planes;
+	u16 bpp;
+	u32 compression;
+	u32 compressed_size;
+	u32 x_pixel_meter;
+	u32 y_pixel_meter;
+	u32 colors;
+	u32 important_colors;
+};
+
+struct bmp {
+	vec2 size;
+	u8 *data;
+	u32 bpp;
+	u32 pitch;
+};
+
+struct bmp *bmp_load(const char *path);
+
+#endif
diff --git a/libs/libgui/gfx.c b/libs/libgui/gfx.c
new file mode 100644
index 0000000..259b03f
--- /dev/null
+++ b/libs/libgui/gfx.c
@@ -0,0 +1,267 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+// Some GFX functions
+// TODO: Better support for bpp < 32
+// TODO: Use efficient redrawing
+
+#include <assert.h>
+#include <libgui/bmp.h>
+#include <libgui/gfx.h>
+#include <libgui/msg.h>
+#include <libgui/png.h>
+#include <libgui/psf.h>
+#include <libgui/vesa.h>
+#include <mem.h>
+#include <str.h>
+#include <sys.h>
+
+// TODO: Move to some global config file
+#define FONT_COUNT 6
+#define FONT_8_PATH "/font/spleen-5x8.psfu"
+#define FONT_12_PATH "/font/spleen-6x12.psfu"
+#define FONT_16_PATH "/font/spleen-8x16.psfu"
+#define FONT_24_PATH "/font/spleen-12x24.psfu"
+#define FONT_32_PATH "/font/spleen-16x32.psfu"
+#define FONT_64_PATH "/font/spleen-32x64.psfu"
+
+struct font *fonts[FONT_COUNT] = { 0 };
+
+static void load_font(enum font_type font_type)
+{
+	if (fonts[font_type])
+		return;
+
+	const char *path = NULL;
+
+	switch (font_type) {
+	case FONT_8:
+		path = FONT_8_PATH;
+		break;
+	case FONT_12:
+		path = FONT_12_PATH;
+		break;
+	case FONT_16:
+		path = FONT_16_PATH;
+		break;
+	case FONT_24:
+		path = FONT_24_PATH;
+		break;
+	case FONT_32:
+		path = FONT_32_PATH;
+		break;
+	case FONT_64:
+		path = FONT_64_PATH;
+		break;
+	default:
+		return;
+	}
+
+	fonts[font_type] = psf_parse(sread(path));
+	assert(fonts[font_type]);
+}
+
+static void write_char(struct context *ctx, vec2 pos, struct font *font, u32 c, char ch)
+{
+	int bypp = ctx->bpp >> 3;
+
+	char *draw = (char *)&ctx->fb[pos.x * bypp + pos.y * ctx->pitch];
+
+	u32 stride = font->char_size / font->size.y;
+	for (u32 cy = 0; cy < font->size.y; cy++) {
+		for (u32 cx = 0; cx < font->size.x; cx++) {
+			u8 bits = font->chars[ch * font->char_size + cy * stride + cx / 8];
+			u8 bit = bits >> (7 - cx % 8) & 1;
+			if (bit) {
+				draw[bypp * cx] = GET_BLUE(c);
+				draw[bypp * cx + 1] = GET_GREEN(c);
+				draw[bypp * cx + 2] = GET_RED(c);
+				draw[bypp * cx + 3] = GET_ALPHA(c);
+			}
+		}
+		draw += ctx->pitch;
+	}
+}
+
+static void draw_rectangle(struct context *ctx, vec2 pos1, vec2 pos2, u32 c)
+{
+	int bypp = ctx->bpp >> 3;
+	u8 *draw = &ctx->fb[pos1.x * bypp + pos1.y * ctx->pitch];
+	for (u32 i = 0; i < pos2.y - pos1.y; i++) {
+		for (u32 j = 0; j < pos2.x - pos1.x; j++) {
+			draw[bypp * j] = GET_BLUE(c);
+			draw[bypp * j + 1] = GET_GREEN(c);
+			draw[bypp * j + 2] = GET_RED(c);
+			draw[bypp * j + 3] = GET_ALPHA(c);
+		}
+		draw += ctx->pitch;
+	}
+}
+
+struct context *gfx_new_ctx(struct context *ctx)
+{
+	/* struct message msg = { 0 }; */
+	assert(0);
+	/* assert(msg_send(pidof(WM_PATH), GFX_NEW_CONTEXT, ctx) > 0); */
+	/* assert(msg_receive(&msg) > 0); */
+	/* memcpy(ctx, msg.data, sizeof(*ctx)); */
+	return ctx;
+}
+
+// On-demand font loading
+struct font *gfx_resolve_font(enum font_type font_type)
+{
+	if (!fonts[font_type])
+		load_font(font_type);
+	return fonts[font_type];
+}
+
+void gfx_write_char(struct context *ctx, vec2 pos, enum font_type font_type, u32 c, char ch)
+{
+	struct font *font = gfx_resolve_font(font_type);
+	write_char(ctx, pos, font, c, ch);
+}
+
+void gfx_write(struct context *ctx, vec2 pos, enum font_type font_type, u32 c, const char *text)
+{
+	struct font *font = gfx_resolve_font(font_type);
+	u32 cnt = 0;
+	for (u32 i = 0; i < strlen(text); i++) {
+		// TODO: Should this be here?
+		if (text[i] == '\r') {
+			cnt = 0;
+		} else if (text[i] == '\n') {
+			cnt = 0;
+			pos.y += font->size.y;
+		} else if (text[i] == '\t') {
+			cnt += 4;
+		} else {
+			// TODO: Overflow on single line input
+			if ((cnt + 1) * font->size.x > ctx->size.x) {
+				cnt = 0;
+				pos.y += font->size.y;
+			}
+			write_char(ctx, vec2(pos.x + cnt * font->size.x, pos.y), font, c, text[i]);
+			cnt++;
+		}
+	}
+}
+
+void gfx_load_image(struct context *ctx, vec2 pos, const char *path)
+{
+	// TODO: Support x, y
+	// TODO: Detect image type
+	struct bmp bmp = { 0 };
+
+	u32 error = png_decode32_file(&bmp.data, &bmp.size.x, &bmp.size.y, path);
+	if (error)
+		err(1, "error %u: %s\n", error, png_error_text(error));
+
+	assert(bmp.size.x + pos.x <= ctx->size.x);
+	assert(bmp.size.y + pos.y <= ctx->size.y);
+
+	bmp.bpp = 32;
+	bmp.pitch = bmp.size.x * (bmp.bpp >> 3);
+
+	// TODO: Fix reversed png in decoder
+	int bypp = bmp.bpp >> 3;
+	// u8 *srcfb = &bmp->data[bypp + (bmp->size.y - 1) * bmp->pitch];
+	u8 *srcfb = bmp.data;
+	u8 *destfb = &ctx->fb[bypp];
+	for (u32 cy = 0; cy < bmp.size.y; cy++) {
+		memcpy(destfb, srcfb, bmp.pitch);
+		// srcfb -= bmp->pitch;
+		srcfb += bmp.pitch;
+		destfb += ctx->pitch;
+	}
+}
+
+void gfx_load_wallpaper(struct context *ctx, const char *path)
+{
+	gfx_load_image(ctx, vec2(0, 0), path);
+}
+
+void gfx_copy(struct context *dest, struct context *src, vec2 pos, vec2 size)
+{
+	int bypp = dest->bpp >> 3;
+	u8 *srcfb = &src->fb[pos.x * bypp + pos.y * src->pitch];
+	u8 *destfb = &dest->fb[pos.x * bypp + pos.y * dest->pitch];
+	for (u32 cy = 0; cy < size.y; cy++) {
+		memcpy(destfb, srcfb, size.x * bypp);
+		srcfb += src->pitch;
+		destfb += dest->pitch;
+	}
+}
+
+// TODO: Support alpha values other than 0x0 and 0xff (blending)
+// TODO: Optimize!
+void gfx_ctx_on_ctx(struct context *dest, struct context *src, vec2 pos)
+{
+	if (src->size.x == dest->size.x && src->size.y == dest->size.y) {
+		memcpy(dest->fb, src->fb, dest->pitch * dest->size.y);
+		return;
+	}
+
+	if (src->size.x > dest->size.x || src->size.y > dest->size.y)
+		return;
+
+	// TODO: Negative x and y
+	int bypp = dest->bpp >> 3;
+	u8 *srcfb = src->fb;
+	u8 *destfb = &dest->fb[pos.x * bypp + pos.y * dest->pitch];
+	for (u32 cy = 0; cy < src->size.y && cy + pos.y < dest->size.y; cy++) {
+		int diff = 0;
+		for (u32 cx = 0; cx < src->size.x && cx + pos.x < dest->size.x; cx++) {
+			if (srcfb[bypp - 1])
+				memcpy(destfb, srcfb, bypp);
+
+			srcfb += bypp;
+			destfb += bypp;
+			diff += bypp;
+		}
+		srcfb += src->pitch - diff;
+		destfb += dest->pitch - diff;
+	}
+}
+
+void gfx_draw_rectangle(struct context *ctx, vec2 pos1, vec2 pos2, u32 c)
+{
+	draw_rectangle(ctx, pos1, pos2, c);
+}
+
+void gfx_fill(struct context *ctx, u32 c)
+{
+	draw_rectangle(ctx, vec2(0, 0), vec2(ctx->size.x, ctx->size.y), c);
+}
+
+void gfx_border(struct context *ctx, u32 c, u32 width)
+{
+	if (width <= 0)
+		return;
+
+	int bypp = ctx->bpp >> 3;
+	u8 *draw = ctx->fb;
+	for (u32 i = 0; i < ctx->size.y; i++) {
+		for (u32 j = 0; j < ctx->size.x; j++) {
+			if (j <= width - 1 || i <= width - 1 ||
+			    j - ctx->size.x + width + 1 <= width ||
+			    i - ctx->size.y + width <= width) {
+				draw[bypp * j + 0] = GET_BLUE(c);
+				draw[bypp * j + 1] = GET_GREEN(c);
+				draw[bypp * j + 2] = GET_RED(c);
+				draw[bypp * j + 3] = GET_ALPHA(c);
+			}
+		}
+		draw += ctx->pitch;
+	}
+}
+
+int gfx_font_height(enum font_type font_type)
+{
+	struct font *font = gfx_resolve_font(font_type);
+	return font->size.y;
+}
+
+int gfx_font_width(enum font_type font_type)
+{
+	struct font *font = gfx_resolve_font(font_type);
+	return font->size.x;
+}
diff --git a/libs/libgui/gfx.h b/libs/libgui/gfx.h
new file mode 100644
index 0000000..8ac01d0
--- /dev/null
+++ b/libs/libgui/gfx.h
@@ -0,0 +1,88 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+// Some GFX functions
+
+#ifndef GFX_H
+#define GFX_H
+
+#include <def.h>
+#include <libgui/vesa.h>
+#include <sys.h>
+#include <vec.h>
+
+#define WM_PATH "/bin/wm"
+
+#define GET_ALPHA(color) ((color >> 24) & 0x000000FF)
+#define GET_RED(color) ((color >> 16) & 0x000000FF)
+#define GET_GREEN(color) ((color >> 8) & 0x000000FF)
+#define GET_BLUE(color) ((color >> 0) & 0X000000FF)
+
+#define COLOR_TRANSPARENT 0x00000000
+#define COLOR_INVISIBLE 0x00000000
+#define COLOR_BLACK 0xff0f0f0f
+#define COLOR_RED 0xfff07f7f
+#define COLOR_GREEN 0xff7ff088
+#define COLOR_YELLOW 0xffeef07f
+#define COLOR_BLUE 0xff7facf0
+#define COLOR_MAGENTA 0xffd67ff0
+#define COLOR_CYAN 0xff7fe7f0
+#define COLOR_WHITE 0xffe9e9e9
+#define COLOR_BRIGHT_BLACK 0xff928374
+#define COLOR_BRIGHT_RED 0xffed9a9a
+#define COLOR_BRIGHT_GREEN 0xff9ef0a5
+#define COLOR_BRIGHT_YELLOW 0xffe7e897
+#define COLOR_BRIGHT_BLUE 0xff98b9eb
+#define COLOR_BRIGHT_MAGENTA 0xffd196e3
+#define COLOR_BRIGHT_CYAN 0xff94dae0
+#define COLOR_BRIGHT_WHITE 0xffe3e3e3
+#define COLOR_FG COLOR_WHITE
+#define COLOR_BG COLOR_BLACK
+
+#define WF_DEFAULT (0 << 0)
+#define WF_NO_FOCUS (1 << 0)
+#define WF_NO_DRAG (1 << 1)
+#define WF_NO_RESIZE (1 << 2)
+#define WF_NO_FB (1 << 3)
+#define WF_NO_WINDOW (1 << 4)
+/* #define WF_RELATIVE (1 << 4) */
+
+enum font_type { FONT_8, FONT_12, FONT_16, FONT_24, FONT_32, FONT_64 };
+
+// Generalized font struct
+struct font {
+	char *chars;
+	vec2 size;
+	int char_size;
+};
+
+struct context {
+	vec2 size;
+	u8 *fb;
+	u32 bpp;
+	u32 pitch;
+	u32 bytes;
+};
+
+struct context *gfx_new_ctx(struct context *ctx);
+struct font *gfx_resolve_font(enum font_type font_type);
+void gfx_write_char(struct context *ctx, vec2 pos, enum font_type font_type, u32 c, char ch);
+void gfx_write(struct context *ctx, vec2 pos, enum font_type font_type, u32 c, const char *text);
+void gfx_load_image(struct context *ctx, vec2 pos, const char *path);
+void gfx_load_wallpaper(struct context *ctx, const char *path);
+void gfx_copy(struct context *dest, struct context *src, vec2 pos, vec2 size);
+void gfx_ctx_on_ctx(struct context *dest, struct context *src, vec2 pos);
+void gfx_draw_rectangle(struct context *ctx, vec2 pos1, vec2 pos2, u32 c);
+void gfx_fill(struct context *ctx, u32 c);
+void gfx_border(struct context *ctx, u32 c, u32 width);
+
+int gfx_font_height(enum font_type);
+int gfx_font_width(enum font_type);
+
+/**
+ * Wrappers
+ */
+
+/* #define gfx_redraw()                                                                               \ */
+/* 	(msg_send(pidof(WM_PATH), GFX_REDRAW, NULL)) // TODO: Partial redraw (optimization) */
+/* #define gfx_redraw_focused() (msg_send(pidof(WM_PATH), GFX_REDRAW_FOCUSED, NULL)) */
+
+#endif
diff --git a/libs/libgui/gui.c b/libs/libgui/gui.c
new file mode 100644
index 0000000..d8435e6
--- /dev/null
+++ b/libs/libgui/gui.c
@@ -0,0 +1,153 @@
+// MIT License, Copyright (c) 2021 Marvin Borner
+
+#include <assert.h>
+#include <def.h>
+#include <errno.h>
+#include <libgui/gui.h>
+#include <libgui/msg.h>
+#include <list.h>
+#include <print.h>
+
+#define WM_PATH "/bin/wm"
+
+struct gui_window {
+	u32 id;
+	struct context ctx;
+};
+
+struct list *windows = NULL;
+
+static struct gui_window *win_by_id(u32 id)
+{
+	assert(windows);
+	struct node *iterator = windows->head;
+	while (iterator) {
+		struct gui_window *win = iterator->data;
+		if (win->id == id)
+			return iterator->data;
+		iterator = iterator->next;
+	}
+	return NULL;
+}
+
+/**
+ * GFX wrappers
+ */
+
+static res fill(u32 id, u32 c)
+{
+	struct gui_window *win = win_by_id(id);
+	if (!win)
+		return -ENOENT;
+	gfx_fill(&win->ctx, c);
+	return EOK;
+}
+
+/**
+ * Program interface
+ */
+
+res gui_new_window(void)
+{
+	if (!windows)
+		windows = list_new();
+
+	struct gui_window *win = zalloc(sizeof(*win));
+
+	struct message_new_window msg = { .header.state = MSG_NEED_ANSWER };
+	if (msg_send(pidof(WM_PATH), GUI_NEW_WINDOW, &msg, sizeof(msg)) > 0 &&
+	    msg_receive(&msg, sizeof(msg)) > 0 &&
+	    msg.header.type == (GUI_NEW_WINDOW | MSG_SUCCESS)) {
+		win->id = msg.id;
+		win->ctx = msg.ctx;
+		u32 size;
+		res ret = shaccess(msg.shid, (u32 *)&win->ctx.fb, &size);
+		if (ret < 0 || !win->ctx.fb)
+			return MIN(ret, -EFAULT);
+		list_add(windows, win);
+		assert(fill(win->id, COLOR_BLACK) == EOK);
+		gui_redraw_window(win->id);
+		return win->id;
+	}
+	return -EINVAL;
+}
+
+res gui_redraw_window(u32 id)
+{
+	struct message_redraw_window msg = { .id = id, .header.state = MSG_NEED_ANSWER };
+	if (msg_send(pidof(WM_PATH), GUI_REDRAW_WINDOW, &msg, sizeof(msg)) > 0 &&
+	    msg_receive(&msg, sizeof(msg)) > 0 &&
+	    msg.header.type == (GUI_REDRAW_WINDOW | MSG_SUCCESS))
+		return id;
+	return -EINVAL;
+}
+
+/**
+ * Message handling
+ */
+
+static res handle_error(const char *op, res code)
+{
+	log("GUI error at '%s': %s (%d)\n", op, strerror(code), code);
+	return code;
+}
+
+static res handle_ping(struct message_ping *msg)
+{
+	if (msg->ping != MSG_PING_SEND)
+		return handle_error("ping", EINVAL);
+
+	msg->header.type |= MSG_SUCCESS;
+	msg->ping = MSG_PING_RECV;
+	msg_send(msg->header.src, GUI_PING, &msg, sizeof(msg));
+
+	return errno;
+}
+
+static res handle_mouse(struct message_mouse *msg)
+{
+	if (msg->header.state == MSG_NEED_ANSWER)
+		msg_send(msg->header.src, msg->header.type | MSG_SUCCESS, msg, sizeof(*msg));
+
+	return errno;
+}
+
+static void handle_exit(void)
+{
+	if (!windows)
+		return;
+
+	struct node *iterator = windows->head;
+	while (iterator) {
+		struct gui_window *win = iterator->data;
+		struct message_destroy_window msg = { .id = win->id };
+		msg_send(pidof(WM_PATH), GUI_DESTROY_WINDOW, &msg, sizeof(msg));
+		iterator = iterator->next;
+	}
+
+	list_destroy(windows);
+}
+
+/**
+ * Main loop
+ */
+
+void gui_loop(void)
+{
+	atexit(handle_exit);
+
+	void *msg = zalloc(4096);
+	while (msg_receive(msg, 4096)) {
+		struct message_header *head = msg;
+		switch (head->type) {
+		case GUI_PING:
+			handle_ping(msg);
+			break;
+		case GUI_MOUSE:
+			handle_mouse(msg);
+			break;
+		default:
+			handle_error("loop", EINVAL);
+		}
+	}
+}
diff --git a/libs/libgui/gui.h b/libs/libgui/gui.h
new file mode 100644
index 0000000..a3f74f8
--- /dev/null
+++ b/libs/libgui/gui.h
@@ -0,0 +1,13 @@
+// MIT License, Copyright (c) 2021 Marvin Borner
+
+#ifndef GUI_H
+#define GUI_H
+
+#include <def.h>
+#include <errno.h>
+
+res gui_new_window(void);
+res gui_redraw_window(u32 id);
+void gui_loop(void);
+
+#endif
diff --git a/libs/libgui/msg.c b/libs/libgui/msg.c
new file mode 100644
index 0000000..73af242
--- /dev/null
+++ b/libs/libgui/msg.c
@@ -0,0 +1,31 @@
+// MIT License, Copyright (c) 2021 Marvin Borner
+
+#include <assert.h>
+#include <errno.h>
+#include <libgui/msg.h>
+#include <print.h>
+#include <sys.h>
+
+res msg_send(u32 pid, enum message_type type, void *data, u32 size)
+{
+	if (!data)
+		return -EFAULT;
+	assert((signed)pid != -1 && size >= sizeof(struct message_header));
+	char path[32] = { 0 };
+	sprintf(path, "/proc/%d/msg", pid);
+	struct message_header *header = data;
+	header->magic = MSG_MAGIC;
+	header->src = getpid();
+	header->type = type;
+	return write(path, data, 0, size);
+}
+
+res msg_receive(void *buf, u32 size)
+{
+	int ret = read("/proc/self/msg", buf, 0, size);
+	struct message_header *header = buf;
+	if (header->magic == MSG_MAGIC)
+		return ret;
+	else
+		return -1;
+}
diff --git a/libs/libgui/msg.h b/libs/libgui/msg.h
new file mode 100644
index 0000000..7326135
--- /dev/null
+++ b/libs/libgui/msg.h
@@ -0,0 +1,71 @@
+// MIT License, Copyright (c) 2021 Marvin Borner
+
+#ifndef MSG_H
+#define MSG_H
+
+#include <def.h>
+#include <libgui/gfx.h>
+
+#define MSG_PING_SEND 0x07734
+#define MSG_PING_RECV 0x7474
+
+#define MSG_MAGIC 0x42042069
+#define MSG_SUCCESS (1 << 29)
+#define MSG_FAILURE (1 << 30)
+
+enum message_state {
+	MSG_GO_ON,
+	MSG_NEED_ANSWER,
+};
+
+struct message_header {
+	u32 magic;
+	u32 src;
+	u32 type;
+	enum message_state state;
+};
+
+struct message_ping {
+	struct message_header header;
+	u32 ping;
+};
+
+struct message_new_window {
+	struct message_header header;
+	u32 id;
+	u32 shid;
+	struct context ctx;
+};
+
+struct message_redraw_window {
+	struct message_header header;
+	u32 id;
+};
+
+struct message_destroy_window {
+	struct message_header header;
+	u32 id;
+};
+
+struct message_mouse {
+	struct message_header header;
+	vec2 pos;
+	struct {
+		u8 click : 1;
+	} bits;
+};
+
+enum message_type {
+	GUI_PING,
+	GUI_NEW_WINDOW,
+	GUI_REDRAW_WINDOW,
+	GUI_DESTROY_WINDOW,
+
+	GUI_MOUSE,
+	GUI_KEYBOARD,
+};
+
+res msg_send(u32 pid, enum message_type type, void *data, u32 size);
+res msg_receive(void *buf, u32 size);
+
+#endif
diff --git a/libs/libgui/png.c b/libs/libgui/png.c
new file mode 100644
index 0000000..6f8f4b5
--- /dev/null
+++ b/libs/libgui/png.c
@@ -0,0 +1,7472 @@
+/*
+LodePNG version 20201017
+
+Copyright (c) 2005-2020 Lode Vandevenne
+Copyright (c) 2021 Marvin Borner
+
+This software is provided 'as-is', without any express or implied
+warranty. In no event will the authors be held liable for any damages
+arising from the use of this software.
+
+Permission is granted to anyone to use this software for any purpose,
+including commercial applications, and to alter it and redistribute it
+freely, subject to the following restrictions:
+
+    1. The origin of this software must not be misrepresented; you must not
+    claim that you wrote the original software. If you use this software
+    in a product, an acknowledgment in the product documentation would be
+    appreciated but is not required.
+
+    2. Altered source versions must be plainly marked as such, and must not be
+    misrepresented as being the original software.
+
+    3. This notice may not be removed or altered from any source
+    distribution.
+*/
+
+/*
+The manual and changelog are in the header file "png.h"
+Rename this file to png.cpp to use it for C++, or to png.c to use it for C.
+*/
+
+#include <def.h>
+#include <libgui/png.h>
+#include <mem.h>
+#include <sys.h>
+
+#if defined(_MSC_VER) &&                                                                           \
+	(_MSC_VER >= 1310) /*Visual Studio: A few warning types are not desired here.*/
+#pragma warning(                                                                                   \
+	disable : 4244) /*implicit conversions: not warned by gcc -Wall -Wextra and requires too much casts*/
+#pragma warning(                                                                                   \
+	disable : 4996) /*VS does not like fopen, but fopen_s is not standard C so unusable here*/
+#endif /*_MSC_VER */
+
+const char *PNG_VERSION_STRING = "20201017";
+
+/*
+This source file is built up in the following large parts. The code sections
+with the "PNG_COMPILE_" #defines divide this up further in an intermixed way.
+-Tools for C and common code for PNG and Zlib
+-C Code for Zlib (huffman, deflate, ...)
+-C Code for PNG (file format chunks, adam7, PNG filters, color conversions, ...)
+-The C++ wrapper around all of the above
+*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* // Tools for C, and common code for PNG and Zlib.                       // */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*The malloc, realloc and free functions defined here with "png_" in front
+of the name, so that you can easily change them to others related to your
+platform if needed. Everything else in the code calls these. Pass
+-DPNG_NO_COMPILE_ALLOCATORS to the compiler, or comment out
+#define PNG_COMPILE_ALLOCATORS in the header, to disable the ones here and
+define them in your own project's source files without needing to change
+png source code. Don't forget to remove "static" if you copypaste them
+from here.*/
+
+#ifdef PNG_COMPILE_ALLOCATORS
+static void *png_malloc(u32 size)
+{
+#ifdef PNG_MAX_ALLOC
+	if (size > PNG_MAX_ALLOC)
+		return 0;
+#endif
+	return malloc(size);
+}
+
+/* NOTE: when realloc returns NULL, it leaves the original memory untouched */
+static void *png_realloc(void *ptr, u32 new_size)
+{
+#ifdef PNG_MAX_ALLOC
+	if (new_size > PNG_MAX_ALLOC)
+		return 0;
+#endif
+	return realloc(ptr, new_size);
+}
+
+static void png_free(void *ptr)
+{
+	free(ptr);
+}
+#else /*PNG_COMPILE_ALLOCATORS*/
+/* TODO: support giving additional void* payload to the custom allocators */
+void *png_malloc(u32 size);
+void *png_realloc(void *ptr, u32 new_size);
+void png_free(void *ptr);
+#endif /*PNG_COMPILE_ALLOCATORS*/
+
+/* convince the compiler to inline a function, for use when this measurably improves performance */
+/* inline is not available in C90, but use it when supported by the compiler */
+#if (defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L)) ||                                \
+	(defined(__cplusplus) && (__cplusplus >= 199711L))
+#define PNG_INLINE inline
+#else
+#define PNG_INLINE /* not available */
+#endif
+
+/* restrict is not available in C90, but use it when supported by the compiler */
+#if (defined(__GNUC__) && (__GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ >= 1))) ||             \
+	(defined(_MSC_VER) && (_MSC_VER >= 1400)) ||                                               \
+	(defined(__WATCOMC__) && (__WATCOMC__ >= 1250) && !defined(__cplusplus))
+#define PNG_RESTRICT __restrict
+#else
+#define PNG_RESTRICT /* not available */
+#endif
+
+/* Replacements for C library functions such as memcpy and strlen, to support platforms
+where a full C library is not available. The compiler can recognize them and compile
+to something as fast. */
+
+static void png_memcpy(void *PNG_RESTRICT dst, const void *PNG_RESTRICT src, u32 size)
+{
+	u32 i;
+	for (i = 0; i < size; i++)
+		((char *)dst)[i] = ((const char *)src)[i];
+}
+
+static void png_memset(void *PNG_RESTRICT dst, int value, u32 num)
+{
+	u32 i;
+	for (i = 0; i < num; i++)
+		((char *)dst)[i] = (char)value;
+}
+
+/* does not check memory out of bounds, do not use on untrusted data */
+static u32 png_strlen(const char *a)
+{
+	const char *orig = a;
+	/* avoid warning about unused function in case of disabled COMPILE... macros */
+	(void)(&png_strlen);
+	while (*a)
+		a++;
+	return (u32)(a - orig);
+}
+
+#define PNG_MAX(a, b) (((a) > (b)) ? (a) : (b))
+#define PNG_MIN(a, b) (((a) < (b)) ? (a) : (b))
+#define PNG_ABS(x) ((x) < 0 ? -(x) : (x))
+
+#if defined(PNG_COMPILE_PNG) || defined(PNG_COMPILE_DECODER)
+/* Safely check if adding two integers will overflow (no undefined
+behavior, compiler removing the code, etc...) and output result. */
+static int png_addofl(u32 a, u32 b, u32 *result)
+{
+	*result = a + b; /* u32 addition is well defined and safe in C90 */
+	return *result < a;
+}
+#endif /*defined(PNG_COMPILE_PNG) || defined(PNG_COMPILE_DECODER)*/
+
+#ifdef PNG_COMPILE_DECODER
+/* Safely check if multiplying two integers will overflow (no undefined
+behavior, compiler removing the code, etc...) and output result. */
+static int png_mulofl(u32 a, u32 b, u32 *result)
+{
+	*result = a * b; /* u32 multiplication is well defined and safe in C90 */
+	return (a != 0 && *result / a != b);
+}
+
+#ifdef PNG_COMPILE_ZLIB
+/* Safely check if a + b > c, even if overflow could happen. */
+static int png_gtofl(u32 a, u32 b, u32 c)
+{
+	u32 d;
+	if (png_addofl(a, b, &d))
+		return 1;
+	return d > c;
+}
+#endif /*PNG_COMPILE_ZLIB*/
+#endif /*PNG_COMPILE_DECODER*/
+
+/*
+Often in case of an error a value is assigned to a variable and then it breaks
+out of a loop (to go to the cleanup phase of a function). This macro does that.
+It makes the error handling code shorter and more readable.
+
+Example: if(!uivector_resize(&lz77_encoded, datasize)) ERROR_BREAK(83);
+*/
+#define CERROR_BREAK(errorvar, code)                                                               \
+	{                                                                                          \
+		errorvar = code;                                                                   \
+		break;                                                                             \
+	}
+
+/*version of CERROR_BREAK that assumes the common case where the error variable is named "error"*/
+#define ERROR_BREAK(code) CERROR_BREAK(error, code)
+
+/*Set error var to the error code, and return it.*/
+#define CERROR_RETURN_ERROR(errorvar, code)                                                        \
+	{                                                                                          \
+		errorvar = code;                                                                   \
+		return code;                                                                       \
+	}
+
+/*Try the code, if it returns error, also return the error.*/
+#define CERROR_TRY_RETURN(call)                                                                    \
+	{                                                                                          \
+		u32 error = call;                                                                  \
+		if (error)                                                                         \
+			return error;                                                              \
+	}
+
+/*Set error var to the error code, and return from the void function.*/
+#define CERROR_RETURN(errorvar, code)                                                              \
+	{                                                                                          \
+		errorvar = code;                                                                   \
+		return;                                                                            \
+	}
+
+/*
+About uivector, ucvector and string:
+-All of them wrap dynamic arrays or text strings in a similar way.
+-png was originally written in C++. The vectors replace the std::vectors that were used in the C++ version.
+-The string tools are made to avoid problems with compilers that declare things like strncat as deprecated.
+-They're not used in the interface, only internally in this file as static functions.
+-As with many other structs in this file, the init and cleanup functions serve as ctor and dtor.
+*/
+
+#ifdef PNG_COMPILE_ZLIB
+#ifdef PNG_COMPILE_ENCODER
+/*dynamic vector of u32 ints*/
+typedef struct uivector {
+	u32 *data;
+	u32 size; /*size in number of u32 longs*/
+	u32 allocsize; /*allocated size in bytes*/
+} uivector;
+
+static void uivector_cleanup(void *p)
+{
+	((uivector *)p)->size = ((uivector *)p)->allocsize = 0;
+	png_free(((uivector *)p)->data);
+	((uivector *)p)->data = NULL;
+}
+
+/*returns 1 if success, 0 if failure ==> nothing done*/
+static u32 uivector_resize(uivector *p, u32 size)
+{
+	u32 allocsize = size * sizeof(u32);
+	if (allocsize > p->allocsize) {
+		u32 newsize = allocsize + (p->allocsize >> 1u);
+		void *data = png_realloc(p->data, newsize);
+		if (data) {
+			p->allocsize = newsize;
+			p->data = (u32 *)data;
+		} else
+			return 0; /*error: not enough memory*/
+	}
+	p->size = size;
+	return 1; /*success*/
+}
+
+static void uivector_init(uivector *p)
+{
+	p->data = NULL;
+	p->size = p->allocsize = 0;
+}
+
+/*returns 1 if success, 0 if failure ==> nothing done*/
+static u32 uivector_push_back(uivector *p, u32 c)
+{
+	if (!uivector_resize(p, p->size + 1))
+		return 0;
+	p->data[p->size - 1] = c;
+	return 1;
+}
+#endif /*PNG_COMPILE_ENCODER*/
+#endif /*PNG_COMPILE_ZLIB*/
+
+/* /////////////////////////////////////////////////////////////////////////// */
+
+/*dynamic vector of u8s*/
+typedef struct ucvector {
+	u8 *data;
+	u32 size; /*used size*/
+	u32 allocsize; /*allocated size*/
+} ucvector;
+
+/*returns 1 if success, 0 if failure ==> nothing done*/
+static u32 ucvector_resize(ucvector *p, u32 size)
+{
+	if (size > p->allocsize) {
+		u32 newsize = size + (p->allocsize >> 1u);
+		void *data = png_realloc(p->data, newsize);
+		if (data) {
+			p->allocsize = newsize;
+			p->data = (u8 *)data;
+		} else
+			return 0; /*error: not enough memory*/
+	}
+	p->size = size;
+	return 1; /*success*/
+}
+
+static ucvector ucvector_init(u8 *buffer, u32 size)
+{
+	ucvector v;
+	v.data = buffer;
+	v.allocsize = v.size = size;
+	return v;
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef PNG_COMPILE_PNG
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+
+/*free string pointer and set it to NULL*/
+static void string_cleanup(char **out)
+{
+	png_free(*out);
+	*out = NULL;
+}
+
+/*also appends null termination character*/
+static char *alloc_string_sized(const char *in, u32 insize)
+{
+	char *out = (char *)png_malloc(insize + 1);
+	if (out) {
+		png_memcpy(out, in, insize);
+		out[insize] = 0;
+	}
+	return out;
+}
+
+/* dynamically allocates a new string with a copy of the null terminated input text */
+static char *alloc_string(const char *in)
+{
+	return alloc_string_sized(in, png_strlen(in));
+}
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+#endif /*PNG_COMPILE_PNG*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#if defined(PNG_COMPILE_DECODER) || defined(PNG_COMPILE_PNG)
+static u32 png_read32bitInt(const u8 *buffer)
+{
+	return (((u32)buffer[0] << 24u) | ((u32)buffer[1] << 16u) | ((u32)buffer[2] << 8u) |
+		(u32)buffer[3]);
+}
+#endif /*defined(PNG_COMPILE_DECODER) || defined(PNG_COMPILE_PNG)*/
+
+#if defined(PNG_COMPILE_PNG) || defined(PNG_COMPILE_ENCODER)
+/*buffer must have at least 4 allocated bytes available*/
+static void png_set32bitInt(u8 *buffer, u32 value)
+{
+	buffer[0] = (u8)((value >> 24) & 0xff);
+	buffer[1] = (u8)((value >> 16) & 0xff);
+	buffer[2] = (u8)((value >> 8) & 0xff);
+	buffer[3] = (u8)((value)&0xff);
+}
+#endif /*defined(PNG_COMPILE_PNG) || defined(PNG_COMPILE_ENCODER)*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / File IO                                                                / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef PNG_COMPILE_DISK
+
+/* returns negative value on error. This should be pure C compatible, so no fstat. */
+static long png_filesize(const char *filename)
+{
+	struct stat s = { 0 };
+	stat(filename, &s);
+	return s.size;
+}
+
+/* load file into buffer that already has the correct allocated size. Returns error code.*/
+static u32 png_buffer_file(u8 *out, u32 size, const char *filename)
+{
+	u32 readsize;
+	readsize = read(filename, out, 0, size);
+
+	if (readsize != size)
+		return 78;
+	return 0;
+}
+
+u32 png_load_file(u8 **out, u32 *outsize, const char *filename)
+{
+	long size = png_filesize(filename);
+	if (size < 0)
+		return 78;
+	*outsize = (u32)size;
+
+	*out = (u8 *)png_malloc((u32)size);
+	if (!(*out) && size > 0)
+		return 83; /*the above malloc failed*/
+
+	return png_buffer_file(*out, (u32)size, filename);
+}
+
+/*write given buffer to the file, overwriting the file, it doesn't append to it.*/
+u32 png_save_file(const u8 *buffer, u32 buffersize, const char *filename)
+{
+	err(1, "Not implemented!\n");
+	if (write(filename, buffer, 0, buffersize) <= 0)
+		return 79;
+	return 0;
+}
+
+#endif /*PNG_COMPILE_DISK*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* // End of common code and tools. Begin of Zlib related code.            // */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef PNG_COMPILE_ZLIB
+#ifdef PNG_COMPILE_ENCODER
+
+typedef struct {
+	ucvector *data;
+	u8 bp; /*ok to overflow, indicates bit pos inside byte*/
+} pngBitWriter;
+
+static void pngBitWriter_init(pngBitWriter *writer, ucvector *data)
+{
+	writer->data = data;
+	writer->bp = 0;
+}
+
+/*TODO: this ignores potential out of memory errors*/
+#define WRITEBIT(writer, bit)                                                                      \
+	{                                                                                          \
+		/* append new byte */                                                              \
+		if (((writer->bp) & 7u) == 0) {                                                    \
+			if (!ucvector_resize(writer->data, writer->data->size + 1))                \
+				return;                                                            \
+			writer->data->data[writer->data->size - 1] = 0;                            \
+		}                                                                                  \
+		(writer->data->data[writer->data->size - 1]) |= (bit << ((writer->bp) & 7u));      \
+		++writer->bp;                                                                      \
+	}
+
+/* LSB of value is written first, and LSB of bytes is used first */
+static void writeBits(pngBitWriter *writer, u32 value, u32 nbits)
+{
+	if (nbits == 1) { /* compiler should statically compile this case if nbits == 1 */
+		WRITEBIT(writer, value);
+	} else {
+		/* TODO: increase output size only once here rather than in each WRITEBIT */
+		u32 i;
+		for (i = 0; i != nbits; ++i) {
+			WRITEBIT(writer, (u8)((value >> i) & 1));
+		}
+	}
+}
+
+/* This one is to use for adding huffman symbol, the value bits are written MSB first */
+static void writeBitsReversed(pngBitWriter *writer, u32 value, u32 nbits)
+{
+	u32 i;
+	for (i = 0; i != nbits; ++i) {
+		/* TODO: increase output size only once here rather than in each WRITEBIT */
+		WRITEBIT(writer, (u8)((value >> (nbits - 1u - i)) & 1u));
+	}
+}
+#endif /*PNG_COMPILE_ENCODER*/
+
+#ifdef PNG_COMPILE_DECODER
+
+typedef struct {
+	const u8 *data;
+	u32 size; /*size of data in bytes*/
+	u32 bitsize; /*size of data in bits, end of valid bp values, should be 8*size*/
+	u32 bp;
+	u32 buffer; /*buffer for reading bits. NOTE: 'u32' must support at least 32 bits*/
+} pngBitReader;
+
+/* data size argument is in bytes. Returns error if size too large causing overflow */
+static u32 pngBitReader_init(pngBitReader *reader, const u8 *data, u32 size)
+{
+	u32 temp;
+	reader->data = data;
+	reader->size = size;
+	/* size in bits, return error if overflow (if u32 is 32 bit this supports up to 500MB)  */
+	if (png_mulofl(size, 8u, &reader->bitsize))
+		return 105;
+	/*ensure incremented bp can be compared to bitsize without overflow even when it would be incremented 32 too much and
+  trying to ensure 32 more bits*/
+	if (png_addofl(reader->bitsize, 64u, &temp))
+		return 105;
+	reader->bp = 0;
+	reader->buffer = 0;
+	return 0; /*ok*/
+}
+
+/*
+ensureBits functions:
+Ensures the reader can at least read nbits bits in one or more readBits calls,
+safely even if not enough bits are available.
+Returns 1 if there are enough bits available, 0 if not.
+*/
+
+/*See ensureBits documentation above. This one ensures exactly 1 bit */
+/*static u32 ensureBits1(pngBitReader* reader) {
+  if(reader->bp >= reader->bitsize) return 0;
+  reader->buffer = (u32)reader->data[reader->bp >> 3u] >> (reader->bp & 7u);
+  return 1;
+}*/
+
+/*See ensureBits documentation above. This one ensures up to 9 bits */
+static u32 ensureBits9(pngBitReader *reader, u32 nbits)
+{
+	u32 start = reader->bp >> 3u;
+	u32 size = reader->size;
+	if (start + 1u < size) {
+		reader->buffer =
+			(u32)reader->data[start + 0] | ((u32)reader->data[start + 1] << 8u);
+		reader->buffer >>= (reader->bp & 7u);
+		return 1;
+	} else {
+		reader->buffer = 0;
+		if (start + 0u < size)
+			reader->buffer |= reader->data[start + 0];
+		reader->buffer >>= (reader->bp & 7u);
+		return reader->bp + nbits <= reader->bitsize;
+	}
+}
+
+/*See ensureBits documentation above. This one ensures up to 17 bits */
+static u32 ensureBits17(pngBitReader *reader, u32 nbits)
+{
+	u32 start = reader->bp >> 3u;
+	u32 size = reader->size;
+	if (start + 2u < size) {
+		reader->buffer = (u32)reader->data[start + 0] |
+				 ((u32)reader->data[start + 1] << 8u) |
+				 ((u32)reader->data[start + 2] << 16u);
+		reader->buffer >>= (reader->bp & 7u);
+		return 1;
+	} else {
+		reader->buffer = 0;
+		if (start + 0u < size)
+			reader->buffer |= reader->data[start + 0];
+		if (start + 1u < size)
+			reader->buffer |= ((u32)reader->data[start + 1] << 8u);
+		reader->buffer >>= (reader->bp & 7u);
+		return reader->bp + nbits <= reader->bitsize;
+	}
+}
+
+/*See ensureBits documentation above. This one ensures up to 25 bits */
+static PNG_INLINE u32 ensureBits25(pngBitReader *reader, u32 nbits)
+{
+	u32 start = reader->bp >> 3u;
+	u32 size = reader->size;
+	if (start + 3u < size) {
+		reader->buffer = (u32)reader->data[start + 0] |
+				 ((u32)reader->data[start + 1] << 8u) |
+				 ((u32)reader->data[start + 2] << 16u) |
+				 ((u32)reader->data[start + 3] << 24u);
+		reader->buffer >>= (reader->bp & 7u);
+		return 1;
+	} else {
+		reader->buffer = 0;
+		if (start + 0u < size)
+			reader->buffer |= reader->data[start + 0];
+		if (start + 1u < size)
+			reader->buffer |= ((u32)reader->data[start + 1] << 8u);
+		if (start + 2u < size)
+			reader->buffer |= ((u32)reader->data[start + 2] << 16u);
+		reader->buffer >>= (reader->bp & 7u);
+		return reader->bp + nbits <= reader->bitsize;
+	}
+}
+
+/*See ensureBits documentation above. This one ensures up to 32 bits */
+static PNG_INLINE u32 ensureBits32(pngBitReader *reader, u32 nbits)
+{
+	u32 start = reader->bp >> 3u;
+	u32 size = reader->size;
+	if (start + 4u < size) {
+		reader->buffer = (u32)reader->data[start + 0] |
+				 ((u32)reader->data[start + 1] << 8u) |
+				 ((u32)reader->data[start + 2] << 16u) |
+				 ((u32)reader->data[start + 3] << 24u);
+		reader->buffer >>= (reader->bp & 7u);
+		reader->buffer |=
+			(((u32)reader->data[start + 4] << 24u) << (8u - (reader->bp & 7u)));
+		return 1;
+	} else {
+		reader->buffer = 0;
+		if (start + 0u < size)
+			reader->buffer |= reader->data[start + 0];
+		if (start + 1u < size)
+			reader->buffer |= ((u32)reader->data[start + 1] << 8u);
+		if (start + 2u < size)
+			reader->buffer |= ((u32)reader->data[start + 2] << 16u);
+		if (start + 3u < size)
+			reader->buffer |= ((u32)reader->data[start + 3] << 24u);
+		reader->buffer >>= (reader->bp & 7u);
+		return reader->bp + nbits <= reader->bitsize;
+	}
+}
+
+/* Get bits without advancing the bit pointer. Must have enough bits available with ensureBits. Max nbits is 31. */
+static u32 peekBits(pngBitReader *reader, u32 nbits)
+{
+	/* The shift allows nbits to be only up to 31. */
+	return reader->buffer & ((1u << nbits) - 1u);
+}
+
+/* Must have enough bits available with ensureBits */
+static void advanceBits(pngBitReader *reader, u32 nbits)
+{
+	reader->buffer >>= nbits;
+	reader->bp += nbits;
+}
+
+/* Must have enough bits available with ensureBits */
+static u32 readBits(pngBitReader *reader, u32 nbits)
+{
+	u32 result = peekBits(reader, nbits);
+	advanceBits(reader, nbits);
+	return result;
+}
+
+/* Public for testing only. steps and result must have numsteps values. */
+/*static u32 png_test_bitreader(const u8 *data, u32 size, u32 numsteps, const u32 *steps, u32 *result)
+{
+	u32 i;
+	pngBitReader reader;
+	u32 error = pngBitReader_init(&reader, data, size);
+	if (error)
+		return 0;
+	for (i = 0; i < numsteps; i++) {
+		u32 step = steps[i];
+		u32 ok;
+		if (step > 25)
+			ok = ensureBits32(&reader, step);
+		else if (step > 17)
+			ok = ensureBits25(&reader, step);
+		else if (step > 9)
+			ok = ensureBits17(&reader, step);
+		else
+			ok = ensureBits9(&reader, step);
+		if (!ok)
+			return 0;
+		result[i] = readBits(&reader, step);
+	}
+	return 1;
+}*/
+#endif /*PNG_COMPILE_DECODER*/
+
+static u32 reverseBits(u32 bits, u32 num)
+{
+	/*TODO: implement faster lookup table based version when needed*/
+	u32 i, result = 0;
+	for (i = 0; i < num; i++)
+		result |= ((bits >> (num - i - 1u)) & 1u) << i;
+	return result;
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Deflate - Huffman                                                      / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#define FIRST_LENGTH_CODE_INDEX 257
+#define LAST_LENGTH_CODE_INDEX 285
+/*256 literals, the end code, some length codes, and 2 unused codes*/
+#define NUM_DEFLATE_CODE_SYMBOLS 288
+/*the distance codes have their own symbols, 30 used, 2 unused*/
+#define NUM_DISTANCE_SYMBOLS 32
+/*the code length codes. 0-15: code lengths, 16: copy previous 3-6 times, 17: 3-10 zeros, 18: 11-138 zeros*/
+#define NUM_CODE_LENGTH_CODES 19
+
+/*the base lengths represented by codes 257-285*/
+static const u32 LENGTHBASE[29] = { 3,	4,  5,	6,  7,	8,  9,	10, 11,	 13,  15,  17,	19,  23, 27,
+				    31, 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258 };
+
+/*the extra bits used by codes 257-285 (added to base length)*/
+static const u32 LENGTHEXTRA[29] = { 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2,
+				     2, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 0 };
+
+/*the base backwards distances (the bits of distance codes appear after length codes and use their own huffman tree)*/
+static const u32 DISTANCEBASE[30] = {
+	1,   2,	  3,   4,   5,	 7,    9,    13,   17,	 25,   33,   49,   65,	  97,	 129,
+	193, 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, 8193, 12289, 16385, 24577
+};
+
+/*the extra bits of backwards distances (added to base)*/
+static const u32 DISTANCEEXTRA[30] = { 0, 0, 0, 0, 1, 1, 2, 2,	3,  3,	4,  4,	5,  5,	6,
+				       6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13 };
+
+/*the order in which "code length alphabet code lengths" are stored as specified by deflate, out of this the huffman
+tree of the dynamic huffman tree lengths is generated*/
+static const u32 CLCL_ORDER[NUM_CODE_LENGTH_CODES] = { 16, 17, 18, 0, 8,  7, 9,	 6, 10, 5,
+						       11, 4,  12, 3, 13, 2, 14, 1, 15 };
+
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*
+Huffman tree struct, containing multiple representations of the tree
+*/
+typedef struct HuffmanTree {
+	u32 *codes; /*the huffman codes (bit patterns representing the symbols)*/
+	u32 *lengths; /*the lengths of the huffman codes*/
+	u32 maxbitlen; /*maximum number of bits a single code can get*/
+	u32 numcodes; /*number of symbols in the alphabet = number of codes*/
+	/* for reading only */
+	u8 *table_len; /*length of symbol from lookup table, or max length if secondary lookup needed*/
+	u16 *table_value; /*value of symbol from lookup table, or pointer to secondary table if needed*/
+} HuffmanTree;
+
+static void HuffmanTree_init(HuffmanTree *tree)
+{
+	tree->codes = 0;
+	tree->lengths = 0;
+	tree->table_len = 0;
+	tree->table_value = 0;
+}
+
+static void HuffmanTree_cleanup(HuffmanTree *tree)
+{
+	png_free(tree->codes);
+	png_free(tree->lengths);
+	png_free(tree->table_len);
+	png_free(tree->table_value);
+}
+
+/* amount of bits for first huffman table lookup (aka root bits), see HuffmanTree_makeTable and huffmanDecodeSymbol.*/
+/* values 8u and 9u work the fastest */
+#define FIRSTBITS 9u
+
+/* a symbol value too big to represent any valid symbol, to indicate reading disallowed huffman bits combination,
+which is possible in case of only 0 or 1 present symbols. */
+#define INVALIDSYMBOL 65535u
+
+/* make table for huffman decoding */
+static u32 HuffmanTree_makeTable(HuffmanTree *tree)
+{
+	static const u32 headsize = 1u << FIRSTBITS; /*size of the first table*/
+	static const u32 mask = (1u << FIRSTBITS) /*headsize*/ - 1u;
+	u32 i, numpresent, pointer, size; /*total table size*/
+	u32 *maxlens = (u32 *)png_malloc(headsize * sizeof(u32));
+	if (!maxlens)
+		return 83; /*alloc fail*/
+
+	/* compute maxlens: max total bit length of symbols sharing prefix in the first table*/
+	png_memset(maxlens, 0, headsize * sizeof(*maxlens));
+	for (i = 0; i < tree->numcodes; i++) {
+		u32 symbol = tree->codes[i];
+		u32 l = tree->lengths[i];
+		u32 index;
+		if (l <= FIRSTBITS)
+			continue; /*symbols that fit in first table don't increase secondary table size*/
+		/*get the FIRSTBITS MSBs, the MSBs of the symbol are encoded first. See later comment about the reversing*/
+		index = reverseBits(symbol >> (l - FIRSTBITS), FIRSTBITS);
+		maxlens[index] = PNG_MAX(maxlens[index], l);
+	}
+	/* compute total table size: size of first table plus all secondary tables for symbols longer than FIRSTBITS */
+	size = headsize;
+	for (i = 0; i < headsize; ++i) {
+		u32 l = maxlens[i];
+		if (l > FIRSTBITS)
+			size += (1u << (l - FIRSTBITS));
+	}
+	tree->table_len = (u8 *)png_malloc(size * sizeof(*tree->table_len));
+	tree->table_value = (u16 *)png_malloc(size * sizeof(*tree->table_value));
+	if (!tree->table_len || !tree->table_value) {
+		png_free(maxlens);
+		/* freeing tree->table values is done at a higher scope */
+		return 83; /*alloc fail*/
+	}
+	/*initialize with an invalid length to indicate unused entries*/
+	for (i = 0; i < size; ++i)
+		tree->table_len[i] = 16;
+
+	/*fill in the first table for long symbols: max prefix size and pointer to secondary tables*/
+	pointer = headsize;
+	for (i = 0; i < headsize; ++i) {
+		u32 l = maxlens[i];
+		if (l <= FIRSTBITS)
+			continue;
+		tree->table_len[i] = l;
+		tree->table_value[i] = pointer;
+		pointer += (1u << (l - FIRSTBITS));
+	}
+	png_free(maxlens);
+
+	/*fill in the first table for short symbols, or secondary table for long symbols*/
+	numpresent = 0;
+	for (i = 0; i < tree->numcodes; ++i) {
+		u32 l = tree->lengths[i];
+		u32 symbol = tree->codes[i]; /*the huffman bit pattern. i itself is the value.*/
+		/*reverse bits, because the huffman bits are given in MSB first order but the bit reader reads LSB first*/
+		u32 reverse = reverseBits(symbol, l);
+		if (l == 0)
+			continue;
+		numpresent++;
+
+		if (l <= FIRSTBITS) {
+			/*short symbol, fully in first table, replicated num times if l < FIRSTBITS*/
+			u32 num = 1u << (FIRSTBITS - l);
+			u32 j;
+			for (j = 0; j < num; ++j) {
+				/*bit reader will read the l bits of symbol first, the remaining FIRSTBITS - l bits go to the MSB's*/
+				u32 index = reverse | (j << l);
+				if (tree->table_len[index] != 16)
+					return 55; /*invalid tree: long symbol shares prefix with short symbol*/
+				tree->table_len[index] = l;
+				tree->table_value[index] = i;
+			}
+		} else {
+			/*long symbol, shares prefix with other long symbols in first lookup table, needs second lookup*/
+			/*the FIRSTBITS MSBs of the symbol are the first table index*/
+			u32 index = reverse & mask;
+			u32 maxlen = tree->table_len[index];
+			/*log2 of secondary table length, should be >= l - FIRSTBITS*/
+			u32 tablelen = maxlen - FIRSTBITS;
+			u32 start = tree->table_value[index]; /*starting index in secondary table*/
+			u32 num =
+				1u
+				<< (tablelen -
+				    (l -
+				     FIRSTBITS)); /*amount of entries of this symbol in secondary table*/
+			u32 j;
+			if (maxlen < l)
+				return 55; /*invalid tree: long symbol shares prefix with short symbol*/
+			for (j = 0; j < num; ++j) {
+				u32 reverse2 = reverse >> FIRSTBITS; /* l - FIRSTBITS bits */
+				u32 index2 = start + (reverse2 | (j << (l - FIRSTBITS)));
+				tree->table_len[index2] = l;
+				tree->table_value[index2] = i;
+			}
+		}
+	}
+
+	if (numpresent < 2) {
+		/* In case of exactly 1 symbol, in theory the huffman symbol needs 0 bits,
+    but deflate uses 1 bit instead. In case of 0 symbols, no symbols can
+    appear at all, but such huffman tree could still exist (e.g. if distance
+    codes are never used). In both cases, not all symbols of the table will be
+    filled in. Fill them in with an invalid symbol value so returning them from
+    huffmanDecodeSymbol will cause error. */
+		for (i = 0; i < size; ++i) {
+			if (tree->table_len[i] == 16) {
+				/* As length, use a value smaller than FIRSTBITS for the head table,
+        and a value larger than FIRSTBITS for the secondary table, to ensure
+        valid behavior for advanceBits when reading this symbol. */
+				tree->table_len[i] = (i < headsize) ? 1 : (FIRSTBITS + 1);
+				tree->table_value[i] = INVALIDSYMBOL;
+			}
+		}
+	} else {
+		/* A good huffman tree has N * 2 - 1 nodes, of which N - 1 are internal nodes.
+    If that is not the case (due to too long length codes), the table will not
+    have been fully used, and this is an error (not all bit combinations can be
+    decoded): an oversubscribed huffman tree, indicated by error 55. */
+		for (i = 0; i < size; ++i) {
+			if (tree->table_len[i] == 16)
+				return 55;
+		}
+	}
+
+	return 0;
+}
+
+/*
+Second step for the ...makeFromLengths and ...makeFromFrequencies functions.
+numcodes, lengths and maxbitlen must already be filled in correctly. return
+value is error.
+*/
+static u32 HuffmanTree_makeFromLengths2(HuffmanTree *tree)
+{
+	u32 *blcount;
+	u32 *nextcode;
+	u32 error = 0;
+	u32 bits, n;
+
+	tree->codes = (u32 *)png_malloc(tree->numcodes * sizeof(u32));
+	blcount = (u32 *)png_malloc((tree->maxbitlen + 1) * sizeof(u32));
+	nextcode = (u32 *)png_malloc((tree->maxbitlen + 1) * sizeof(u32));
+	if (!tree->codes || !blcount || !nextcode)
+		error = 83; /*alloc fail*/
+
+	if (!error) {
+		for (n = 0; n != tree->maxbitlen + 1; n++)
+			blcount[n] = nextcode[n] = 0;
+		/*step 1: count number of instances of each code length*/
+		for (bits = 0; bits != tree->numcodes; ++bits)
+			++blcount[tree->lengths[bits]];
+		/*step 2: generate the nextcode values*/
+		for (bits = 1; bits <= tree->maxbitlen; ++bits) {
+			nextcode[bits] = (nextcode[bits - 1] + blcount[bits - 1]) << 1u;
+		}
+		/*step 3: generate all the codes*/
+		for (n = 0; n != tree->numcodes; ++n) {
+			if (tree->lengths[n] != 0) {
+				tree->codes[n] = nextcode[tree->lengths[n]]++;
+				/*remove superfluous bits from the code*/
+				tree->codes[n] &= ((1u << tree->lengths[n]) - 1u);
+			}
+		}
+	}
+
+	png_free(blcount);
+	png_free(nextcode);
+
+	if (!error)
+		error = HuffmanTree_makeTable(tree);
+	return error;
+}
+
+/*
+given the code lengths (as stored in the PNG file), generate the tree as defined
+by Deflate. maxbitlen is the maximum bits that a code in the tree can have.
+return value is error.
+*/
+static u32 HuffmanTree_makeFromLengths(HuffmanTree *tree, const u32 *bitlen, u32 numcodes,
+				       u32 maxbitlen)
+{
+	u32 i;
+	tree->lengths = (u32 *)png_malloc(numcodes * sizeof(u32));
+	if (!tree->lengths)
+		return 83; /*alloc fail*/
+	for (i = 0; i != numcodes; ++i)
+		tree->lengths[i] = bitlen[i];
+	tree->numcodes = (u32)numcodes; /*number of symbols*/
+	tree->maxbitlen = maxbitlen;
+	return HuffmanTree_makeFromLengths2(tree);
+}
+
+#ifdef PNG_COMPILE_ENCODER
+
+/*BPM: Boundary Package Merge, see "A Fast and Space-Economical Algorithm for Length-Limited Coding",
+Jyrki Katajainen, Alistair Moffat, Andrew Turpin, 1995.*/
+
+/*chain node for boundary package merge*/
+typedef struct BPMNode {
+	int weight; /*the sum of all weights in this chain*/
+	u32 index; /*index of this leaf node (called "count" in the paper)*/
+	struct BPMNode *tail; /*the next nodes in this chain (null if last)*/
+	int in_use;
+} BPMNode;
+
+/*lists of chains*/
+typedef struct BPMLists {
+	/*memory pool*/
+	u32 memsize;
+	BPMNode *memory;
+	u32 numfree;
+	u32 nextfree;
+	BPMNode **freelist;
+	/*two heads of lookahead chains per list*/
+	u32 listsize;
+	BPMNode **chains0;
+	BPMNode **chains1;
+} BPMLists;
+
+/*creates a new chain node with the given parameters, from the memory in the lists */
+static BPMNode *bpmnode_create(BPMLists *lists, int weight, u32 index, BPMNode *tail)
+{
+	u32 i;
+	BPMNode *result;
+
+	/*memory full, so garbage collect*/
+	if (lists->nextfree >= lists->numfree) {
+		/*mark only those that are in use*/
+		for (i = 0; i != lists->memsize; ++i)
+			lists->memory[i].in_use = 0;
+		for (i = 0; i != lists->listsize; ++i) {
+			BPMNode *node;
+			for (node = lists->chains0[i]; node != 0; node = node->tail)
+				node->in_use = 1;
+			for (node = lists->chains1[i]; node != 0; node = node->tail)
+				node->in_use = 1;
+		}
+		/*collect those that are free*/
+		lists->numfree = 0;
+		for (i = 0; i != lists->memsize; ++i) {
+			if (!lists->memory[i].in_use)
+				lists->freelist[lists->numfree++] = &lists->memory[i];
+		}
+		lists->nextfree = 0;
+	}
+
+	result = lists->freelist[lists->nextfree++];
+	result->weight = weight;
+	result->index = index;
+	result->tail = tail;
+	return result;
+}
+
+/*sort the leaves with stable mergesort*/
+static void bpmnode_sort(BPMNode *leaves, u32 num)
+{
+	BPMNode *mem = (BPMNode *)png_malloc(sizeof(*leaves) * num);
+	u32 width, counter = 0;
+	for (width = 1; width < num; width *= 2) {
+		BPMNode *a = (counter & 1) ? mem : leaves;
+		BPMNode *b = (counter & 1) ? leaves : mem;
+		u32 p;
+		for (p = 0; p < num; p += 2 * width) {
+			u32 q = (p + width > num) ? num : (p + width);
+			u32 r = (p + 2 * width > num) ? num : (p + 2 * width);
+			u32 i = p, j = q, k;
+			for (k = p; k < r; k++) {
+				if (i < q && (j >= r || a[i].weight <= a[j].weight))
+					b[k] = a[i++];
+				else
+					b[k] = a[j++];
+			}
+		}
+		counter++;
+	}
+	if (counter & 1)
+		png_memcpy(leaves, mem, sizeof(*leaves) * num);
+	png_free(mem);
+}
+
+/*Boundary Package Merge step, numpresent is the amount of leaves, and c is the current chain.*/
+static void boundaryPM(BPMLists *lists, BPMNode *leaves, u32 numpresent, int c, int num)
+{
+	u32 lastindex = lists->chains1[c]->index;
+
+	if (c == 0) {
+		if (lastindex >= numpresent)
+			return;
+		lists->chains0[c] = lists->chains1[c];
+		lists->chains1[c] =
+			bpmnode_create(lists, leaves[lastindex].weight, lastindex + 1, 0);
+	} else {
+		/*sum of the weights of the head nodes of the previous lookahead chains.*/
+		int sum = lists->chains0[c - 1]->weight + lists->chains1[c - 1]->weight;
+		lists->chains0[c] = lists->chains1[c];
+		if (lastindex < numpresent && sum > leaves[lastindex].weight) {
+			lists->chains1[c] = bpmnode_create(lists, leaves[lastindex].weight,
+							   lastindex + 1, lists->chains1[c]->tail);
+			return;
+		}
+		lists->chains1[c] = bpmnode_create(lists, sum, lastindex, lists->chains1[c - 1]);
+		/*in the end we are only interested in the chain of the last list, so no
+    need to recurse if we're at the last one (this gives measurable speedup)*/
+		if (num + 1 < (int)(2 * numpresent - 2)) {
+			boundaryPM(lists, leaves, numpresent, c - 1, num);
+			boundaryPM(lists, leaves, numpresent, c - 1, num);
+		}
+	}
+}
+
+u32 png_huffman_code_lengths(u32 *lengths, const u32 *frequencies, u32 numcodes, u32 maxbitlen)
+{
+	u32 error = 0;
+	u32 i;
+	u32 numpresent = 0; /*number of symbols with non-zero frequency*/
+	BPMNode *leaves; /*the symbols, only those with > 0 frequency*/
+
+	if (numcodes == 0)
+		return 80; /*error: a tree of 0 symbols is not supposed to be made*/
+	if ((1u << maxbitlen) < (u32)numcodes)
+		return 80; /*error: represent all symbols*/
+
+	leaves = (BPMNode *)png_malloc(numcodes * sizeof(*leaves));
+	if (!leaves)
+		return 83; /*alloc fail*/
+
+	for (i = 0; i != numcodes; ++i) {
+		if (frequencies[i] > 0) {
+			leaves[numpresent].weight = (int)frequencies[i];
+			leaves[numpresent].index = i;
+			++numpresent;
+		}
+	}
+
+	png_memset(lengths, 0, numcodes * sizeof(*lengths));
+
+	/*ensure at least two present symbols. There should be at least one symbol
+  according to RFC 1951 section 3.2.7. Some decoders incorrectly require two. To
+  make these work as well ensure there are at least two symbols. The
+  Package-Merge code below also doesn't work correctly if there's only one
+  symbol, it'd give it the theoretical 0 bits but in practice zlib wants 1 bit*/
+	if (numpresent == 0) {
+		lengths[0] = lengths[1] =
+			1; /*note that for RFC 1951 section 3.2.7, only lengths[0] = 1 is needed*/
+	} else if (numpresent == 1) {
+		lengths[leaves[0].index] = 1;
+		lengths[leaves[0].index == 0 ? 1 : 0] = 1;
+	} else {
+		BPMLists lists;
+		BPMNode *node;
+
+		bpmnode_sort(leaves, numpresent);
+
+		lists.listsize = maxbitlen;
+		lists.memsize = 2 * maxbitlen * (maxbitlen + 1);
+		lists.nextfree = 0;
+		lists.numfree = lists.memsize;
+		lists.memory = (BPMNode *)png_malloc(lists.memsize * sizeof(*lists.memory));
+		lists.freelist = (BPMNode **)png_malloc(lists.memsize * sizeof(BPMNode *));
+		lists.chains0 = (BPMNode **)png_malloc(lists.listsize * sizeof(BPMNode *));
+		lists.chains1 = (BPMNode **)png_malloc(lists.listsize * sizeof(BPMNode *));
+		if (!lists.memory || !lists.freelist || !lists.chains0 || !lists.chains1)
+			error = 83; /*alloc fail*/
+
+		if (!error) {
+			for (i = 0; i != lists.memsize; ++i)
+				lists.freelist[i] = &lists.memory[i];
+
+			bpmnode_create(&lists, leaves[0].weight, 1, 0);
+			bpmnode_create(&lists, leaves[1].weight, 2, 0);
+
+			for (i = 0; i != lists.listsize; ++i) {
+				lists.chains0[i] = &lists.memory[0];
+				lists.chains1[i] = &lists.memory[1];
+			}
+
+			/*each boundaryPM call adds one chain to the last list, and we need 2 * numpresent - 2 chains.*/
+			for (i = 2; i != 2 * numpresent - 2; ++i)
+				boundaryPM(&lists, leaves, numpresent, (int)maxbitlen - 1, (int)i);
+
+			for (node = lists.chains1[maxbitlen - 1]; node; node = node->tail) {
+				for (i = 0; i != node->index; ++i)
+					++lengths[leaves[i].index];
+			}
+		}
+
+		png_free(lists.memory);
+		png_free(lists.freelist);
+		png_free(lists.chains0);
+		png_free(lists.chains1);
+	}
+
+	png_free(leaves);
+	return error;
+}
+
+/*Create the Huffman tree given the symbol frequencies*/
+static u32 HuffmanTree_makeFromFrequencies(HuffmanTree *tree, const u32 *frequencies, u32 mincodes,
+					   u32 numcodes, u32 maxbitlen)
+{
+	u32 error = 0;
+	while (!frequencies[numcodes - 1] && numcodes > mincodes)
+		--numcodes; /*trim zeroes*/
+	tree->lengths = (u32 *)png_malloc(numcodes * sizeof(u32));
+	if (!tree->lengths)
+		return 83; /*alloc fail*/
+	tree->maxbitlen = maxbitlen;
+	tree->numcodes = (u32)numcodes; /*number of symbols*/
+
+	error = png_huffman_code_lengths(tree->lengths, frequencies, numcodes, maxbitlen);
+	if (!error)
+		error = HuffmanTree_makeFromLengths2(tree);
+	return error;
+}
+#endif /*PNG_COMPILE_ENCODER*/
+
+/*get the literal and length code tree of a deflated block with fixed tree, as per the deflate specification*/
+static u32 generateFixedLitLenTree(HuffmanTree *tree)
+{
+	u32 i, error = 0;
+	u32 *bitlen = (u32 *)png_malloc(NUM_DEFLATE_CODE_SYMBOLS * sizeof(u32));
+	if (!bitlen)
+		return 83; /*alloc fail*/
+
+	/*288 possible codes: 0-255=literals, 256=endcode, 257-285=lengthcodes, 286-287=unused*/
+	for (i = 0; i <= 143; ++i)
+		bitlen[i] = 8;
+	for (i = 144; i <= 255; ++i)
+		bitlen[i] = 9;
+	for (i = 256; i <= 279; ++i)
+		bitlen[i] = 7;
+	for (i = 280; i <= 287; ++i)
+		bitlen[i] = 8;
+
+	error = HuffmanTree_makeFromLengths(tree, bitlen, NUM_DEFLATE_CODE_SYMBOLS, 15);
+
+	png_free(bitlen);
+	return error;
+}
+
+/*get the distance code tree of a deflated block with fixed tree, as specified in the deflate specification*/
+static u32 generateFixedDistanceTree(HuffmanTree *tree)
+{
+	u32 i, error = 0;
+	u32 *bitlen = (u32 *)png_malloc(NUM_DISTANCE_SYMBOLS * sizeof(u32));
+	if (!bitlen)
+		return 83; /*alloc fail*/
+
+	/*there are 32 distance codes, but 30-31 are unused*/
+	for (i = 0; i != NUM_DISTANCE_SYMBOLS; ++i)
+		bitlen[i] = 5;
+	error = HuffmanTree_makeFromLengths(tree, bitlen, NUM_DISTANCE_SYMBOLS, 15);
+
+	png_free(bitlen);
+	return error;
+}
+
+#ifdef PNG_COMPILE_DECODER
+
+/*
+returns the code. The bit reader must already have been ensured at least 15 bits
+*/
+static u32 huffmanDecodeSymbol(pngBitReader *reader, const HuffmanTree *codetree)
+{
+	u16 code = peekBits(reader, FIRSTBITS);
+	u16 l = codetree->table_len[code];
+	u16 value = codetree->table_value[code];
+	if (l <= FIRSTBITS) {
+		advanceBits(reader, l);
+		return value;
+	} else {
+		u32 index2;
+		advanceBits(reader, FIRSTBITS);
+		index2 = value + peekBits(reader, l - FIRSTBITS);
+		advanceBits(reader, codetree->table_len[index2] - FIRSTBITS);
+		return codetree->table_value[index2];
+	}
+}
+#endif /*PNG_COMPILE_DECODER*/
+
+#ifdef PNG_COMPILE_DECODER
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Inflator (Decompressor)                                                / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*get the tree of a deflated block with fixed tree, as specified in the deflate specification
+Returns error code.*/
+static u32 getTreeInflateFixed(HuffmanTree *tree_ll, HuffmanTree *tree_d)
+{
+	u32 error = generateFixedLitLenTree(tree_ll);
+	if (error)
+		return error;
+	return generateFixedDistanceTree(tree_d);
+}
+
+/*get the tree of a deflated block with dynamic tree, the tree itself is also Huffman compressed with a known tree*/
+static u32 getTreeInflateDynamic(HuffmanTree *tree_ll, HuffmanTree *tree_d, pngBitReader *reader)
+{
+	/*make sure that length values that aren't filled in will be 0, or a wrong tree will be generated*/
+	u32 error = 0;
+	u32 n, HLIT, HDIST, HCLEN, i;
+
+	/*see comments in deflateDynamic for explanation of the context and these variables, it is analogous*/
+	u32 *bitlen_ll = 0; /*lit,len code lengths*/
+	u32 *bitlen_d = 0; /*dist code lengths*/
+	/*code length code lengths ("clcl"), the bit lengths of the huffman tree used to compress bitlen_ll and bitlen_d*/
+	u32 *bitlen_cl = 0;
+	HuffmanTree
+		tree_cl; /*the code tree for code length codes (the huffman tree for compressed huffman trees)*/
+
+	if (!ensureBits17(reader, 14))
+		return 49; /*error: the bit pointer is or will go past the memory*/
+
+	/*number of literal/length codes + 257. Unlike the spec, the value 257 is added to it here already*/
+	HLIT = readBits(reader, 5) + 257;
+	/*number of distance codes. Unlike the spec, the value 1 is added to it here already*/
+	HDIST = readBits(reader, 5) + 1;
+	/*number of code length codes. Unlike the spec, the value 4 is added to it here already*/
+	HCLEN = readBits(reader, 4) + 4;
+
+	bitlen_cl = (u32 *)png_malloc(NUM_CODE_LENGTH_CODES * sizeof(u32));
+	if (!bitlen_cl)
+		return 83 /*alloc fail*/;
+
+	HuffmanTree_init(&tree_cl);
+
+	while (!error) {
+		/*read the code length codes out of 3 * (amount of code length codes) bits*/
+		if (png_gtofl(reader->bp, HCLEN * 3, reader->bitsize)) {
+			ERROR_BREAK(50); /*error: the bit pointer is or will go past the memory*/
+		}
+		for (i = 0; i != HCLEN; ++i) {
+			ensureBits9(reader, 3); /*out of bounds already checked above */
+			bitlen_cl[CLCL_ORDER[i]] = readBits(reader, 3);
+		}
+		for (i = HCLEN; i != NUM_CODE_LENGTH_CODES; ++i) {
+			bitlen_cl[CLCL_ORDER[i]] = 0;
+		}
+
+		error = HuffmanTree_makeFromLengths(&tree_cl, bitlen_cl, NUM_CODE_LENGTH_CODES, 7);
+		if (error)
+			break;
+
+		/*now we can use this tree to read the lengths for the tree that this function will return*/
+		bitlen_ll = (u32 *)png_malloc(NUM_DEFLATE_CODE_SYMBOLS * sizeof(u32));
+		bitlen_d = (u32 *)png_malloc(NUM_DISTANCE_SYMBOLS * sizeof(u32));
+		if (!bitlen_ll || !bitlen_d)
+			ERROR_BREAK(83 /*alloc fail*/);
+		png_memset(bitlen_ll, 0, NUM_DEFLATE_CODE_SYMBOLS * sizeof(*bitlen_ll));
+		png_memset(bitlen_d, 0, NUM_DISTANCE_SYMBOLS * sizeof(*bitlen_d));
+
+		/*i is the current symbol we're reading in the part that contains the code lengths of lit/len and dist codes*/
+		i = 0;
+		while (i < HLIT + HDIST) {
+			u32 code;
+			ensureBits25(
+				reader,
+				22); /* up to 15 bits for huffman code, up to 7 extra bits below*/
+			code = huffmanDecodeSymbol(reader, &tree_cl);
+			if (code <= 15) /*a length code*/ {
+				if (i < HLIT)
+					bitlen_ll[i] = code;
+				else
+					bitlen_d[i - HLIT] = code;
+				++i;
+			} else if (code == 16) /*repeat previous*/ {
+				u32 replength =
+					3; /*read in the 2 bits that indicate repeat length (3-6)*/
+				u32 value; /*set value to the previous code*/
+
+				if (i == 0)
+					ERROR_BREAK(54); /*can't repeat previous if i is 0*/
+
+				replength += readBits(reader, 2);
+
+				if (i < HLIT + 1)
+					value = bitlen_ll[i - 1];
+				else
+					value = bitlen_d[i - HLIT - 1];
+				/*repeat this value in the next lengths*/
+				for (n = 0; n < replength; ++n) {
+					if (i >= HLIT + HDIST)
+						ERROR_BREAK(
+							13); /*error: i is larger than the amount of codes*/
+					if (i < HLIT)
+						bitlen_ll[i] = value;
+					else
+						bitlen_d[i - HLIT] = value;
+					++i;
+				}
+			} else if (code == 17) /*repeat "0" 3-10 times*/ {
+				u32 replength = 3; /*read in the bits that indicate repeat length*/
+				replength += readBits(reader, 3);
+
+				/*repeat this value in the next lengths*/
+				for (n = 0; n < replength; ++n) {
+					if (i >= HLIT + HDIST)
+						ERROR_BREAK(
+							14); /*error: i is larger than the amount of codes*/
+
+					if (i < HLIT)
+						bitlen_ll[i] = 0;
+					else
+						bitlen_d[i - HLIT] = 0;
+					++i;
+				}
+			} else if (code == 18) /*repeat "0" 11-138 times*/ {
+				u32 replength = 11; /*read in the bits that indicate repeat length*/
+				replength += readBits(reader, 7);
+
+				/*repeat this value in the next lengths*/
+				for (n = 0; n < replength; ++n) {
+					if (i >= HLIT + HDIST)
+						ERROR_BREAK(
+							15); /*error: i is larger than the amount of codes*/
+
+					if (i < HLIT)
+						bitlen_ll[i] = 0;
+					else
+						bitlen_d[i - HLIT] = 0;
+					++i;
+				}
+			} else /*if(code == INVALIDSYMBOL)*/ {
+				ERROR_BREAK(16); /*error: tried to read disallowed huffman symbol*/
+			}
+			/*check if any of the ensureBits above went out of bounds*/
+			if (reader->bp > reader->bitsize) {
+				/*return error code 10 or 11 depending on the situation that happened in huffmanDecodeSymbol
+        (10=no endcode, 11=wrong jump outside of tree)*/
+				/* TODO: revise error codes 10,11,50: the above comment is no longer valid */
+				ERROR_BREAK(50); /*error, bit pointer jumps past memory*/
+			}
+		}
+		if (error)
+			break;
+
+		if (bitlen_ll[256] == 0)
+			ERROR_BREAK(64); /*the length of the end code 256 must be larger than 0*/
+
+		/*now we've finally got HLIT and HDIST, so generate the code trees, and the function is done*/
+		error = HuffmanTree_makeFromLengths(tree_ll, bitlen_ll, NUM_DEFLATE_CODE_SYMBOLS,
+						    15);
+		if (error)
+			break;
+		error = HuffmanTree_makeFromLengths(tree_d, bitlen_d, NUM_DISTANCE_SYMBOLS, 15);
+
+		break; /*end of error-while*/
+	}
+
+	png_free(bitlen_cl);
+	png_free(bitlen_ll);
+	png_free(bitlen_d);
+	HuffmanTree_cleanup(&tree_cl);
+
+	return error;
+}
+
+/*inflate a block with dynamic of fixed Huffman tree. btype must be 1 or 2.*/
+static u32 inflateHuffmanBlock(ucvector *out, pngBitReader *reader, u32 btype, u32 max_output_size)
+{
+	u32 error = 0;
+	HuffmanTree tree_ll; /*the huffman tree for literal and length codes*/
+	HuffmanTree tree_d; /*the huffman tree for distance codes*/
+
+	HuffmanTree_init(&tree_ll);
+	HuffmanTree_init(&tree_d);
+
+	if (btype == 1)
+		error = getTreeInflateFixed(&tree_ll, &tree_d);
+	else /*if(btype == 2)*/
+		error = getTreeInflateDynamic(&tree_ll, &tree_d, reader);
+
+	while (!error) /*decode all symbols until end reached, breaks at end code*/ {
+		/*code_ll is literal, length or end code*/
+		u32 code_ll;
+		ensureBits25(
+			reader,
+			20); /* up to 15 for the huffman symbol, up to 5 for the length extra bits */
+		code_ll = huffmanDecodeSymbol(reader, &tree_ll);
+		if (code_ll <= 255) /*literal symbol*/ {
+			if (!ucvector_resize(out, out->size + 1))
+				ERROR_BREAK(83 /*alloc fail*/);
+			out->data[out->size - 1] = (u8)code_ll;
+		} else if (code_ll >= FIRST_LENGTH_CODE_INDEX &&
+			   code_ll <= LAST_LENGTH_CODE_INDEX) /*length code*/ {
+			u32 code_d, distance;
+			u32 numextrabits_l, numextrabits_d; /*extra bits for length and distance*/
+			u32 start, backward, length;
+
+			/*part 1: get length base*/
+			length = LENGTHBASE[code_ll - FIRST_LENGTH_CODE_INDEX];
+
+			/*part 2: get extra bits and add the value of that to length*/
+			numextrabits_l = LENGTHEXTRA[code_ll - FIRST_LENGTH_CODE_INDEX];
+			if (numextrabits_l != 0) {
+				/* bits already ensured above */
+				length += readBits(reader, numextrabits_l);
+			}
+
+			/*part 3: get distance code*/
+			ensureBits32(
+				reader,
+				28); /* up to 15 for the huffman symbol, up to 13 for the extra bits */
+			code_d = huffmanDecodeSymbol(reader, &tree_d);
+			if (code_d > 29) {
+				if (code_d <= 31) {
+					ERROR_BREAK(
+						18); /*error: invalid distance code (30-31 are never used)*/
+				} else /* if(code_d == INVALIDSYMBOL) */ {
+					ERROR_BREAK(
+						16); /*error: tried to read disallowed huffman symbol*/
+				}
+			}
+			distance = DISTANCEBASE[code_d];
+
+			/*part 4: get extra bits from distance*/
+			numextrabits_d = DISTANCEEXTRA[code_d];
+			if (numextrabits_d != 0) {
+				/* bits already ensured above */
+				distance += readBits(reader, numextrabits_d);
+			}
+
+			/*part 5: fill in all the out[n] values based on the length and dist*/
+			start = out->size;
+			if (distance > start)
+				ERROR_BREAK(52); /*too long backward distance*/
+			backward = start - distance;
+
+			if (!ucvector_resize(out, out->size + length))
+				ERROR_BREAK(83 /*alloc fail*/);
+			if (distance < length) {
+				u32 forward;
+				png_memcpy(out->data + start, out->data + backward, distance);
+				start += distance;
+				for (forward = distance; forward < length; ++forward) {
+					out->data[start++] = out->data[backward++];
+				}
+			} else {
+				png_memcpy(out->data + start, out->data + backward, length);
+			}
+		} else if (code_ll == 256) {
+			break; /*end code, break the loop*/
+		} else /*if(code_ll == INVALIDSYMBOL)*/ {
+			ERROR_BREAK(16); /*error: tried to read disallowed huffman symbol*/
+		}
+		/*check if any of the ensureBits above went out of bounds*/
+		if (reader->bp > reader->bitsize) {
+			/*return error code 10 or 11 depending on the situation that happened in huffmanDecodeSymbol
+      (10=no endcode, 11=wrong jump outside of tree)*/
+			/* TODO: revise error codes 10,11,50: the above comment is no longer valid */
+			ERROR_BREAK(51); /*error, bit pointer jumps past memory*/
+		}
+		if (max_output_size && out->size > max_output_size) {
+			ERROR_BREAK(109); /*error, larger than max size*/
+		}
+	}
+
+	HuffmanTree_cleanup(&tree_ll);
+	HuffmanTree_cleanup(&tree_d);
+
+	return error;
+}
+
+static u32 inflateNoCompression(ucvector *out, pngBitReader *reader,
+				const pngDecompressSettings *settings)
+{
+	u32 bytepos;
+	u32 size = reader->size;
+	u32 LEN, NLEN, error = 0;
+
+	/*go to first boundary of byte*/
+	bytepos = (reader->bp + 7u) >> 3u;
+
+	/*read LEN (2 bytes) and NLEN (2 bytes)*/
+	if (bytepos + 4 >= size)
+		return 52; /*error, bit pointer will jump past memory*/
+	LEN = (u32)reader->data[bytepos] + ((u32)reader->data[bytepos + 1] << 8u);
+	bytepos += 2;
+	NLEN = (u32)reader->data[bytepos] + ((u32)reader->data[bytepos + 1] << 8u);
+	bytepos += 2;
+
+	/*check if 16-bit NLEN is really the one's complement of LEN*/
+	if (!settings->ignore_nlen && LEN + NLEN != 65535) {
+		return 21; /*error: NLEN is not one's complement of LEN*/
+	}
+
+	if (!ucvector_resize(out, out->size + LEN))
+		return 83; /*alloc fail*/
+
+	/*read the literal data: LEN bytes are now stored in the out buffer*/
+	if (bytepos + LEN > size)
+		return 23; /*error: reading outside of in buffer*/
+
+	png_memcpy(out->data + out->size - LEN, reader->data + bytepos, LEN);
+	bytepos += LEN;
+
+	reader->bp = bytepos << 3u;
+
+	return error;
+}
+
+static u32 png_inflatev(ucvector *out, const u8 *in, u32 insize,
+			const pngDecompressSettings *settings)
+{
+	u32 BFINAL = 0;
+	pngBitReader reader;
+	u32 error = pngBitReader_init(&reader, in, insize);
+
+	if (error)
+		return error;
+
+	while (!BFINAL) {
+		u32 BTYPE;
+		if (!ensureBits9(&reader, 3))
+			return 52; /*error, bit pointer will jump past memory*/
+		BFINAL = readBits(&reader, 1);
+		BTYPE = readBits(&reader, 2);
+
+		if (BTYPE == 3)
+			return 20; /*error: invalid BTYPE*/
+		else if (BTYPE == 0)
+			error = inflateNoCompression(out, &reader, settings); /*no compression*/
+		else
+			error = inflateHuffmanBlock(
+				out, &reader, BTYPE,
+				settings->max_output_size); /*compression, BTYPE 01 or 10*/
+		if (!error && settings->max_output_size && out->size > settings->max_output_size)
+			error = 109;
+		if (error)
+			break;
+	}
+
+	return error;
+}
+
+u32 png_inflate(u8 **out, u32 *outsize, const u8 *in, u32 insize,
+		const pngDecompressSettings *settings)
+{
+	ucvector v = ucvector_init(*out, *outsize);
+	u32 error = png_inflatev(&v, in, insize, settings);
+	*out = v.data;
+	*outsize = v.size;
+	return error;
+}
+
+static u32 inflatev(ucvector *out, const u8 *in, u32 insize, const pngDecompressSettings *settings)
+{
+	if (settings->custom_inflate) {
+		u32 error = settings->custom_inflate(&out->data, &out->size, in, insize, settings);
+		out->allocsize = out->size;
+		if (error) {
+			/*the custom inflate is allowed to have its own error codes, however, we translate it to code 110*/
+			error = 110;
+			/*if there's a max output size, and the custom zlib returned error, then indicate that error instead*/
+			if (settings->max_output_size && out->size > settings->max_output_size)
+				error = 109;
+		}
+		return error;
+	} else {
+		return png_inflatev(out, in, insize, settings);
+	}
+}
+
+#endif /*PNG_COMPILE_DECODER*/
+
+#ifdef PNG_COMPILE_ENCODER
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Deflator (Compressor)                                                  / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+static const u32 MAX_SUPPORTED_DEFLATE_LENGTH = 258;
+
+/*search the index in the array, that has the largest value smaller than or equal to the given value,
+given array must be sorted (if no value is smaller, it returns the size of the given array)*/
+static u32 searchCodeIndex(const u32 *array, u32 array_size, u32 value)
+{
+	/*binary search (only small gain over linear). TODO: use CPU log2 instruction for getting symbols instead*/
+	u32 left = 1;
+	u32 right = array_size - 1;
+
+	while (left <= right) {
+		u32 mid = (left + right) >> 1;
+		if (array[mid] >= value)
+			right = mid - 1;
+		else
+			left = mid + 1;
+	}
+	if (left >= array_size || array[left] > value)
+		left--;
+	return left;
+}
+
+static void addLengthDistance(uivector *values, u32 length, u32 distance)
+{
+	/*values in encoded vector are those used by deflate:
+  0-255: literal bytes
+  256: end
+  257-285: length/distance pair (length code, followed by extra length bits, distance code, extra distance bits)
+  286-287: invalid*/
+
+	u32 length_code = (u32)searchCodeIndex(LENGTHBASE, 29, length);
+	u32 extra_length = (u32)(length - LENGTHBASE[length_code]);
+	u32 dist_code = (u32)searchCodeIndex(DISTANCEBASE, 30, distance);
+	u32 extra_distance = (u32)(distance - DISTANCEBASE[dist_code]);
+
+	u32 pos = values->size;
+	/*TODO: return error when this fails (out of memory)*/
+	u32 ok = uivector_resize(values, values->size + 4);
+	if (ok) {
+		values->data[pos + 0] = length_code + FIRST_LENGTH_CODE_INDEX;
+		values->data[pos + 1] = extra_length;
+		values->data[pos + 2] = dist_code;
+		values->data[pos + 3] = extra_distance;
+	}
+}
+
+/*3 bytes of data get encoded into two bytes. The hash cannot use more than 3
+bytes as input because 3 is the minimum match length for deflate*/
+static const u32 HASH_NUM_VALUES = 65536;
+static const u32 HASH_BIT_MASK =
+	65535; /*HASH_NUM_VALUES - 1, but C90 does not like that as initializer*/
+
+typedef struct Hash {
+	int *head; /*hash value to head circular pos - can be outdated if went around window*/
+	/*circular pos to prev circular pos*/
+	u16 *chain;
+	int *val; /*circular pos to hash value*/
+
+	/*TODO: do this not only for zeros but for any repeated byte. However for PNG
+  it's always going to be the zeros that dominate, so not important for PNG*/
+	int *headz; /*similar to head, but for chainz*/
+	u16 *chainz; /*those with same amount of zeros*/
+	u16 *zeros; /*length of zeros streak, used as a second hash chain*/
+} Hash;
+
+static u32 hash_init(Hash *hash, u32 windowsize)
+{
+	u32 i;
+	hash->head = (int *)png_malloc(sizeof(int) * HASH_NUM_VALUES);
+	hash->val = (int *)png_malloc(sizeof(int) * windowsize);
+	hash->chain = (u16 *)png_malloc(sizeof(u16) * windowsize);
+
+	hash->zeros = (u16 *)png_malloc(sizeof(u16) * windowsize);
+	hash->headz = (int *)png_malloc(sizeof(int) * (MAX_SUPPORTED_DEFLATE_LENGTH + 1));
+	hash->chainz = (u16 *)png_malloc(sizeof(u16) * windowsize);
+
+	if (!hash->head || !hash->chain || !hash->val || !hash->headz || !hash->chainz ||
+	    !hash->zeros) {
+		return 83; /*alloc fail*/
+	}
+
+	/*initialize hash table*/
+	for (i = 0; i != HASH_NUM_VALUES; ++i)
+		hash->head[i] = -1;
+	for (i = 0; i != windowsize; ++i)
+		hash->val[i] = -1;
+	for (i = 0; i != windowsize; ++i)
+		hash->chain[i] = i; /*same value as index indicates uninitialized*/
+
+	for (i = 0; i <= MAX_SUPPORTED_DEFLATE_LENGTH; ++i)
+		hash->headz[i] = -1;
+	for (i = 0; i != windowsize; ++i)
+		hash->chainz[i] = i; /*same value as index indicates uninitialized*/
+
+	return 0;
+}
+
+static void hash_cleanup(Hash *hash)
+{
+	png_free(hash->head);
+	png_free(hash->val);
+	png_free(hash->chain);
+
+	png_free(hash->zeros);
+	png_free(hash->headz);
+	png_free(hash->chainz);
+}
+
+static u32 getHash(const u8 *data, u32 size, u32 pos)
+{
+	u32 result = 0;
+	if (pos + 2 < size) {
+		/*A simple shift and xor hash is used. Since the data of PNGs is dominated
+    by zeroes due to the filters, a better hash does not have a significant
+    effect on speed in traversing the chain, and causes more time spend on
+    calculating the hash.*/
+		result ^= ((u32)data[pos + 0] << 0u);
+		result ^= ((u32)data[pos + 1] << 4u);
+		result ^= ((u32)data[pos + 2] << 8u);
+	} else {
+		u32 amount, i;
+		if (pos >= size)
+			return 0;
+		amount = size - pos;
+		for (i = 0; i != amount; ++i)
+			result ^= ((u32)data[pos + i] << (i * 8u));
+	}
+	return result & HASH_BIT_MASK;
+}
+
+static u32 countZeros(const u8 *data, u32 size, u32 pos)
+{
+	const u8 *start = data + pos;
+	const u8 *end = start + MAX_SUPPORTED_DEFLATE_LENGTH;
+	if (end > data + size)
+		end = data + size;
+	data = start;
+	while (data != end && *data == 0)
+		++data;
+	/*subtracting two addresses returned as 32-bit number (max value is MAX_SUPPORTED_DEFLATE_LENGTH)*/
+	return (u32)(data - start);
+}
+
+/*wpos = pos & (windowsize - 1)*/
+static void updateHashChain(Hash *hash, u32 wpos, u32 hashval, u16 numzeros)
+{
+	hash->val[wpos] = (int)hashval;
+	if (hash->head[hashval] != -1)
+		hash->chain[wpos] = hash->head[hashval];
+	hash->head[hashval] = (int)wpos;
+
+	hash->zeros[wpos] = numzeros;
+	if (hash->headz[numzeros] != -1)
+		hash->chainz[wpos] = hash->headz[numzeros];
+	hash->headz[numzeros] = (int)wpos;
+}
+
+/*
+LZ77-encode the data. Return value is error code. The input are raw bytes, the output
+is in the form of u32 integers with codes representing for example literal bytes, or
+length/distance pairs.
+It uses a hash table technique to let it encode faster. When doing LZ77 encoding, a
+sliding window (of windowsize) is used, and all past bytes in that window can be used as
+the "dictionary". A brute force search through all possible distances would be slow, and
+this hash technique is one out of several ways to speed this up.
+*/
+static u32 encodeLZ77(uivector *out, Hash *hash, const u8 *in, u32 inpos, u32 insize,
+		      u32 windowsize, u32 minmatch, u32 nicematch, u32 lazymatching)
+{
+	u32 pos;
+	u32 i, error = 0;
+	/*for large window lengths, assume the user wants no compression loss. Otherwise, max hash chain length speedup.*/
+	u32 maxchainlength = windowsize >= 8192 ? windowsize : windowsize / 8u;
+	u32 maxlazymatch = windowsize >= 8192 ? MAX_SUPPORTED_DEFLATE_LENGTH : 64;
+
+	u32 usezeros =
+		1; /*not sure if setting it to false for windowsize < 8192 is better or worse*/
+	u32 numzeros = 0;
+
+	u32 offset; /*the offset represents the distance in LZ77 terminology*/
+	u32 length;
+	u32 lazy = 0;
+	u32 lazylength = 0, lazyoffset = 0;
+	u32 hashval;
+	u32 current_offset, current_length;
+	u32 prev_offset;
+	const u8 *lastptr, *foreptr, *backptr;
+	u32 hashpos;
+
+	if (windowsize == 0 || windowsize > 32768)
+		return 60; /*error: windowsize smaller/larger than allowed*/
+	if ((windowsize & (windowsize - 1)) != 0)
+		return 90; /*error: must be power of two*/
+
+	if (nicematch > MAX_SUPPORTED_DEFLATE_LENGTH)
+		nicematch = MAX_SUPPORTED_DEFLATE_LENGTH;
+
+	for (pos = inpos; pos < insize; ++pos) {
+		u32 wpos = pos & (windowsize - 1); /*position for in 'circular' hash buffers*/
+		u32 chainlength = 0;
+
+		hashval = getHash(in, insize, pos);
+
+		if (usezeros && hashval == 0) {
+			if (numzeros == 0)
+				numzeros = countZeros(in, insize, pos);
+			else if (pos + numzeros > insize || in[pos + numzeros - 1] != 0)
+				--numzeros;
+		} else {
+			numzeros = 0;
+		}
+
+		updateHashChain(hash, wpos, hashval, numzeros);
+
+		/*the length and offset found for the current position*/
+		length = 0;
+		offset = 0;
+
+		hashpos = hash->chain[wpos];
+
+		lastptr = &in[insize < pos + MAX_SUPPORTED_DEFLATE_LENGTH ?
+					    insize :
+					    pos + MAX_SUPPORTED_DEFLATE_LENGTH];
+
+		/*search for the longest string*/
+		prev_offset = 0;
+		for (;;) {
+			if (chainlength++ >= maxchainlength)
+				break;
+			current_offset = (u32)(hashpos <= wpos ? wpos - hashpos :
+								       wpos - hashpos + windowsize);
+
+			if (current_offset < prev_offset)
+				break; /*stop when went completely around the circular buffer*/
+			prev_offset = current_offset;
+			if (current_offset > 0) {
+				/*test the next characters*/
+				foreptr = &in[pos];
+				backptr = &in[pos - current_offset];
+
+				/*common case in PNGs is lots of zeros. Quickly skip over them as a speedup*/
+				if (numzeros >= 3) {
+					u32 skip = hash->zeros[hashpos];
+					if (skip > numzeros)
+						skip = numzeros;
+					backptr += skip;
+					foreptr += skip;
+				}
+
+				while (foreptr != lastptr &&
+				       *backptr ==
+					       *foreptr) /*maximum supported length by deflate is max length*/
+				{
+					++backptr;
+					++foreptr;
+				}
+				current_length = (u32)(foreptr - &in[pos]);
+
+				if (current_length > length) {
+					length = current_length; /*the longest length*/
+					offset =
+						current_offset; /*the offset that is related to this longest length*/
+					/*jump out once a length of max length is found (speed gain). This also jumps
+          out if length is MAX_SUPPORTED_DEFLATE_LENGTH*/
+					if (current_length >= nicematch)
+						break;
+				}
+			}
+
+			if (hashpos == hash->chain[hashpos])
+				break;
+
+			if (numzeros >= 3 && length > numzeros) {
+				hashpos = hash->chainz[hashpos];
+				if (hash->zeros[hashpos] != numzeros)
+					break;
+			} else {
+				hashpos = hash->chain[hashpos];
+				/*outdated hash value, happens if particular value was not encountered in whole last window*/
+				if (hash->val[hashpos] != (int)hashval)
+					break;
+			}
+		}
+
+		if (lazymatching) {
+			if (!lazy && length >= 3 && length <= maxlazymatch &&
+			    length < MAX_SUPPORTED_DEFLATE_LENGTH) {
+				lazy = 1;
+				lazylength = length;
+				lazyoffset = offset;
+				continue; /*try the next byte*/
+			}
+			if (lazy) {
+				lazy = 0;
+				if (pos == 0)
+					ERROR_BREAK(81);
+				if (length > lazylength + 1) {
+					/*push the previous character as literal*/
+					if (!uivector_push_back(out, in[pos - 1]))
+						ERROR_BREAK(83 /*alloc fail*/);
+				} else {
+					length = lazylength;
+					offset = lazyoffset;
+					hash->head[hashval] =
+						-1; /*the same hashchain update will be done, this ensures no wrong alteration*/
+					hash->headz[numzeros] = -1; /*idem*/
+					--pos;
+				}
+			}
+		}
+		if (length >= 3 && offset > windowsize)
+			ERROR_BREAK(86 /*too big (or overflown negative) offset*/);
+
+		/*encode it as length/distance pair or literal value*/
+		if (length <
+		    3) /*only lengths of 3 or higher are supported as length/distance pair*/ {
+			if (!uivector_push_back(out, in[pos]))
+				ERROR_BREAK(83 /*alloc fail*/);
+		} else if (length < minmatch || (length == 3 && offset > 4096)) {
+			/*compensate for the fact that longer offsets have more extra bits, a
+      length of only 3 may be not worth it then*/
+			if (!uivector_push_back(out, in[pos]))
+				ERROR_BREAK(83 /*alloc fail*/);
+		} else {
+			addLengthDistance(out, length, offset);
+			for (i = 1; i < length; ++i) {
+				++pos;
+				wpos = pos & (windowsize - 1);
+				hashval = getHash(in, insize, pos);
+				if (usezeros && hashval == 0) {
+					if (numzeros == 0)
+						numzeros = countZeros(in, insize, pos);
+					else if (pos + numzeros > insize ||
+						 in[pos + numzeros - 1] != 0)
+						--numzeros;
+				} else {
+					numzeros = 0;
+				}
+				updateHashChain(hash, wpos, hashval, numzeros);
+			}
+		}
+	} /*end of the loop through each character of input*/
+
+	return error;
+}
+
+/* /////////////////////////////////////////////////////////////////////////// */
+
+static u32 deflateNoCompression(ucvector *out, const u8 *data, u32 datasize)
+{
+	/*non compressed deflate block data: 1 bit BFINAL,2 bits BTYPE,(5 bits): it jumps to start of next byte,
+  2 bytes LEN, 2 bytes NLEN, LEN bytes literal DATA*/
+
+	u32 i, numdeflateblocks = (datasize + 65534u) / 65535u;
+	u32 datapos = 0;
+	for (i = 0; i != numdeflateblocks; ++i) {
+		u32 BFINAL, BTYPE, LEN, NLEN;
+		u8 firstbyte;
+		u32 pos = out->size;
+
+		BFINAL = (i == numdeflateblocks - 1);
+		BTYPE = 0;
+
+		LEN = 65535;
+		if (datasize - datapos < 65535u)
+			LEN = (u32)datasize - datapos;
+		NLEN = 65535 - LEN;
+
+		if (!ucvector_resize(out, out->size + LEN + 5))
+			return 83; /*alloc fail*/
+
+		firstbyte = (u8)(BFINAL + ((BTYPE & 1u) << 1u) + ((BTYPE & 2u) << 1u));
+		out->data[pos + 0] = firstbyte;
+		out->data[pos + 1] = (u8)(LEN & 255);
+		out->data[pos + 2] = (u8)(LEN >> 8u);
+		out->data[pos + 3] = (u8)(NLEN & 255);
+		out->data[pos + 4] = (u8)(NLEN >> 8u);
+		png_memcpy(out->data + pos + 5, data + datapos, LEN);
+		datapos += LEN;
+	}
+
+	return 0;
+}
+
+/*
+write the lz77-encoded data, which has lit, len and dist codes, to compressed stream using huffman trees.
+tree_ll: the tree for lit and len codes.
+tree_d: the tree for distance codes.
+*/
+static void writeLZ77data(pngBitWriter *writer, const uivector *lz77_encoded,
+			  const HuffmanTree *tree_ll, const HuffmanTree *tree_d)
+{
+	u32 i = 0;
+	for (i = 0; i != lz77_encoded->size; ++i) {
+		u32 val = lz77_encoded->data[i];
+		writeBitsReversed(writer, tree_ll->codes[val], tree_ll->lengths[val]);
+		if (val > 256) /*for a length code, 3 more things have to be added*/ {
+			u32 length_index = val - FIRST_LENGTH_CODE_INDEX;
+			u32 n_length_extra_bits = LENGTHEXTRA[length_index];
+			u32 length_extra_bits = lz77_encoded->data[++i];
+
+			u32 distance_code = lz77_encoded->data[++i];
+
+			u32 distance_index = distance_code;
+			u32 n_distance_extra_bits = DISTANCEEXTRA[distance_index];
+			u32 distance_extra_bits = lz77_encoded->data[++i];
+
+			writeBits(writer, length_extra_bits, n_length_extra_bits);
+			writeBitsReversed(writer, tree_d->codes[distance_code],
+					  tree_d->lengths[distance_code]);
+			writeBits(writer, distance_extra_bits, n_distance_extra_bits);
+		}
+	}
+}
+
+/*Deflate for a block of type "dynamic", that is, with freely, optimally, created huffman trees*/
+static u32 deflateDynamic(pngBitWriter *writer, Hash *hash, const u8 *data, u32 datapos,
+			  u32 dataend, const pngCompressSettings *settings, u32 final)
+{
+	u32 error = 0;
+
+	/*
+  A block is compressed as follows: The PNG data is lz77 encoded, resulting in
+  literal bytes and length/distance pairs. This is then huffman compressed with
+  two huffman trees. One huffman tree is used for the lit and len values ("ll"),
+  another huffman tree is used for the dist values ("d"). These two trees are
+  stored using their code lengths, and to compress even more these code lengths
+  are also run-length encoded and huffman compressed. This gives a huffman tree
+  of code lengths "cl". The code lengths used to describe this third tree are
+  the code length code lengths ("clcl").
+  */
+
+	/*The lz77 encoded data, represented with integers since there will also be length and distance codes in it*/
+	uivector lz77_encoded;
+	HuffmanTree tree_ll; /*tree for lit,len values*/
+	HuffmanTree tree_d; /*tree for distance codes*/
+	HuffmanTree tree_cl; /*tree for encoding the code lengths representing tree_ll and tree_d*/
+	u32 *frequencies_ll = 0; /*frequency of lit,len codes*/
+	u32 *frequencies_d = 0; /*frequency of dist codes*/
+	u32 *frequencies_cl = 0; /*frequency of code length codes*/
+	u32 *bitlen_lld =
+		0; /*lit,len,dist code lengths (int bits), literally (without repeat codes).*/
+	u32 *bitlen_lld_e =
+		0; /*bitlen_lld encoded with repeat codes (this is a rudimentary run length compression)*/
+	u32 datasize = dataend - datapos;
+
+	/*
+  If we could call "bitlen_cl" the the code length code lengths ("clcl"), that is the bit lengths of codes to represent
+  tree_cl in CLCL_ORDER, then due to the huffman compression of huffman tree representations ("two levels"), there are
+  some analogies:
+  bitlen_lld is to tree_cl what data is to tree_ll and tree_d.
+  bitlen_lld_e is to bitlen_lld what lz77_encoded is to data.
+  bitlen_cl is to bitlen_lld_e what bitlen_lld is to lz77_encoded.
+  */
+
+	u32 BFINAL = final;
+	u32 i;
+	u32 numcodes_ll, numcodes_d, numcodes_lld, numcodes_lld_e, numcodes_cl;
+	u32 HLIT, HDIST, HCLEN;
+
+	uivector_init(&lz77_encoded);
+	HuffmanTree_init(&tree_ll);
+	HuffmanTree_init(&tree_d);
+	HuffmanTree_init(&tree_cl);
+	/* could fit on stack, but >1KB is on the larger side so allocate instead */
+	frequencies_ll = (u32 *)png_malloc(286 * sizeof(*frequencies_ll));
+	frequencies_d = (u32 *)png_malloc(30 * sizeof(*frequencies_d));
+	frequencies_cl = (u32 *)png_malloc(NUM_CODE_LENGTH_CODES * sizeof(*frequencies_cl));
+
+	if (!frequencies_ll || !frequencies_d || !frequencies_cl)
+		error = 83; /*alloc fail*/
+
+	/*This while loop never loops due to a break at the end, it is here to
+  allow breaking out of it to the cleanup phase on error conditions.*/
+	while (!error) {
+		png_memset(frequencies_ll, 0, 286 * sizeof(*frequencies_ll));
+		png_memset(frequencies_d, 0, 30 * sizeof(*frequencies_d));
+		png_memset(frequencies_cl, 0, NUM_CODE_LENGTH_CODES * sizeof(*frequencies_cl));
+
+		if (settings->use_lz77) {
+			error = encodeLZ77(&lz77_encoded, hash, data, datapos, dataend,
+					   settings->windowsize, settings->minmatch,
+					   settings->nicematch, settings->lazymatching);
+			if (error)
+				break;
+		} else {
+			if (!uivector_resize(&lz77_encoded, datasize))
+				ERROR_BREAK(83 /*alloc fail*/);
+			for (i = datapos; i < dataend; ++i)
+				lz77_encoded.data[i - datapos] =
+					data[i]; /*no LZ77, but still will be Huffman compressed*/
+		}
+
+		/*Count the frequencies of lit, len and dist codes*/
+		for (i = 0; i != lz77_encoded.size; ++i) {
+			u32 symbol = lz77_encoded.data[i];
+			++frequencies_ll[symbol];
+			if (symbol > 256) {
+				u32 dist = lz77_encoded.data[i + 2];
+				++frequencies_d[dist];
+				i += 3;
+			}
+		}
+		frequencies_ll[256] =
+			1; /*there will be exactly 1 end code, at the end of the block*/
+
+		/*Make both huffman trees, one for the lit and len codes, one for the dist codes*/
+		error = HuffmanTree_makeFromFrequencies(&tree_ll, frequencies_ll, 257, 286, 15);
+		if (error)
+			break;
+		/*2, not 1, is chosen for mincodes: some buggy PNG decoders require at least 2 symbols in the dist tree*/
+		error = HuffmanTree_makeFromFrequencies(&tree_d, frequencies_d, 2, 30, 15);
+		if (error)
+			break;
+
+		numcodes_ll = PNG_MIN(tree_ll.numcodes, 286);
+		numcodes_d = PNG_MIN(tree_d.numcodes, 30);
+		/*store the code lengths of both generated trees in bitlen_lld*/
+		numcodes_lld = numcodes_ll + numcodes_d;
+		bitlen_lld = (u32 *)png_malloc(numcodes_lld * sizeof(*bitlen_lld));
+		/*numcodes_lld_e never needs more size than bitlen_lld*/
+		bitlen_lld_e = (u32 *)png_malloc(numcodes_lld * sizeof(*bitlen_lld_e));
+		if (!bitlen_lld || !bitlen_lld_e)
+			ERROR_BREAK(83); /*alloc fail*/
+		numcodes_lld_e = 0;
+
+		for (i = 0; i != numcodes_ll; ++i)
+			bitlen_lld[i] = tree_ll.lengths[i];
+		for (i = 0; i != numcodes_d; ++i)
+			bitlen_lld[numcodes_ll + i] = tree_d.lengths[i];
+
+		/*run-length compress bitlen_ldd into bitlen_lld_e by using repeat codes 16 (copy length 3-6 times),
+    17 (3-10 zeroes), 18 (11-138 zeroes)*/
+		for (i = 0; i != numcodes_lld; ++i) {
+			u32 j = 0; /*amount of repetitions*/
+			while (i + j + 1 < numcodes_lld && bitlen_lld[i + j + 1] == bitlen_lld[i])
+				++j;
+
+			if (bitlen_lld[i] == 0 && j >= 2) /*repeat code for zeroes*/ {
+				++j; /*include the first zero*/
+				if (j <= 10) /*repeat code 17 supports max 10 zeroes*/ {
+					bitlen_lld_e[numcodes_lld_e++] = 17;
+					bitlen_lld_e[numcodes_lld_e++] = j - 3;
+				} else /*repeat code 18 supports max 138 zeroes*/ {
+					if (j > 138)
+						j = 138;
+					bitlen_lld_e[numcodes_lld_e++] = 18;
+					bitlen_lld_e[numcodes_lld_e++] = j - 11;
+				}
+				i += (j - 1);
+			} else if (j >= 3) /*repeat code for value other than zero*/ {
+				u32 k;
+				u32 num = j / 6u, rest = j % 6u;
+				bitlen_lld_e[numcodes_lld_e++] = bitlen_lld[i];
+				for (k = 0; k < num; ++k) {
+					bitlen_lld_e[numcodes_lld_e++] = 16;
+					bitlen_lld_e[numcodes_lld_e++] = 6 - 3;
+				}
+				if (rest >= 3) {
+					bitlen_lld_e[numcodes_lld_e++] = 16;
+					bitlen_lld_e[numcodes_lld_e++] = rest - 3;
+				} else
+					j -= rest;
+				i += j;
+			} else /*too short to benefit from repeat code*/ {
+				bitlen_lld_e[numcodes_lld_e++] = bitlen_lld[i];
+			}
+		}
+
+		/*generate tree_cl, the huffmantree of huffmantrees*/
+		for (i = 0; i != numcodes_lld_e; ++i) {
+			++frequencies_cl[bitlen_lld_e[i]];
+			/*after a repeat code come the bits that specify the number of repetitions,
+      those don't need to be in the frequencies_cl calculation*/
+			if (bitlen_lld_e[i] >= 16)
+				++i;
+		}
+
+		error = HuffmanTree_makeFromFrequencies(
+			&tree_cl, frequencies_cl, NUM_CODE_LENGTH_CODES, NUM_CODE_LENGTH_CODES, 7);
+		if (error)
+			break;
+
+		/*compute amount of code-length-code-lengths to output*/
+		numcodes_cl = NUM_CODE_LENGTH_CODES;
+		/*trim zeros at the end (using CLCL_ORDER), but minimum size must be 4 (see HCLEN below)*/
+		while (numcodes_cl > 4u && tree_cl.lengths[CLCL_ORDER[numcodes_cl - 1u]] == 0) {
+			numcodes_cl--;
+		}
+
+		/*
+    Write everything into the output
+
+    After the BFINAL and BTYPE, the dynamic block consists out of the following:
+    - 5 bits HLIT, 5 bits HDIST, 4 bits HCLEN
+    - (HCLEN+4)*3 bits code lengths of code length alphabet
+    - HLIT + 257 code lengths of lit/length alphabet (encoded using the code length
+      alphabet, + possible repetition codes 16, 17, 18)
+    - HDIST + 1 code lengths of distance alphabet (encoded using the code length
+      alphabet, + possible repetition codes 16, 17, 18)
+    - compressed data
+    - 256 (end code)
+    */
+
+		/*Write block type*/
+		writeBits(writer, BFINAL, 1);
+		writeBits(writer, 0, 1); /*first bit of BTYPE "dynamic"*/
+		writeBits(writer, 1, 1); /*second bit of BTYPE "dynamic"*/
+
+		/*write the HLIT, HDIST and HCLEN values*/
+		/*all three sizes take trimmed ending zeroes into account, done either by HuffmanTree_makeFromFrequencies
+    or in the loop for numcodes_cl above, which saves space. */
+		HLIT = (u32)(numcodes_ll - 257);
+		HDIST = (u32)(numcodes_d - 1);
+		HCLEN = (u32)(numcodes_cl - 4);
+		writeBits(writer, HLIT, 5);
+		writeBits(writer, HDIST, 5);
+		writeBits(writer, HCLEN, 4);
+
+		/*write the code lengths of the code length alphabet ("bitlen_cl")*/
+		for (i = 0; i != numcodes_cl; ++i)
+			writeBits(writer, tree_cl.lengths[CLCL_ORDER[i]], 3);
+
+		/*write the lengths of the lit/len AND the dist alphabet*/
+		for (i = 0; i != numcodes_lld_e; ++i) {
+			writeBitsReversed(writer, tree_cl.codes[bitlen_lld_e[i]],
+					  tree_cl.lengths[bitlen_lld_e[i]]);
+			/*extra bits of repeat codes*/
+			if (bitlen_lld_e[i] == 16)
+				writeBits(writer, bitlen_lld_e[++i], 2);
+			else if (bitlen_lld_e[i] == 17)
+				writeBits(writer, bitlen_lld_e[++i], 3);
+			else if (bitlen_lld_e[i] == 18)
+				writeBits(writer, bitlen_lld_e[++i], 7);
+		}
+
+		/*write the compressed data symbols*/
+		writeLZ77data(writer, &lz77_encoded, &tree_ll, &tree_d);
+		/*error: the length of the end code 256 must be larger than 0*/
+		if (tree_ll.lengths[256] == 0)
+			ERROR_BREAK(64);
+
+		/*write the end code*/
+		writeBitsReversed(writer, tree_ll.codes[256], tree_ll.lengths[256]);
+
+		break; /*end of error-while*/
+	}
+
+	/*cleanup*/
+	uivector_cleanup(&lz77_encoded);
+	HuffmanTree_cleanup(&tree_ll);
+	HuffmanTree_cleanup(&tree_d);
+	HuffmanTree_cleanup(&tree_cl);
+	png_free(frequencies_ll);
+	png_free(frequencies_d);
+	png_free(frequencies_cl);
+	png_free(bitlen_lld);
+	png_free(bitlen_lld_e);
+
+	return error;
+}
+
+static u32 deflateFixed(pngBitWriter *writer, Hash *hash, const u8 *data, u32 datapos, u32 dataend,
+			const pngCompressSettings *settings, u32 final)
+{
+	HuffmanTree tree_ll; /*tree for literal values and length codes*/
+	HuffmanTree tree_d; /*tree for distance codes*/
+
+	u32 BFINAL = final;
+	u32 error = 0;
+	u32 i;
+
+	HuffmanTree_init(&tree_ll);
+	HuffmanTree_init(&tree_d);
+
+	error = generateFixedLitLenTree(&tree_ll);
+	if (!error)
+		error = generateFixedDistanceTree(&tree_d);
+
+	if (!error) {
+		writeBits(writer, BFINAL, 1);
+		writeBits(writer, 1, 1); /*first bit of BTYPE*/
+		writeBits(writer, 0, 1); /*second bit of BTYPE*/
+
+		if (settings->use_lz77) /*LZ77 encoded*/ {
+			uivector lz77_encoded;
+			uivector_init(&lz77_encoded);
+			error = encodeLZ77(&lz77_encoded, hash, data, datapos, dataend,
+					   settings->windowsize, settings->minmatch,
+					   settings->nicematch, settings->lazymatching);
+			if (!error)
+				writeLZ77data(writer, &lz77_encoded, &tree_ll, &tree_d);
+			uivector_cleanup(&lz77_encoded);
+		} else /*no LZ77, but still will be Huffman compressed*/ {
+			for (i = datapos; i < dataend; ++i) {
+				writeBitsReversed(writer, tree_ll.codes[data[i]],
+						  tree_ll.lengths[data[i]]);
+			}
+		}
+		/*add END code*/
+		if (!error)
+			writeBitsReversed(writer, tree_ll.codes[256], tree_ll.lengths[256]);
+	}
+
+	/*cleanup*/
+	HuffmanTree_cleanup(&tree_ll);
+	HuffmanTree_cleanup(&tree_d);
+
+	return error;
+}
+
+static u32 png_deflatev(ucvector *out, const u8 *in, u32 insize,
+			const pngCompressSettings *settings)
+{
+	u32 error = 0;
+	u32 i, blocksize, numdeflateblocks;
+	Hash hash;
+	pngBitWriter writer;
+
+	pngBitWriter_init(&writer, out);
+
+	if (settings->btype > 2)
+		return 61;
+	else if (settings->btype == 0)
+		return deflateNoCompression(out, in, insize);
+	else if (settings->btype == 1)
+		blocksize = insize;
+	else /*if(settings->btype == 2)*/ {
+		/*on PNGs, deflate blocks of 65-262k seem to give most dense encoding*/
+		blocksize = insize / 8u + 8;
+		if (blocksize < 65536)
+			blocksize = 65536;
+		if (blocksize > 262144)
+			blocksize = 262144;
+	}
+
+	numdeflateblocks = (insize + blocksize - 1) / blocksize;
+	if (numdeflateblocks == 0)
+		numdeflateblocks = 1;
+
+	error = hash_init(&hash, settings->windowsize);
+
+	if (!error) {
+		for (i = 0; i != numdeflateblocks && !error; ++i) {
+			u32 final = (i == numdeflateblocks - 1);
+			u32 start = i * blocksize;
+			u32 end = start + blocksize;
+			if (end > insize)
+				end = insize;
+
+			if (settings->btype == 1)
+				error = deflateFixed(&writer, &hash, in, start, end, settings,
+						     final);
+			else if (settings->btype == 2)
+				error = deflateDynamic(&writer, &hash, in, start, end, settings,
+						       final);
+		}
+	}
+
+	hash_cleanup(&hash);
+
+	return error;
+}
+
+u32 png_deflate(u8 **out, u32 *outsize, const u8 *in, u32 insize,
+		const pngCompressSettings *settings)
+{
+	ucvector v = ucvector_init(*out, *outsize);
+	u32 error = png_deflatev(&v, in, insize, settings);
+	*out = v.data;
+	*outsize = v.size;
+	return error;
+}
+
+static u32 deflate(u8 **out, u32 *outsize, const u8 *in, u32 insize,
+		   const pngCompressSettings *settings)
+{
+	if (settings->custom_deflate) {
+		u32 error = settings->custom_deflate(out, outsize, in, insize, settings);
+		/*the custom deflate is allowed to have its own error codes, however, we translate it to code 111*/
+		return error ? 111 : 0;
+	} else {
+		return png_deflate(out, outsize, in, insize, settings);
+	}
+}
+
+#endif /*PNG_COMPILE_DECODER*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Adler32                                                                / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+static u32 update_adler32(u32 adler, const u8 *data, u32 len)
+{
+	u32 s1 = adler & 0xffffu;
+	u32 s2 = (adler >> 16u) & 0xffffu;
+
+	while (len != 0u) {
+		u32 i;
+		/*at least 5552 sums can be done before the sums overflow, saving a lot of module divisions*/
+		u32 amount = len > 5552u ? 5552u : len;
+		len -= amount;
+		for (i = 0; i != amount; ++i) {
+			s1 += (*data++);
+			s2 += s1;
+		}
+		s1 %= 65521u;
+		s2 %= 65521u;
+	}
+
+	return (s2 << 16u) | s1;
+}
+
+/*Return the adler32 of the bytes data[0..len-1]*/
+static u32 adler32(const u8 *data, u32 len)
+{
+	return update_adler32(1u, data, len);
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Zlib                                                                   / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef PNG_COMPILE_DECODER
+
+static u32 png_zlib_decompressv(ucvector *out, const u8 *in, u32 insize,
+				const pngDecompressSettings *settings)
+{
+	u32 error = 0;
+	u32 CM, CINFO, FDICT;
+
+	if (insize < 2)
+		return 53; /*error, size of zlib data too small*/
+	/*read information from zlib header*/
+	if ((in[0] * 256 + in[1]) % 31 != 0) {
+		/*error: 256 * in[0] + in[1] must be a multiple of 31, the FCHECK value is supposed to be made that way*/
+		return 24;
+	}
+
+	CM = in[0] & 15;
+	CINFO = (in[0] >> 4) & 15;
+	/*FCHECK = in[1] & 31;*/ /*FCHECK is already tested above*/
+	FDICT = (in[1] >> 5) & 1;
+	/*FLEVEL = (in[1] >> 6) & 3;*/ /*FLEVEL is not used here*/
+
+	if (CM != 8 || CINFO > 7) {
+		/*error: only compression method 8: inflate with sliding window of 32k is supported by the PNG spec*/
+		return 25;
+	}
+	if (FDICT != 0) {
+		/*error: the specification of PNG says about the zlib stream:
+      "The additional flags shall not specify a preset dictionary."*/
+		return 26;
+	}
+
+	error = inflatev(out, in + 2, insize - 2, settings);
+	if (error)
+		return error;
+
+	if (!settings->ignore_adler32) {
+		u32 ADLER32 = png_read32bitInt(&in[insize - 4]);
+		u32 checksum = adler32(out->data, (u32)(out->size));
+		if (checksum != ADLER32)
+			return 58; /*error, adler checksum not correct, data must be corrupted*/
+	}
+
+	return 0; /*no error*/
+}
+
+u32 png_zlib_decompress(u8 **out, u32 *outsize, const u8 *in, u32 insize,
+			const pngDecompressSettings *settings)
+{
+	ucvector v = ucvector_init(*out, *outsize);
+	u32 error = png_zlib_decompressv(&v, in, insize, settings);
+	*out = v.data;
+	*outsize = v.size;
+	return error;
+}
+
+/*expected_size is expected output size, to avoid intermediate allocations. Set to 0 if not known. */
+static u32 zlib_decompress(u8 **out, u32 *outsize, u32 expected_size, const u8 *in, u32 insize,
+			   const pngDecompressSettings *settings)
+{
+	u32 error;
+	if (settings->custom_zlib) {
+		error = settings->custom_zlib(out, outsize, in, insize, settings);
+		if (error) {
+			/*the custom zlib is allowed to have its own error codes, however, we translate it to code 110*/
+			error = 110;
+			/*if there's a max output size, and the custom zlib returned error, then indicate that error instead*/
+			if (settings->max_output_size && *outsize > settings->max_output_size)
+				error = 109;
+		}
+	} else {
+		ucvector v = ucvector_init(*out, *outsize);
+		if (expected_size) {
+			/*reserve the memory to avoid intermediate reallocations*/
+			ucvector_resize(&v, *outsize + expected_size);
+			v.size = *outsize;
+		}
+		error = png_zlib_decompressv(&v, in, insize, settings);
+		*out = v.data;
+		*outsize = v.size;
+	}
+	return error;
+}
+
+#endif /*PNG_COMPILE_DECODER*/
+
+#ifdef PNG_COMPILE_ENCODER
+
+u32 png_zlib_compress(u8 **out, u32 *outsize, const u8 *in, u32 insize,
+		      const pngCompressSettings *settings)
+{
+	u32 i;
+	u32 error;
+	u8 *deflatedata = 0;
+	u32 deflatesize = 0;
+
+	error = deflate(&deflatedata, &deflatesize, in, insize, settings);
+
+	*out = NULL;
+	*outsize = 0;
+	if (!error) {
+		*outsize = deflatesize + 6;
+		*out = (u8 *)png_malloc(*outsize);
+		if (!*out)
+			error = 83; /*alloc fail*/
+	}
+
+	if (!error) {
+		u32 ADLER32 = adler32(in, (u32)insize);
+		/*zlib data: 1 byte CMF (CM+CINFO), 1 byte FLG, deflate data, 4 byte ADLER32 checksum of the Decompressed data*/
+		u32 CMF =
+			120; /*0b01111000: CM 8, CINFO 7. With CINFO 7, any window size up to 32768 can be used.*/
+		u32 FLEVEL = 0;
+		u32 FDICT = 0;
+		u32 CMFFLG = 256 * CMF + FDICT * 32 + FLEVEL * 64;
+		u32 FCHECK = 31 - CMFFLG % 31;
+		CMFFLG += FCHECK;
+
+		(*out)[0] = (u8)(CMFFLG >> 8);
+		(*out)[1] = (u8)(CMFFLG & 255);
+		for (i = 0; i != deflatesize; ++i)
+			(*out)[i + 2] = deflatedata[i];
+		png_set32bitInt(&(*out)[*outsize - 4], ADLER32);
+	}
+
+	png_free(deflatedata);
+	return error;
+}
+
+/* compress using the default or custom zlib function */
+static u32 zlib_compress(u8 **out, u32 *outsize, const u8 *in, u32 insize,
+			 const pngCompressSettings *settings)
+{
+	if (settings->custom_zlib) {
+		u32 error = settings->custom_zlib(out, outsize, in, insize, settings);
+		/*the custom zlib is allowed to have its own error codes, however, we translate it to code 111*/
+		return error ? 111 : 0;
+	} else {
+		return png_zlib_compress(out, outsize, in, insize, settings);
+	}
+}
+
+#endif /*PNG_COMPILE_ENCODER*/
+
+#else /*no PNG_COMPILE_ZLIB*/
+
+#ifdef PNG_COMPILE_DECODER
+static u32 zlib_decompress(u8 **out, u32 *outsize, u32 expected_size, const u8 *in, u32 insize,
+			   const pngDecompressSettings *settings)
+{
+	if (!settings->custom_zlib)
+		return 87; /*no custom zlib function provided */
+	(void)expected_size;
+	return settings->custom_zlib(out, outsize, in, insize, settings);
+}
+#endif /*PNG_COMPILE_DECODER*/
+#ifdef PNG_COMPILE_ENCODER
+static u32 zlib_compress(u8 **out, u32 *outsize, const u8 *in, u32 insize,
+			 const pngCompressSettings *settings)
+{
+	if (!settings->custom_zlib)
+		return 87; /*no custom zlib function provided */
+	return settings->custom_zlib(out, outsize, in, insize, settings);
+}
+#endif /*PNG_COMPILE_ENCODER*/
+
+#endif /*PNG_COMPILE_ZLIB*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef PNG_COMPILE_ENCODER
+
+/*this is a good tradeoff between speed and compression ratio*/
+#define DEFAULT_WINDOWSIZE 2048
+
+void png_compress_settings_init(pngCompressSettings *settings)
+{
+	/*compress with dynamic huffman tree (not in the mathematical sense, just not the predefined one)*/
+	settings->btype = 2;
+	settings->use_lz77 = 1;
+	settings->windowsize = DEFAULT_WINDOWSIZE;
+	settings->minmatch = 3;
+	settings->nicematch = 128;
+	settings->lazymatching = 1;
+
+	settings->custom_zlib = 0;
+	settings->custom_deflate = 0;
+	settings->custom_context = 0;
+}
+
+const pngCompressSettings png_default_compress_settings = { 2, 1, DEFAULT_WINDOWSIZE, 3, 128, 1, 0,
+							    0, 0 };
+
+#endif /*PNG_COMPILE_ENCODER*/
+
+#ifdef PNG_COMPILE_DECODER
+
+void png_decompress_settings_init(pngDecompressSettings *settings)
+{
+	settings->ignore_adler32 = 0;
+	settings->ignore_nlen = 0;
+	settings->max_output_size = 0;
+
+	settings->custom_zlib = 0;
+	settings->custom_inflate = 0;
+	settings->custom_context = 0;
+}
+
+const pngDecompressSettings png_default_decompress_settings = { 0, 0, 0, 0, 0, 0 };
+
+#endif /*PNG_COMPILE_DECODER*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* // End of Zlib related code. Begin of PNG related code.                 // */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef PNG_COMPILE_PNG
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / CRC32                                                                  / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifndef PNG_NO_COMPILE_CRC
+/* CRC polynomial: 0xedb88320 */
+static u32 png_crc32_table[256] = {
+	0u,	     1996959894u, 3993919788u, 2567524794u, 124634137u,	 1886057615u, 3915621685u,
+	2657392035u, 249268274u,  2044508324u, 3772115230u, 2547177864u, 162941995u,  2125561021u,
+	3887607047u, 2428444049u, 498536548u,  1789927666u, 4089016648u, 2227061214u, 450548861u,
+	1843258603u, 4107580753u, 2211677639u, 325883990u,  1684777152u, 4251122042u, 2321926636u,
+	335633487u,  1661365465u, 4195302755u, 2366115317u, 997073096u,	 1281953886u, 3579855332u,
+	2724688242u, 1006888145u, 1258607687u, 3524101629u, 2768942443u, 901097722u,  1119000684u,
+	3686517206u, 2898065728u, 853044451u,  1172266101u, 3705015759u, 2882616665u, 651767980u,
+	1373503546u, 3369554304u, 3218104598u, 565507253u,  1454621731u, 3485111705u, 3099436303u,
+	671266974u,  1594198024u, 3322730930u, 2970347812u, 795835527u,	 1483230225u, 3244367275u,
+	3060149565u, 1994146192u, 31158534u,   2563907772u, 4023717930u, 1907459465u, 112637215u,
+	2680153253u, 3904427059u, 2013776290u, 251722036u,  2517215374u, 3775830040u, 2137656763u,
+	141376813u,  2439277719u, 3865271297u, 1802195444u, 476864866u,	 2238001368u, 4066508878u,
+	1812370925u, 453092731u,  2181625025u, 4111451223u, 1706088902u, 314042704u,  2344532202u,
+	4240017532u, 1658658271u, 366619977u,  2362670323u, 4224994405u, 1303535960u, 984961486u,
+	2747007092u, 3569037538u, 1256170817u, 1037604311u, 2765210733u, 3554079995u, 1131014506u,
+	879679996u,  2909243462u, 3663771856u, 1141124467u, 855842277u,	 2852801631u, 3708648649u,
+	1342533948u, 654459306u,  3188396048u, 3373015174u, 1466479909u, 544179635u,  3110523913u,
+	3462522015u, 1591671054u, 702138776u,  2966460450u, 3352799412u, 1504918807u, 783551873u,
+	3082640443u, 3233442989u, 3988292384u, 2596254646u, 62317068u,	 1957810842u, 3939845945u,
+	2647816111u, 81470997u,	  1943803523u, 3814918930u, 2489596804u, 225274430u,  2053790376u,
+	3826175755u, 2466906013u, 167816743u,  2097651377u, 4027552580u, 2265490386u, 503444072u,
+	1762050814u, 4150417245u, 2154129355u, 426522225u,  1852507879u, 4275313526u, 2312317920u,
+	282753626u,  1742555852u, 4189708143u, 2394877945u, 397917763u,	 1622183637u, 3604390888u,
+	2714866558u, 953729732u,  1340076626u, 3518719985u, 2797360999u, 1068828381u, 1219638859u,
+	3624741850u, 2936675148u, 906185462u,  1090812512u, 3747672003u, 2825379669u, 829329135u,
+	1181335161u, 3412177804u, 3160834842u, 628085408u,  1382605366u, 3423369109u, 3138078467u,
+	570562233u,  1426400815u, 3317316542u, 2998733608u, 733239954u,	 1555261956u, 3268935591u,
+	3050360625u, 752459403u,  1541320221u, 2607071920u, 3965973030u, 1969922972u, 40735498u,
+	2617837225u, 3943577151u, 1913087877u, 83908371u,   2512341634u, 3803740692u, 2075208622u,
+	213261112u,  2463272603u, 3855990285u, 2094854071u, 198958881u,	 2262029012u, 4057260610u,
+	1759359992u, 534414190u,  2176718541u, 4139329115u, 1873836001u, 414664567u,  2282248934u,
+	4279200368u, 1711684554u, 285281116u,  2405801727u, 4167216745u, 1634467795u, 376229701u,
+	2685067896u, 3608007406u, 1308918612u, 956543938u,  2808555105u, 3495958263u, 1231636301u,
+	1047427035u, 2932959818u, 3654703836u, 1088359270u, 936918000u,	 2847714899u, 3736837829u,
+	1202900863u, 817233897u,  3183342108u, 3401237130u, 1404277552u, 615818150u,  3134207493u,
+	3453421203u, 1423857449u, 601450431u,  3009837614u, 3294710456u, 1567103746u, 711928724u,
+	3020668471u, 3272380065u, 1510334235u, 755167117u
+};
+
+/*Return the CRC of the bytes buf[0..len-1].*/
+u32 png_crc32(const u8 *data, u32 length)
+{
+	u32 r = 0xffffffffu;
+	u32 i;
+	for (i = 0; i < length; ++i) {
+		r = png_crc32_table[(r ^ data[i]) & 0xffu] ^ (r >> 8u);
+	}
+	return r ^ 0xffffffffu;
+}
+#else /* !PNG_NO_COMPILE_CRC */
+u32 png_crc32(const u8 *data, u32 length);
+#endif /* !PNG_NO_COMPILE_CRC */
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Reading and writing PNG color channel bits                             / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/* The color channel bits of less-than-8-bit pixels are read with the MSB of bytes first,
+so pngBitWriter and pngBitReader can't be used for those. */
+
+static u8 readBitFromReversedStream(u32 *bitpointer, const u8 *bitstream)
+{
+	u8 result = (u8)((bitstream[(*bitpointer) >> 3] >> (7 - ((*bitpointer) & 0x7))) & 1);
+	++(*bitpointer);
+	return result;
+}
+
+/* TODO: make this faster */
+static u32 readBitsFromReversedStream(u32 *bitpointer, const u8 *bitstream, u32 nbits)
+{
+	u32 result = 0;
+	u32 i;
+	for (i = 0; i < nbits; ++i) {
+		result <<= 1u;
+		result |= (u32)readBitFromReversedStream(bitpointer, bitstream);
+	}
+	return result;
+}
+
+static void setBitOfReversedStream(u32 *bitpointer, u8 *bitstream, u8 bit)
+{
+	/*the current bit in bitstream may be 0 or 1 for this to work*/
+	if (bit == 0)
+		bitstream[(*bitpointer) >> 3u] &= (u8)(~(1u << (7u - ((*bitpointer) & 7u))));
+	else
+		bitstream[(*bitpointer) >> 3u] |= (1u << (7u - ((*bitpointer) & 7u)));
+	++(*bitpointer);
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / PNG chunks                                                             / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+u32 png_chunk_length(const u8 *chunk)
+{
+	return png_read32bitInt(&chunk[0]);
+}
+
+void png_chunk_type(char type[5], const u8 *chunk)
+{
+	u32 i;
+	for (i = 0; i != 4; ++i)
+		type[i] = (char)chunk[4 + i];
+	type[4] = 0; /*null termination char*/
+}
+
+u8 png_chunk_type_equals(const u8 *chunk, const char *type)
+{
+	if (png_strlen(type) != 4)
+		return 0;
+	return (chunk[4] == type[0] && chunk[5] == type[1] && chunk[6] == type[2] &&
+		chunk[7] == type[3]);
+}
+
+u8 png_chunk_ancillary(const u8 *chunk)
+{
+	return ((chunk[4] & 32) != 0);
+}
+
+u8 png_chunk_private(const u8 *chunk)
+{
+	return ((chunk[6] & 32) != 0);
+}
+
+u8 png_chunk_safetocopy(const u8 *chunk)
+{
+	return ((chunk[7] & 32) != 0);
+}
+
+u8 *png_chunk_data(u8 *chunk)
+{
+	return &chunk[8];
+}
+
+const u8 *png_chunk_data_const(const u8 *chunk)
+{
+	return &chunk[8];
+}
+
+u32 png_chunk_check_crc(const u8 *chunk)
+{
+	u32 length = png_chunk_length(chunk);
+	u32 CRC = png_read32bitInt(&chunk[length + 8]);
+	/*the CRC is taken of the data and the 4 chunk type letters, not the length*/
+	u32 checksum = png_crc32(&chunk[4], length + 4);
+	if (CRC != checksum)
+		return 1;
+	else
+		return 0;
+}
+
+void png_chunk_generate_crc(u8 *chunk)
+{
+	u32 length = png_chunk_length(chunk);
+	u32 CRC = png_crc32(&chunk[4], length + 4);
+	png_set32bitInt(chunk + 8 + length, CRC);
+}
+
+u8 *png_chunk_next(u8 *chunk, u8 *end)
+{
+	if (chunk >= end || end - chunk < 12)
+		return end; /*too small to contain a chunk*/
+	if (chunk[0] == 0x89 && chunk[1] == 0x50 && chunk[2] == 0x4e && chunk[3] == 0x47 &&
+	    chunk[4] == 0x0d && chunk[5] == 0x0a && chunk[6] == 0x1a && chunk[7] == 0x0a) {
+		/* Is PNG magic header at start of PNG file. Jump to first actual chunk. */
+		return chunk + 8;
+	} else {
+		u32 total_chunk_length;
+		u8 *result;
+		if (png_addofl(png_chunk_length(chunk), 12, &total_chunk_length))
+			return end;
+		result = chunk + total_chunk_length;
+		if (result < chunk)
+			return end; /*pointer overflow*/
+		return result;
+	}
+}
+
+const u8 *png_chunk_next_const(const u8 *chunk, const u8 *end)
+{
+	if (chunk >= end || end - chunk < 12)
+		return end; /*too small to contain a chunk*/
+	if (chunk[0] == 0x89 && chunk[1] == 0x50 && chunk[2] == 0x4e && chunk[3] == 0x47 &&
+	    chunk[4] == 0x0d && chunk[5] == 0x0a && chunk[6] == 0x1a && chunk[7] == 0x0a) {
+		/* Is PNG magic header at start of PNG file. Jump to first actual chunk. */
+		return chunk + 8;
+	} else {
+		u32 total_chunk_length;
+		const u8 *result;
+		if (png_addofl(png_chunk_length(chunk), 12, &total_chunk_length))
+			return end;
+		result = chunk + total_chunk_length;
+		if (result < chunk)
+			return end; /*pointer overflow*/
+		return result;
+	}
+}
+
+u8 *png_chunk_find(u8 *chunk, u8 *end, const char type[5])
+{
+	for (;;) {
+		if (chunk >= end || end - chunk < 12)
+			return 0; /* past file end: chunk + 12 > end */
+		if (png_chunk_type_equals(chunk, type))
+			return chunk;
+		chunk = png_chunk_next(chunk, end);
+	}
+}
+
+const u8 *png_chunk_find_const(const u8 *chunk, const u8 *end, const char type[5])
+{
+	for (;;) {
+		if (chunk >= end || end - chunk < 12)
+			return 0; /* past file end: chunk + 12 > end */
+		if (png_chunk_type_equals(chunk, type))
+			return chunk;
+		chunk = png_chunk_next_const(chunk, end);
+	}
+}
+
+u32 png_chunk_append(u8 **out, u32 *outsize, const u8 *chunk)
+{
+	u32 i;
+	u32 total_chunk_length, new_length;
+	u8 *chunk_start, *new_buffer;
+
+	if (png_addofl(png_chunk_length(chunk), 12, &total_chunk_length))
+		return 77;
+	if (png_addofl(*outsize, total_chunk_length, &new_length))
+		return 77;
+
+	new_buffer = (u8 *)png_realloc(*out, new_length);
+	if (!new_buffer)
+		return 83; /*alloc fail*/
+	(*out) = new_buffer;
+	(*outsize) = new_length;
+	chunk_start = &(*out)[new_length - total_chunk_length];
+
+	for (i = 0; i != total_chunk_length; ++i)
+		chunk_start[i] = chunk[i];
+
+	return 0;
+}
+
+/*Sets length and name and allocates the space for data and crc but does not
+set data or crc yet. Returns the start of the chunk in chunk. The start of
+the data is at chunk + 8. To finalize chunk, add the data, then use
+png_chunk_generate_crc */
+static u32 png_chunk_init(u8 **chunk, ucvector *out, u32 length, const char *type)
+{
+	u32 new_length = out->size;
+	if (png_addofl(new_length, length, &new_length))
+		return 77;
+	if (png_addofl(new_length, 12, &new_length))
+		return 77;
+	if (!ucvector_resize(out, new_length))
+		return 83; /*alloc fail*/
+	*chunk = out->data + new_length - length - 12u;
+
+	/*1: length*/
+	png_set32bitInt(*chunk, length);
+
+	/*2: chunk name (4 letters)*/
+	png_memcpy(*chunk + 4, type, 4);
+
+	return 0;
+}
+
+/* like png_chunk_create but with custom allocsize */
+static u32 png_chunk_createv(ucvector *out, u32 length, const char *type, const u8 *data)
+{
+	u8 *chunk;
+	CERROR_TRY_RETURN(png_chunk_init(&chunk, out, length, type));
+
+	/*3: the data*/
+	png_memcpy(chunk + 8, data, length);
+
+	/*4: CRC (of the chunkname characters and the data)*/
+	png_chunk_generate_crc(chunk);
+
+	return 0;
+}
+
+u32 png_chunk_create(u8 **out, u32 *outsize, u32 length, const char *type, const u8 *data)
+{
+	ucvector v = ucvector_init(*out, *outsize);
+	u32 error = png_chunk_createv(&v, length, type, data);
+	*out = v.data;
+	*outsize = v.size;
+	return error;
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / Color types, channels, bits                                            / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*checks if the colortype is valid and the bitdepth bd is allowed for this colortype.
+Return value is a png error code.*/
+static u32 checkColorValidity(pngColorType colortype, u32 bd)
+{
+	switch (colortype) {
+	case LCT_GREY:
+		if (!(bd == 1 || bd == 2 || bd == 4 || bd == 8 || bd == 16))
+			return 37;
+		break;
+	case LCT_RGB:
+		if (!(bd == 8 || bd == 16))
+			return 37;
+		break;
+	case LCT_PALETTE:
+		if (!(bd == 1 || bd == 2 || bd == 4 || bd == 8))
+			return 37;
+		break;
+	case LCT_GREY_ALPHA:
+		if (!(bd == 8 || bd == 16))
+			return 37;
+		break;
+	case LCT_RGBA:
+		if (!(bd == 8 || bd == 16))
+			return 37;
+		break;
+	case LCT_MAX_OCTET_VALUE:
+		return 31; /* invalid color type */
+	default:
+		return 31; /* invalid color type */
+	}
+	return 0; /*allowed color type / bits combination*/
+}
+
+static u32 getNumColorChannels(pngColorType colortype)
+{
+	switch (colortype) {
+	case LCT_GREY:
+		return 1;
+	case LCT_RGB:
+		return 3;
+	case LCT_PALETTE:
+		return 1;
+	case LCT_GREY_ALPHA:
+		return 2;
+	case LCT_RGBA:
+		return 4;
+	case LCT_MAX_OCTET_VALUE:
+		return 0; /* invalid color type */
+	default:
+		return 0; /*invalid color type*/
+	}
+}
+
+static u32 png_get_bpp_lct(pngColorType colortype, u32 bitdepth)
+{
+	/*bits per pixel is amount of channels * bits per channel*/
+	return getNumColorChannels(colortype) * bitdepth;
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+
+void png_color_mode_init(pngColorMode *info)
+{
+	info->key_defined = 0;
+	info->key_r = info->key_g = info->key_b = 0;
+	info->colortype = LCT_RGBA;
+	info->bitdepth = 8;
+	info->palette = 0;
+	info->palettesize = 0;
+}
+
+/*allocates palette memory if needed, and initializes all colors to black*/
+static void png_color_mode_alloc_palette(pngColorMode *info)
+{
+	u32 i;
+	/*if the palette is already allocated, it will have size 1024 so no reallocation needed in that case*/
+	/*the palette must have room for up to 256 colors with 4 bytes each.*/
+	if (!info->palette)
+		info->palette = (u8 *)png_malloc(1024);
+	if (!info->palette)
+		return; /*alloc fail*/
+	for (i = 0; i != 256; ++i) {
+		/*Initialize all unused colors with black, the value used for invalid palette indices.
+    This is an error according to the PNG spec, but common PNG decoders make it black instead.
+    That makes color conversion slightly faster due to no error handling needed.*/
+		info->palette[i * 4 + 0] = 0;
+		info->palette[i * 4 + 1] = 0;
+		info->palette[i * 4 + 2] = 0;
+		info->palette[i * 4 + 3] = 255;
+	}
+}
+
+void png_color_mode_cleanup(pngColorMode *info)
+{
+	png_palette_clear(info);
+}
+
+u32 png_color_mode_copy(pngColorMode *dest, const pngColorMode *source)
+{
+	png_color_mode_cleanup(dest);
+	png_memcpy(dest, source, sizeof(pngColorMode));
+	if (source->palette) {
+		dest->palette = (u8 *)png_malloc(1024);
+		if (!dest->palette && source->palettesize)
+			return 83; /*alloc fail*/
+		png_memcpy(dest->palette, source->palette, source->palettesize * 4);
+	}
+	return 0;
+}
+
+pngColorMode png_color_mode_make(pngColorType colortype, u32 bitdepth)
+{
+	pngColorMode result;
+	png_color_mode_init(&result);
+	result.colortype = colortype;
+	result.bitdepth = bitdepth;
+	return result;
+}
+
+static int png_color_mode_equal(const pngColorMode *a, const pngColorMode *b)
+{
+	u32 i;
+	if (a->colortype != b->colortype)
+		return 0;
+	if (a->bitdepth != b->bitdepth)
+		return 0;
+	if (a->key_defined != b->key_defined)
+		return 0;
+	if (a->key_defined) {
+		if (a->key_r != b->key_r)
+			return 0;
+		if (a->key_g != b->key_g)
+			return 0;
+		if (a->key_b != b->key_b)
+			return 0;
+	}
+	if (a->palettesize != b->palettesize)
+		return 0;
+	for (i = 0; i != a->palettesize * 4; ++i) {
+		if (a->palette[i] != b->palette[i])
+			return 0;
+	}
+	return 1;
+}
+
+void png_palette_clear(pngColorMode *info)
+{
+	if (info->palette)
+		png_free(info->palette);
+	info->palette = 0;
+	info->palettesize = 0;
+}
+
+u32 png_palette_add(pngColorMode *info, u8 r, u8 g, u8 b, u8 a)
+{
+	if (!info->palette) /*allocate palette if empty*/ {
+		png_color_mode_alloc_palette(info);
+		if (!info->palette)
+			return 83; /*alloc fail*/
+	}
+	if (info->palettesize >= 256) {
+		return 108; /*too many palette values*/
+	}
+	info->palette[4 * info->palettesize + 0] = r;
+	info->palette[4 * info->palettesize + 1] = g;
+	info->palette[4 * info->palettesize + 2] = b;
+	info->palette[4 * info->palettesize + 3] = a;
+	++info->palettesize;
+	return 0;
+}
+
+/*calculate bits per pixel out of colortype and bitdepth*/
+u32 png_get_bpp(const pngColorMode *info)
+{
+	return png_get_bpp_lct(info->colortype, info->bitdepth);
+}
+
+u32 png_get_channels(const pngColorMode *info)
+{
+	return getNumColorChannels(info->colortype);
+}
+
+u32 png_is_greyscale_type(const pngColorMode *info)
+{
+	return info->colortype == LCT_GREY || info->colortype == LCT_GREY_ALPHA;
+}
+
+u32 png_is_alpha_type(const pngColorMode *info)
+{
+	return (info->colortype & 4) != 0; /*4 or 6*/
+}
+
+u32 png_is_palette_type(const pngColorMode *info)
+{
+	return info->colortype == LCT_PALETTE;
+}
+
+u32 png_has_palette_alpha(const pngColorMode *info)
+{
+	u32 i;
+	for (i = 0; i != info->palettesize; ++i) {
+		if (info->palette[i * 4 + 3] < 255)
+			return 1;
+	}
+	return 0;
+}
+
+u32 png_can_have_alpha(const pngColorMode *info)
+{
+	return info->key_defined || png_is_alpha_type(info) || png_has_palette_alpha(info);
+}
+
+static u32 png_get_raw_size_lct(u32 w, u32 h, pngColorType colortype, u32 bitdepth)
+{
+	u32 bpp = png_get_bpp_lct(colortype, bitdepth);
+	u32 n = (u32)w * (u32)h;
+	return ((n / 8u) * bpp) + ((n & 7u) * bpp + 7u) / 8u;
+}
+
+u32 png_get_raw_size(u32 w, u32 h, const pngColorMode *color)
+{
+	return png_get_raw_size_lct(w, h, color->colortype, color->bitdepth);
+}
+
+#ifdef PNG_COMPILE_PNG
+
+/*in an idat chunk, each scanline is a multiple of 8 bits, unlike the png output buffer,
+and in addition has one extra byte per line: the filter byte. So this gives a larger
+result than png_get_raw_size. Set h to 1 to get the size of 1 row including filter byte. */
+static u32 png_get_raw_size_idat(u32 w, u32 h, u32 bpp)
+{
+	/* + 1 for the filter byte, and possibly plus padding bits per line. */
+	/* Ignoring casts, the expression is equal to (w * bpp + 7) / 8 + 1, but avoids overflow of w * bpp */
+	u32 line = ((u32)(w / 8u) * bpp) + 1u + ((w & 7u) * bpp + 7u) / 8u;
+	return (u32)h * line;
+}
+
+#ifdef PNG_COMPILE_DECODER
+/*Safely checks whether u32 overflow can be caused due to amount of pixels.
+This check is overcautious rather than precise. If this check indicates no overflow,
+you can safely compute in a u32 (but not an u32):
+-(u32)w * (u32)h * 8
+-amount of bytes in IDAT (including filter, padding and Adam7 bytes)
+-amount of bytes in raw color model
+Returns 1 if overflow possible, 0 if not.
+*/
+static int png_pixel_overflow(u32 w, u32 h, const pngColorMode *pngcolor,
+			      const pngColorMode *rawcolor)
+{
+	u32 bpp = PNG_MAX(png_get_bpp(pngcolor), png_get_bpp(rawcolor));
+	u32 numpixels, total;
+	u32 line; /* bytes per line in worst case */
+
+	if (png_mulofl((u32)w, (u32)h, &numpixels))
+		return 1;
+	if (png_mulofl(numpixels, 8, &total))
+		return 1; /* bit pointer with 8-bit color, or 8 bytes per channel color */
+
+	/* Bytes per scanline with the expression "(w / 8u) * bpp) + ((w & 7u) * bpp + 7u) / 8u" */
+	if (png_mulofl((u32)(w / 8u), bpp, &line))
+		return 1;
+	if (png_addofl(line, ((w & 7u) * bpp + 7u) / 8u, &line))
+		return 1;
+
+	if (png_addofl(line, 5, &line))
+		return 1; /* 5 bytes overhead per line: 1 filterbyte, 4 for Adam7 worst case */
+	if (png_mulofl(line, h, &total))
+		return 1; /* Total bytes in worst case */
+
+	return 0; /* no overflow */
+}
+#endif /*PNG_COMPILE_DECODER*/
+#endif /*PNG_COMPILE_PNG*/
+
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+
+static void pngUnknownChunks_init(pngInfo *info)
+{
+	u32 i;
+	for (i = 0; i != 3; ++i)
+		info->unknown_chunks_data[i] = 0;
+	for (i = 0; i != 3; ++i)
+		info->unknown_chunks_size[i] = 0;
+}
+
+static void pngUnknownChunks_cleanup(pngInfo *info)
+{
+	u32 i;
+	for (i = 0; i != 3; ++i)
+		png_free(info->unknown_chunks_data[i]);
+}
+
+static u32 pngUnknownChunks_copy(pngInfo *dest, const pngInfo *src)
+{
+	u32 i;
+
+	pngUnknownChunks_cleanup(dest);
+
+	for (i = 0; i != 3; ++i) {
+		u32 j;
+		dest->unknown_chunks_size[i] = src->unknown_chunks_size[i];
+		dest->unknown_chunks_data[i] = (u8 *)png_malloc(src->unknown_chunks_size[i]);
+		if (!dest->unknown_chunks_data[i] && dest->unknown_chunks_size[i])
+			return 83; /*alloc fail*/
+		for (j = 0; j < src->unknown_chunks_size[i]; ++j) {
+			dest->unknown_chunks_data[i][j] = src->unknown_chunks_data[i][j];
+		}
+	}
+
+	return 0;
+}
+
+/******************************************************************************/
+
+static void pngText_init(pngInfo *info)
+{
+	info->text_num = 0;
+	info->text_keys = NULL;
+	info->text_strings = NULL;
+}
+
+static void pngText_cleanup(pngInfo *info)
+{
+	u32 i;
+	for (i = 0; i != info->text_num; ++i) {
+		string_cleanup(&info->text_keys[i]);
+		string_cleanup(&info->text_strings[i]);
+	}
+	png_free(info->text_keys);
+	png_free(info->text_strings);
+}
+
+static u32 pngText_copy(pngInfo *dest, const pngInfo *source)
+{
+	u32 i = 0;
+	dest->text_keys = NULL;
+	dest->text_strings = NULL;
+	dest->text_num = 0;
+	for (i = 0; i != source->text_num; ++i) {
+		CERROR_TRY_RETURN(
+			png_add_text(dest, source->text_keys[i], source->text_strings[i]));
+	}
+	return 0;
+}
+
+static u32 png_add_text_sized(pngInfo *info, const char *key, const char *str, u32 size)
+{
+	char **new_keys =
+		(char **)(png_realloc(info->text_keys, sizeof(char *) * (info->text_num + 1)));
+	char **new_strings =
+		(char **)(png_realloc(info->text_strings, sizeof(char *) * (info->text_num + 1)));
+
+	if (new_keys)
+		info->text_keys = new_keys;
+	if (new_strings)
+		info->text_strings = new_strings;
+
+	if (!new_keys || !new_strings)
+		return 83; /*alloc fail*/
+
+	++info->text_num;
+	info->text_keys[info->text_num - 1] = alloc_string(key);
+	info->text_strings[info->text_num - 1] = alloc_string_sized(str, size);
+	if (!info->text_keys[info->text_num - 1] || !info->text_strings[info->text_num - 1])
+		return 83; /*alloc fail*/
+
+	return 0;
+}
+
+u32 png_add_text(pngInfo *info, const char *key, const char *str)
+{
+	return png_add_text_sized(info, key, str, png_strlen(str));
+}
+
+void png_clear_text(pngInfo *info)
+{
+	pngText_cleanup(info);
+}
+
+/******************************************************************************/
+
+static void pngIText_init(pngInfo *info)
+{
+	info->itext_num = 0;
+	info->itext_keys = NULL;
+	info->itext_langtags = NULL;
+	info->itext_transkeys = NULL;
+	info->itext_strings = NULL;
+}
+
+static void pngIText_cleanup(pngInfo *info)
+{
+	u32 i;
+	for (i = 0; i != info->itext_num; ++i) {
+		string_cleanup(&info->itext_keys[i]);
+		string_cleanup(&info->itext_langtags[i]);
+		string_cleanup(&info->itext_transkeys[i]);
+		string_cleanup(&info->itext_strings[i]);
+	}
+	png_free(info->itext_keys);
+	png_free(info->itext_langtags);
+	png_free(info->itext_transkeys);
+	png_free(info->itext_strings);
+}
+
+static u32 pngIText_copy(pngInfo *dest, const pngInfo *source)
+{
+	u32 i = 0;
+	dest->itext_keys = NULL;
+	dest->itext_langtags = NULL;
+	dest->itext_transkeys = NULL;
+	dest->itext_strings = NULL;
+	dest->itext_num = 0;
+	for (i = 0; i != source->itext_num; ++i) {
+		CERROR_TRY_RETURN(
+			png_add_itext(dest, source->itext_keys[i], source->itext_langtags[i],
+				      source->itext_transkeys[i], source->itext_strings[i]));
+	}
+	return 0;
+}
+
+void png_clear_itext(pngInfo *info)
+{
+	pngIText_cleanup(info);
+}
+
+static u32 png_add_itext_sized(pngInfo *info, const char *key, const char *langtag,
+			       const char *transkey, const char *str, u32 size)
+{
+	char **new_keys =
+		(char **)(png_realloc(info->itext_keys, sizeof(char *) * (info->itext_num + 1)));
+	char **new_langtags = (char **)(png_realloc(info->itext_langtags,
+						    sizeof(char *) * (info->itext_num + 1)));
+	char **new_transkeys = (char **)(png_realloc(info->itext_transkeys,
+						     sizeof(char *) * (info->itext_num + 1)));
+	char **new_strings =
+		(char **)(png_realloc(info->itext_strings, sizeof(char *) * (info->itext_num + 1)));
+
+	if (new_keys)
+		info->itext_keys = new_keys;
+	if (new_langtags)
+		info->itext_langtags = new_langtags;
+	if (new_transkeys)
+		info->itext_transkeys = new_transkeys;
+	if (new_strings)
+		info->itext_strings = new_strings;
+
+	if (!new_keys || !new_langtags || !new_transkeys || !new_strings)
+		return 83; /*alloc fail*/
+
+	++info->itext_num;
+
+	info->itext_keys[info->itext_num - 1] = alloc_string(key);
+	info->itext_langtags[info->itext_num - 1] = alloc_string(langtag);
+	info->itext_transkeys[info->itext_num - 1] = alloc_string(transkey);
+	info->itext_strings[info->itext_num - 1] = alloc_string_sized(str, size);
+
+	return 0;
+}
+
+u32 png_add_itext(pngInfo *info, const char *key, const char *langtag, const char *transkey,
+		  const char *str)
+{
+	return png_add_itext_sized(info, key, langtag, transkey, str, png_strlen(str));
+}
+
+/* same as set but does not delete */
+static u32 png_assign_icc(pngInfo *info, const char *name, const u8 *profile, u32 profile_size)
+{
+	if (profile_size == 0)
+		return 100; /*invalid ICC profile size*/
+
+	info->iccp_name = alloc_string(name);
+	info->iccp_profile = (u8 *)png_malloc(profile_size);
+
+	if (!info->iccp_name || !info->iccp_profile)
+		return 83; /*alloc fail*/
+
+	png_memcpy(info->iccp_profile, profile, profile_size);
+	info->iccp_profile_size = profile_size;
+
+	return 0; /*ok*/
+}
+
+u32 png_set_icc(pngInfo *info, const char *name, const u8 *profile, u32 profile_size)
+{
+	if (info->iccp_name)
+		png_clear_icc(info);
+	info->iccp_defined = 1;
+
+	return png_assign_icc(info, name, profile, profile_size);
+}
+
+void png_clear_icc(pngInfo *info)
+{
+	string_cleanup(&info->iccp_name);
+	png_free(info->iccp_profile);
+	info->iccp_profile = NULL;
+	info->iccp_profile_size = 0;
+	info->iccp_defined = 0;
+}
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+
+void png_info_init(pngInfo *info)
+{
+	png_color_mode_init(&info->color);
+	info->interlace_method = 0;
+	info->compression_method = 0;
+	info->filter_method = 0;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+	info->background_defined = 0;
+	info->background_r = info->background_g = info->background_b = 0;
+
+	pngText_init(info);
+	pngIText_init(info);
+
+	info->time_defined = 0;
+	info->phys_defined = 0;
+
+	info->gama_defined = 0;
+	info->chrm_defined = 0;
+	info->srgb_defined = 0;
+	info->iccp_defined = 0;
+	info->iccp_name = NULL;
+	info->iccp_profile = NULL;
+
+	pngUnknownChunks_init(info);
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+}
+
+void png_info_cleanup(pngInfo *info)
+{
+	png_color_mode_cleanup(&info->color);
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+	pngText_cleanup(info);
+	pngIText_cleanup(info);
+
+	png_clear_icc(info);
+
+	pngUnknownChunks_cleanup(info);
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+}
+
+u32 png_info_copy(pngInfo *dest, const pngInfo *source)
+{
+	png_info_cleanup(dest);
+	png_memcpy(dest, source, sizeof(pngInfo));
+	png_color_mode_init(&dest->color);
+	CERROR_TRY_RETURN(png_color_mode_copy(&dest->color, &source->color));
+
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+	CERROR_TRY_RETURN(pngText_copy(dest, source));
+	CERROR_TRY_RETURN(pngIText_copy(dest, source));
+	if (source->iccp_defined) {
+		CERROR_TRY_RETURN(png_assign_icc(dest, source->iccp_name, source->iccp_profile,
+						 source->iccp_profile_size));
+	}
+
+	pngUnknownChunks_init(dest);
+	CERROR_TRY_RETURN(pngUnknownChunks_copy(dest, source));
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+	return 0;
+}
+
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*index: bitgroup index, bits: bitgroup size(1, 2 or 4), in: bitgroup value, out: octet array to add bits to*/
+static void addColorBits(u8 *out, u32 index, u32 bits, u32 in)
+{
+	u32 m = bits == 1 ? 7 : bits == 2 ? 3 : 1; /*8 / bits - 1*/
+	/*p = the partial index in the byte, e.g. with 4 palettebits it is 0 for first half or 1 for second half*/
+	u32 p = index & m;
+	in &= (1u << bits) - 1u; /*filter out any other bits of the input value*/
+	in = in << (bits * (m - p));
+	if (p == 0)
+		out[index * bits / 8u] = in;
+	else
+		out[index * bits / 8u] |= in;
+}
+
+typedef struct ColorTree ColorTree;
+
+/*
+One node of a color tree
+This is the data structure used to count the number of unique colors and to get a palette
+index for a color. It's like an octree, but because the alpha channel is used too, each
+node has 16 instead of 8 children.
+*/
+struct ColorTree {
+	ColorTree *children[16]; /*up to 16 pointers to ColorTree of next level*/
+	int index; /*the payload. Only has a meaningful value if this is in the last level*/
+};
+
+static void color_tree_init(ColorTree *tree)
+{
+	png_memset(tree->children, 0, 16 * sizeof(*tree->children));
+	tree->index = -1;
+}
+
+static void color_tree_cleanup(ColorTree *tree)
+{
+	int i;
+	for (i = 0; i != 16; ++i) {
+		if (tree->children[i]) {
+			color_tree_cleanup(tree->children[i]);
+			png_free(tree->children[i]);
+		}
+	}
+}
+
+/*returns -1 if color not present, its index otherwise*/
+static int color_tree_get(ColorTree *tree, u8 r, u8 g, u8 b, u8 a)
+{
+	int bit = 0;
+	for (bit = 0; bit < 8; ++bit) {
+		int i = 8 * ((r >> bit) & 1) + 4 * ((g >> bit) & 1) + 2 * ((b >> bit) & 1) +
+			1 * ((a >> bit) & 1);
+		if (!tree->children[i])
+			return -1;
+		else
+			tree = tree->children[i];
+	}
+	return tree ? tree->index : -1;
+}
+
+#ifdef PNG_COMPILE_ENCODER
+static int color_tree_has(ColorTree *tree, u8 r, u8 g, u8 b, u8 a)
+{
+	return color_tree_get(tree, r, g, b, a) >= 0;
+}
+#endif /*PNG_COMPILE_ENCODER*/
+
+/*color is not allowed to already exist.
+Index should be >= 0 (it's signed to be compatible with using -1 for "doesn't exist")
+Returns error code, or 0 if ok*/
+static u32 color_tree_add(ColorTree *tree, u8 r, u8 g, u8 b, u8 a, u32 index)
+{
+	int bit;
+	for (bit = 0; bit < 8; ++bit) {
+		int i = 8 * ((r >> bit) & 1) + 4 * ((g >> bit) & 1) + 2 * ((b >> bit) & 1) +
+			1 * ((a >> bit) & 1);
+		if (!tree->children[i]) {
+			tree->children[i] = (ColorTree *)png_malloc(sizeof(ColorTree));
+			if (!tree->children[i])
+				return 83; /*alloc fail*/
+			color_tree_init(tree->children[i]);
+		}
+		tree = tree->children[i];
+	}
+	tree->index = (int)index;
+	return 0;
+}
+
+/*put a pixel, given its RGBA color, into image of any color type*/
+static u32 rgba8ToPixel(u8 *out, u32 i, const pngColorMode *mode, ColorTree *tree /*for palette*/,
+			u8 r, u8 g, u8 b, u8 a)
+{
+	if (mode->colortype == LCT_GREY) {
+		u8 gray = r; /*((u16)r + g + b) / 3u;*/
+		if (mode->bitdepth == 8)
+			out[i] = gray;
+		else if (mode->bitdepth == 16)
+			out[i * 2 + 0] = out[i * 2 + 1] = gray;
+		else {
+			/*take the most significant bits of gray*/
+			gray = ((u32)gray >> (8u - mode->bitdepth)) & ((1u << mode->bitdepth) - 1u);
+			addColorBits(out, i, mode->bitdepth, gray);
+		}
+	} else if (mode->colortype == LCT_RGB) {
+		if (mode->bitdepth == 8) {
+			out[i * 3 + 0] = r;
+			out[i * 3 + 1] = g;
+			out[i * 3 + 2] = b;
+		} else {
+			out[i * 6 + 0] = out[i * 6 + 1] = r;
+			out[i * 6 + 2] = out[i * 6 + 3] = g;
+			out[i * 6 + 4] = out[i * 6 + 5] = b;
+		}
+	} else if (mode->colortype == LCT_PALETTE) {
+		int index = color_tree_get(tree, r, g, b, a);
+		if (index < 0)
+			return 82; /*color not in palette*/
+		if (mode->bitdepth == 8)
+			out[i] = index;
+		else
+			addColorBits(out, i, mode->bitdepth, (u32)index);
+	} else if (mode->colortype == LCT_GREY_ALPHA) {
+		u8 gray = r; /*((u16)r + g + b) / 3u;*/
+		if (mode->bitdepth == 8) {
+			out[i * 2 + 0] = gray;
+			out[i * 2 + 1] = a;
+		} else if (mode->bitdepth == 16) {
+			out[i * 4 + 0] = out[i * 4 + 1] = gray;
+			out[i * 4 + 2] = out[i * 4 + 3] = a;
+		}
+	} else if (mode->colortype == LCT_RGBA) {
+		if (mode->bitdepth == 8) {
+			out[i * 4 + 0] = r;
+			out[i * 4 + 1] = g;
+			out[i * 4 + 2] = b;
+			out[i * 4 + 3] = a;
+		} else {
+			out[i * 8 + 0] = out[i * 8 + 1] = r;
+			out[i * 8 + 2] = out[i * 8 + 3] = g;
+			out[i * 8 + 4] = out[i * 8 + 5] = b;
+			out[i * 8 + 6] = out[i * 8 + 7] = a;
+		}
+	}
+
+	return 0; /*no error*/
+}
+
+/*put a pixel, given its RGBA16 color, into image of any color 16-bitdepth type*/
+static void rgba16ToPixel(u8 *out, u32 i, const pngColorMode *mode, u16 r, u16 g, u16 b, u16 a)
+{
+	if (mode->colortype == LCT_GREY) {
+		u16 gray = r; /*((u32)r + g + b) / 3u;*/
+		out[i * 2 + 0] = (gray >> 8) & 255;
+		out[i * 2 + 1] = gray & 255;
+	} else if (mode->colortype == LCT_RGB) {
+		out[i * 6 + 0] = (r >> 8) & 255;
+		out[i * 6 + 1] = r & 255;
+		out[i * 6 + 2] = (g >> 8) & 255;
+		out[i * 6 + 3] = g & 255;
+		out[i * 6 + 4] = (b >> 8) & 255;
+		out[i * 6 + 5] = b & 255;
+	} else if (mode->colortype == LCT_GREY_ALPHA) {
+		u16 gray = r; /*((u32)r + g + b) / 3u;*/
+		out[i * 4 + 0] = (gray >> 8) & 255;
+		out[i * 4 + 1] = gray & 255;
+		out[i * 4 + 2] = (a >> 8) & 255;
+		out[i * 4 + 3] = a & 255;
+	} else if (mode->colortype == LCT_RGBA) {
+		out[i * 8 + 0] = (r >> 8) & 255;
+		out[i * 8 + 1] = r & 255;
+		out[i * 8 + 2] = (g >> 8) & 255;
+		out[i * 8 + 3] = g & 255;
+		out[i * 8 + 4] = (b >> 8) & 255;
+		out[i * 8 + 5] = b & 255;
+		out[i * 8 + 6] = (a >> 8) & 255;
+		out[i * 8 + 7] = a & 255;
+	}
+}
+
+/*Get RGBA8 color of pixel with index i (y * width + x) from the raw image with given color type.*/
+static void getPixelColorRGBA8(u8 *r, u8 *g, u8 *b, u8 *a, const u8 *in, u32 i,
+			       const pngColorMode *mode)
+{
+	if (mode->colortype == LCT_GREY) {
+		if (mode->bitdepth == 8) {
+			*r = *g = *b = in[i];
+			if (mode->key_defined && *r == mode->key_r)
+				*a = 0;
+			else
+				*a = 255;
+		} else if (mode->bitdepth == 16) {
+			*r = *g = *b = in[i * 2 + 0];
+			if (mode->key_defined &&
+			    256U * in[i * 2 + 0] + in[i * 2 + 1] == mode->key_r)
+				*a = 0;
+			else
+				*a = 255;
+		} else {
+			u32 highest = ((1U << mode->bitdepth) -
+				       1U); /*highest possible value for this bit depth*/
+			u32 j = i * mode->bitdepth;
+			u32 value = readBitsFromReversedStream(&j, in, mode->bitdepth);
+			*r = *g = *b = (value * 255) / highest;
+			if (mode->key_defined && value == mode->key_r)
+				*a = 0;
+			else
+				*a = 255;
+		}
+	} else if (mode->colortype == LCT_RGB) {
+		if (mode->bitdepth == 8) {
+			*r = in[i * 3 + 0];
+			*g = in[i * 3 + 1];
+			*b = in[i * 3 + 2];
+			if (mode->key_defined && *r == mode->key_r && *g == mode->key_g &&
+			    *b == mode->key_b)
+				*a = 0;
+			else
+				*a = 255;
+		} else {
+			*r = in[i * 6 + 0];
+			*g = in[i * 6 + 2];
+			*b = in[i * 6 + 4];
+			if (mode->key_defined &&
+			    256U * in[i * 6 + 0] + in[i * 6 + 1] == mode->key_r &&
+			    256U * in[i * 6 + 2] + in[i * 6 + 3] == mode->key_g &&
+			    256U * in[i * 6 + 4] + in[i * 6 + 5] == mode->key_b)
+				*a = 0;
+			else
+				*a = 255;
+		}
+	} else if (mode->colortype == LCT_PALETTE) {
+		u32 index;
+		if (mode->bitdepth == 8)
+			index = in[i];
+		else {
+			u32 j = i * mode->bitdepth;
+			index = readBitsFromReversedStream(&j, in, mode->bitdepth);
+		}
+		/*out of bounds of palette not checked: see png_color_mode_alloc_palette.*/
+		*r = mode->palette[index * 4 + 0];
+		*g = mode->palette[index * 4 + 1];
+		*b = mode->palette[index * 4 + 2];
+		*a = mode->palette[index * 4 + 3];
+	} else if (mode->colortype == LCT_GREY_ALPHA) {
+		if (mode->bitdepth == 8) {
+			*r = *g = *b = in[i * 2 + 0];
+			*a = in[i * 2 + 1];
+		} else {
+			*r = *g = *b = in[i * 4 + 0];
+			*a = in[i * 4 + 2];
+		}
+	} else if (mode->colortype == LCT_RGBA) {
+		if (mode->bitdepth == 8) {
+			*r = in[i * 4 + 0];
+			*g = in[i * 4 + 1];
+			*b = in[i * 4 + 2];
+			*a = in[i * 4 + 3];
+		} else {
+			*r = in[i * 8 + 0];
+			*g = in[i * 8 + 2];
+			*b = in[i * 8 + 4];
+			*a = in[i * 8 + 6];
+		}
+	}
+}
+
+/*Similar to getPixelColorRGBA8, but with all the for loops inside of the color
+mode test cases, optimized to convert the colors much faster, when converting
+to the common case of RGBA with 8 bit per channel. buffer must be RGBA with
+enough memory.*/
+static void getPixelColorsRGBA8(u8 *PNG_RESTRICT buffer, u32 numpixels, const u8 *PNG_RESTRICT in,
+				const pngColorMode *mode)
+{
+	u32 num_channels = 4;
+	u32 i;
+	if (mode->colortype == LCT_GREY) {
+		if (mode->bitdepth == 8) {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				buffer[0] = buffer[1] = buffer[2] = in[i];
+				buffer[3] = 255;
+			}
+			if (mode->key_defined) {
+				buffer -= numpixels * num_channels;
+				for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+					if (buffer[0] == mode->key_r)
+						buffer[3] = 0;
+				}
+			}
+		} else if (mode->bitdepth == 16) {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				buffer[0] = buffer[1] = buffer[2] = in[i * 2];
+				buffer[3] =
+					mode->key_defined && 256U * in[i * 2 + 0] + in[i * 2 + 1] ==
+								     mode->key_r ?
+						      0 :
+						      255;
+			}
+		} else {
+			u32 highest = ((1U << mode->bitdepth) -
+				       1U); /*highest possible value for this bit depth*/
+			u32 j = 0;
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				u32 value = readBitsFromReversedStream(&j, in, mode->bitdepth);
+				buffer[0] = buffer[1] = buffer[2] = (value * 255) / highest;
+				buffer[3] = mode->key_defined && value == mode->key_r ? 0 : 255;
+			}
+		}
+	} else if (mode->colortype == LCT_RGB) {
+		if (mode->bitdepth == 8) {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				png_memcpy(buffer, &in[i * 3], 3);
+				buffer[3] = 255;
+			}
+			if (mode->key_defined) {
+				buffer -= numpixels * num_channels;
+				for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+					if (buffer[0] == mode->key_r && buffer[1] == mode->key_g &&
+					    buffer[2] == mode->key_b)
+						buffer[3] = 0;
+				}
+			}
+		} else {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				buffer[0] = in[i * 6 + 0];
+				buffer[1] = in[i * 6 + 2];
+				buffer[2] = in[i * 6 + 4];
+				buffer[3] = mode->key_defined &&
+							    256U * in[i * 6 + 0] + in[i * 6 + 1] ==
+								    mode->key_r &&
+							    256U * in[i * 6 + 2] + in[i * 6 + 3] ==
+								    mode->key_g &&
+							    256U * in[i * 6 + 4] + in[i * 6 + 5] ==
+								    mode->key_b ?
+							  0 :
+							  255;
+			}
+		}
+	} else if (mode->colortype == LCT_PALETTE) {
+		if (mode->bitdepth == 8) {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				u32 index = in[i];
+				/*out of bounds of palette not checked: see png_color_mode_alloc_palette.*/
+				png_memcpy(buffer, &mode->palette[index * 4], 4);
+			}
+		} else {
+			u32 j = 0;
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				u32 index = readBitsFromReversedStream(&j, in, mode->bitdepth);
+				/*out of bounds of palette not checked: see png_color_mode_alloc_palette.*/
+				png_memcpy(buffer, &mode->palette[index * 4], 4);
+			}
+		}
+	} else if (mode->colortype == LCT_GREY_ALPHA) {
+		if (mode->bitdepth == 8) {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				buffer[0] = buffer[1] = buffer[2] = in[i * 2 + 0];
+				buffer[3] = in[i * 2 + 1];
+			}
+		} else {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				buffer[0] = buffer[1] = buffer[2] = in[i * 4 + 0];
+				buffer[3] = in[i * 4 + 2];
+			}
+		}
+	} else if (mode->colortype == LCT_RGBA) {
+		if (mode->bitdepth == 8) {
+			png_memcpy(buffer, in, numpixels * 4);
+		} else {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				buffer[0] = in[i * 8 + 0];
+				buffer[1] = in[i * 8 + 2];
+				buffer[2] = in[i * 8 + 4];
+				buffer[3] = in[i * 8 + 6];
+			}
+		}
+	}
+}
+
+/*Similar to getPixelColorsRGBA8, but with 3-channel RGB output.*/
+static void getPixelColorsRGB8(u8 *PNG_RESTRICT buffer, u32 numpixels, const u8 *PNG_RESTRICT in,
+			       const pngColorMode *mode)
+{
+	const u32 num_channels = 3;
+	u32 i;
+	if (mode->colortype == LCT_GREY) {
+		if (mode->bitdepth == 8) {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				buffer[0] = buffer[1] = buffer[2] = in[i];
+			}
+		} else if (mode->bitdepth == 16) {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				buffer[0] = buffer[1] = buffer[2] = in[i * 2];
+			}
+		} else {
+			u32 highest = ((1U << mode->bitdepth) -
+				       1U); /*highest possible value for this bit depth*/
+			u32 j = 0;
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				u32 value = readBitsFromReversedStream(&j, in, mode->bitdepth);
+				buffer[0] = buffer[1] = buffer[2] = (value * 255) / highest;
+			}
+		}
+	} else if (mode->colortype == LCT_RGB) {
+		if (mode->bitdepth == 8) {
+			png_memcpy(buffer, in, numpixels * 3);
+		} else {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				buffer[0] = in[i * 6 + 0];
+				buffer[1] = in[i * 6 + 2];
+				buffer[2] = in[i * 6 + 4];
+			}
+		}
+	} else if (mode->colortype == LCT_PALETTE) {
+		if (mode->bitdepth == 8) {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				u32 index = in[i];
+				/*out of bounds of palette not checked: see png_color_mode_alloc_palette.*/
+				png_memcpy(buffer, &mode->palette[index * 4], 3);
+			}
+		} else {
+			u32 j = 0;
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				u32 index = readBitsFromReversedStream(&j, in, mode->bitdepth);
+				/*out of bounds of palette not checked: see png_color_mode_alloc_palette.*/
+				png_memcpy(buffer, &mode->palette[index * 4], 3);
+			}
+		}
+	} else if (mode->colortype == LCT_GREY_ALPHA) {
+		if (mode->bitdepth == 8) {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				buffer[0] = buffer[1] = buffer[2] = in[i * 2 + 0];
+			}
+		} else {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				buffer[0] = buffer[1] = buffer[2] = in[i * 4 + 0];
+			}
+		}
+	} else if (mode->colortype == LCT_RGBA) {
+		if (mode->bitdepth == 8) {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				png_memcpy(buffer, &in[i * 4], 3);
+			}
+		} else {
+			for (i = 0; i != numpixels; ++i, buffer += num_channels) {
+				buffer[0] = in[i * 8 + 0];
+				buffer[1] = in[i * 8 + 2];
+				buffer[2] = in[i * 8 + 4];
+			}
+		}
+	}
+}
+
+/*Get RGBA16 color of pixel with index i (y * width + x) from the raw image with
+given color type, but the given color type must be 16-bit itself.*/
+static void getPixelColorRGBA16(u16 *r, u16 *g, u16 *b, u16 *a, const u8 *in, u32 i,
+				const pngColorMode *mode)
+{
+	if (mode->colortype == LCT_GREY) {
+		*r = *g = *b = 256 * in[i * 2 + 0] + in[i * 2 + 1];
+		if (mode->key_defined && 256U * in[i * 2 + 0] + in[i * 2 + 1] == mode->key_r)
+			*a = 0;
+		else
+			*a = 65535;
+	} else if (mode->colortype == LCT_RGB) {
+		*r = 256u * in[i * 6 + 0] + in[i * 6 + 1];
+		*g = 256u * in[i * 6 + 2] + in[i * 6 + 3];
+		*b = 256u * in[i * 6 + 4] + in[i * 6 + 5];
+		if (mode->key_defined && 256u * in[i * 6 + 0] + in[i * 6 + 1] == mode->key_r &&
+		    256u * in[i * 6 + 2] + in[i * 6 + 3] == mode->key_g &&
+		    256u * in[i * 6 + 4] + in[i * 6 + 5] == mode->key_b)
+			*a = 0;
+		else
+			*a = 65535;
+	} else if (mode->colortype == LCT_GREY_ALPHA) {
+		*r = *g = *b = 256u * in[i * 4 + 0] + in[i * 4 + 1];
+		*a = 256u * in[i * 4 + 2] + in[i * 4 + 3];
+	} else if (mode->colortype == LCT_RGBA) {
+		*r = 256u * in[i * 8 + 0] + in[i * 8 + 1];
+		*g = 256u * in[i * 8 + 2] + in[i * 8 + 3];
+		*b = 256u * in[i * 8 + 4] + in[i * 8 + 5];
+		*a = 256u * in[i * 8 + 6] + in[i * 8 + 7];
+	}
+}
+
+u32 png_convert(u8 *out, const u8 *in, const pngColorMode *mode_out, const pngColorMode *mode_in,
+		u32 w, u32 h)
+{
+	u32 i;
+	ColorTree tree;
+	u32 numpixels = (u32)w * (u32)h;
+	u32 error = 0;
+
+	if (mode_in->colortype == LCT_PALETTE && !mode_in->palette) {
+		return 107; /* error: must provide palette if input mode is palette */
+	}
+
+	if (png_color_mode_equal(mode_out, mode_in)) {
+		u32 numbytes = png_get_raw_size(w, h, mode_in);
+		png_memcpy(out, in, numbytes);
+		return 0;
+	}
+
+	if (mode_out->colortype == LCT_PALETTE) {
+		u32 palettesize = mode_out->palettesize;
+		const u8 *palette = mode_out->palette;
+		u32 palsize = (u32)1u << mode_out->bitdepth;
+		/*if the user specified output palette but did not give the values, assume
+    they want the values of the input color type (assuming that one is palette).
+    Note that we never create a new palette ourselves.*/
+		if (palettesize == 0) {
+			palettesize = mode_in->palettesize;
+			palette = mode_in->palette;
+			/*if the input was also palette with same bitdepth, then the color types are also
+      equal, so copy literally. This to preserve the exact indices that were in the PNG
+      even in case there are duplicate colors in the palette.*/
+			if (mode_in->colortype == LCT_PALETTE &&
+			    mode_in->bitdepth == mode_out->bitdepth) {
+				u32 numbytes = png_get_raw_size(w, h, mode_in);
+				png_memcpy(out, in, numbytes);
+				return 0;
+			}
+		}
+		if (palettesize < palsize)
+			palsize = palettesize;
+		color_tree_init(&tree);
+		for (i = 0; i != palsize; ++i) {
+			const u8 *p = &palette[i * 4];
+			error = color_tree_add(&tree, p[0], p[1], p[2], p[3], (u32)i);
+			if (error)
+				break;
+		}
+	}
+
+	if (!error) {
+		if (mode_in->bitdepth == 16 && mode_out->bitdepth == 16) {
+			for (i = 0; i != numpixels; ++i) {
+				u16 r = 0, g = 0, b = 0, a = 0;
+				getPixelColorRGBA16(&r, &g, &b, &a, in, i, mode_in);
+				rgba16ToPixel(out, i, mode_out, r, g, b, a);
+			}
+		} else if (mode_out->bitdepth == 8 && mode_out->colortype == LCT_RGBA) {
+			getPixelColorsRGBA8(out, numpixels, in, mode_in);
+		} else if (mode_out->bitdepth == 8 && mode_out->colortype == LCT_RGB) {
+			getPixelColorsRGB8(out, numpixels, in, mode_in);
+		} else {
+			u8 r = 0, g = 0, b = 0, a = 0;
+			for (i = 0; i != numpixels; ++i) {
+				getPixelColorRGBA8(&r, &g, &b, &a, in, i, mode_in);
+				error = rgba8ToPixel(out, i, mode_out, &tree, r, g, b, a);
+				if (error)
+					break;
+			}
+		}
+	}
+
+	if (mode_out->colortype == LCT_PALETTE) {
+		color_tree_cleanup(&tree);
+	}
+
+	return error;
+}
+
+#ifdef PNG_COMPILE_ENCODER
+
+/* Converts a single rgb color without alpha from one type to another, color bits truncated to
+their bitdepth. In case of single channel (gray or palette), only the r channel is used. Slow
+function, do not use to process all pixels of an image. Alpha channel not supported on purpose:
+this is for bKGD, supporting alpha may prevent it from finding a color in the palette, from the
+specification it looks like bKGD should ignore the alpha values of the palette since it can use
+any palette index but doesn't have an alpha channel. Idem with ignoring color key. */
+static u32 png_convert_rgb(u32 *r_out, u32 *g_out, u32 *b_out, u32 r_in, u32 g_in, u32 b_in,
+			   const pngColorMode *mode_out, const pngColorMode *mode_in)
+{
+	u32 r = 0, g = 0, b = 0;
+	u32 mul = 65535 / ((1u << mode_in->bitdepth) - 1u); /*65535, 21845, 4369, 257, 1*/
+	u32 shift = 16 - mode_out->bitdepth;
+
+	if (mode_in->colortype == LCT_GREY || mode_in->colortype == LCT_GREY_ALPHA) {
+		r = g = b = r_in * mul;
+	} else if (mode_in->colortype == LCT_RGB || mode_in->colortype == LCT_RGBA) {
+		r = r_in * mul;
+		g = g_in * mul;
+		b = b_in * mul;
+	} else if (mode_in->colortype == LCT_PALETTE) {
+		if (r_in >= mode_in->palettesize)
+			return 82;
+		r = mode_in->palette[r_in * 4 + 0] * 257u;
+		g = mode_in->palette[r_in * 4 + 1] * 257u;
+		b = mode_in->palette[r_in * 4 + 2] * 257u;
+	} else {
+		return 31;
+	}
+
+	/* now convert to output format */
+	if (mode_out->colortype == LCT_GREY || mode_out->colortype == LCT_GREY_ALPHA) {
+		*r_out = r >> shift;
+	} else if (mode_out->colortype == LCT_RGB || mode_out->colortype == LCT_RGBA) {
+		*r_out = r >> shift;
+		*g_out = g >> shift;
+		*b_out = b >> shift;
+	} else if (mode_out->colortype == LCT_PALETTE) {
+		u32 i;
+		/* a 16-bit color cannot be in the palette */
+		if ((r >> 8) != (r & 255) || (g >> 8) != (g & 255) || (b >> 8) != (b & 255))
+			return 82;
+		for (i = 0; i < mode_out->palettesize; i++) {
+			u32 j = i * 4;
+			if ((r >> 8) == mode_out->palette[j + 0] &&
+			    (g >> 8) == mode_out->palette[j + 1] &&
+			    (b >> 8) == mode_out->palette[j + 2]) {
+				*r_out = i;
+				return 0;
+			}
+		}
+		return 82;
+	} else {
+		return 31;
+	}
+
+	return 0;
+}
+
+void png_color_stats_init(pngColorStats *stats)
+{
+	/*stats*/
+	stats->colored = 0;
+	stats->key = 0;
+	stats->key_r = stats->key_g = stats->key_b = 0;
+	stats->alpha = 0;
+	stats->numcolors = 0;
+	stats->bits = 1;
+	stats->numpixels = 0;
+	/*settings*/
+	stats->allow_palette = 1;
+	stats->allow_greyscale = 1;
+}
+
+/*function used for debug purposes with C++*/
+/*void printColorStats(pngColorStats* p) {
+  std::cout << "colored: " << (int)p->colored << ", ";
+  std::cout << "key: " << (int)p->key << ", ";
+  std::cout << "key_r: " << (int)p->key_r << ", ";
+  std::cout << "key_g: " << (int)p->key_g << ", ";
+  std::cout << "key_b: " << (int)p->key_b << ", ";
+  std::cout << "alpha: " << (int)p->alpha << ", ";
+  std::cout << "numcolors: " << (int)p->numcolors << ", ";
+  std::cout << "bits: " << (int)p->bits << std::endl;
+}*/
+
+/*Returns how many bits needed to represent given value (max 8 bit)*/
+static u32 getValueRequiredBits(u8 value)
+{
+	if (value == 0 || value == 255)
+		return 1;
+	/*The scaling of 2-bit and 4-bit values uses multiples of 85 and 17*/
+	if (value % 17 == 0)
+		return value % 85 == 0 ? 2 : 4;
+	return 8;
+}
+
+/*stats must already have been inited. */
+u32 png_compute_color_stats(pngColorStats *stats, const u8 *in, u32 w, u32 h,
+			    const pngColorMode *mode_in)
+{
+	u32 i;
+	ColorTree tree;
+	u32 numpixels = (u32)w * (u32)h;
+	u32 error = 0;
+
+	/* mark things as done already if it would be impossible to have a more expensive case */
+	u32 colored_done = png_is_greyscale_type(mode_in) ? 1 : 0;
+	u32 alpha_done = png_can_have_alpha(mode_in) ? 0 : 1;
+	u32 numcolors_done = 0;
+	u32 bpp = png_get_bpp(mode_in);
+	u32 bits_done = (stats->bits == 1 && bpp == 1) ? 1 : 0;
+	u32 sixteen = 0; /* whether the input image is 16 bit */
+	u32 maxnumcolors = 257;
+	if (bpp <= 8)
+		maxnumcolors = PNG_MIN(257, stats->numcolors + (1u << bpp));
+
+	stats->numpixels += numpixels;
+
+	/*if palette not allowed, no need to compute numcolors*/
+	if (!stats->allow_palette)
+		numcolors_done = 1;
+
+	color_tree_init(&tree);
+
+	/*If the stats was already filled in from previous data, fill its palette in tree
+  and mark things as done already if we know they are the most expensive case already*/
+	if (stats->alpha)
+		alpha_done = 1;
+	if (stats->colored)
+		colored_done = 1;
+	if (stats->bits == 16)
+		numcolors_done = 1;
+	if (stats->bits >= bpp)
+		bits_done = 1;
+	if (stats->numcolors >= maxnumcolors)
+		numcolors_done = 1;
+
+	if (!numcolors_done) {
+		for (i = 0; i < stats->numcolors; i++) {
+			const u8 *color = &stats->palette[i * 4];
+			error = color_tree_add(&tree, color[0], color[1], color[2], color[3], i);
+			if (error)
+				goto cleanup;
+		}
+	}
+
+	/*Check if the 16-bit input is truly 16-bit*/
+	if (mode_in->bitdepth == 16 && !sixteen) {
+		u16 r = 0, g = 0, b = 0, a = 0;
+		for (i = 0; i != numpixels; ++i) {
+			getPixelColorRGBA16(&r, &g, &b, &a, in, i, mode_in);
+			if ((r & 255) != ((r >> 8) & 255) || (g & 255) != ((g >> 8) & 255) ||
+			    (b & 255) != ((b >> 8) & 255) ||
+			    (a & 255) != ((a >> 8) & 255)) /*first and second byte differ*/ {
+				stats->bits = 16;
+				sixteen = 1;
+				bits_done = 1;
+				numcolors_done =
+					1; /*counting colors no longer useful, palette doesn't support 16-bit*/
+				break;
+			}
+		}
+	}
+
+	if (sixteen) {
+		u16 r = 0, g = 0, b = 0, a = 0;
+
+		for (i = 0; i != numpixels; ++i) {
+			getPixelColorRGBA16(&r, &g, &b, &a, in, i, mode_in);
+
+			if (!colored_done && (r != g || r != b)) {
+				stats->colored = 1;
+				colored_done = 1;
+			}
+
+			if (!alpha_done) {
+				u32 matchkey = (r == stats->key_r && g == stats->key_g &&
+						b == stats->key_b);
+				if (a != 65535 && (a != 0 || (stats->key && !matchkey))) {
+					stats->alpha = 1;
+					stats->key = 0;
+					alpha_done = 1;
+				} else if (a == 0 && !stats->alpha && !stats->key) {
+					stats->key = 1;
+					stats->key_r = r;
+					stats->key_g = g;
+					stats->key_b = b;
+				} else if (a == 65535 && stats->key && matchkey) {
+					/* Color key cannot be used if an opaque pixel also has that RGB color. */
+					stats->alpha = 1;
+					stats->key = 0;
+					alpha_done = 1;
+				}
+			}
+			if (alpha_done && numcolors_done && colored_done && bits_done)
+				break;
+		}
+
+		if (stats->key && !stats->alpha) {
+			for (i = 0; i != numpixels; ++i) {
+				getPixelColorRGBA16(&r, &g, &b, &a, in, i, mode_in);
+				if (a != 0 && r == stats->key_r && g == stats->key_g &&
+				    b == stats->key_b) {
+					/* Color key cannot be used if an opaque pixel also has that RGB color. */
+					stats->alpha = 1;
+					stats->key = 0;
+					alpha_done = 1;
+				}
+			}
+		}
+	} else /* < 16-bit */ {
+		u8 r = 0, g = 0, b = 0, a = 0;
+		for (i = 0; i != numpixels; ++i) {
+			getPixelColorRGBA8(&r, &g, &b, &a, in, i, mode_in);
+
+			if (!bits_done && stats->bits < 8) {
+				/*only r is checked, < 8 bits is only relevant for grayscale*/
+				u32 bits = getValueRequiredBits(r);
+				if (bits > stats->bits)
+					stats->bits = bits;
+			}
+			bits_done = (stats->bits >= bpp);
+
+			if (!colored_done && (r != g || r != b)) {
+				stats->colored = 1;
+				colored_done = 1;
+				if (stats->bits < 8)
+					stats->bits =
+						8; /*PNG has no colored modes with less than 8-bit per channel*/
+			}
+
+			if (!alpha_done) {
+				u32 matchkey = (r == stats->key_r && g == stats->key_g &&
+						b == stats->key_b);
+				if (a != 255 && (a != 0 || (stats->key && !matchkey))) {
+					stats->alpha = 1;
+					stats->key = 0;
+					alpha_done = 1;
+					if (stats->bits < 8)
+						stats->bits =
+							8; /*PNG has no alphachannel modes with less than 8-bit per channel*/
+				} else if (a == 0 && !stats->alpha && !stats->key) {
+					stats->key = 1;
+					stats->key_r = r;
+					stats->key_g = g;
+					stats->key_b = b;
+				} else if (a == 255 && stats->key && matchkey) {
+					/* Color key cannot be used if an opaque pixel also has that RGB color. */
+					stats->alpha = 1;
+					stats->key = 0;
+					alpha_done = 1;
+					if (stats->bits < 8)
+						stats->bits =
+							8; /*PNG has no alphachannel modes with less than 8-bit per channel*/
+				}
+			}
+
+			if (!numcolors_done) {
+				if (!color_tree_has(&tree, r, g, b, a)) {
+					error = color_tree_add(&tree, r, g, b, a, stats->numcolors);
+					if (error)
+						goto cleanup;
+					if (stats->numcolors < 256) {
+						u8 *p = stats->palette;
+						u32 n = stats->numcolors;
+						p[n * 4 + 0] = r;
+						p[n * 4 + 1] = g;
+						p[n * 4 + 2] = b;
+						p[n * 4 + 3] = a;
+					}
+					++stats->numcolors;
+					numcolors_done = stats->numcolors >= maxnumcolors;
+				}
+			}
+
+			if (alpha_done && numcolors_done && colored_done && bits_done)
+				break;
+		}
+
+		if (stats->key && !stats->alpha) {
+			for (i = 0; i != numpixels; ++i) {
+				getPixelColorRGBA8(&r, &g, &b, &a, in, i, mode_in);
+				if (a != 0 && r == stats->key_r && g == stats->key_g &&
+				    b == stats->key_b) {
+					/* Color key cannot be used if an opaque pixel also has that RGB color. */
+					stats->alpha = 1;
+					stats->key = 0;
+					alpha_done = 1;
+					if (stats->bits < 8)
+						stats->bits =
+							8; /*PNG has no alphachannel modes with less than 8-bit per channel*/
+				}
+			}
+		}
+
+		/*make the stats's key always 16-bit for consistency - repeat each byte twice*/
+		stats->key_r += (stats->key_r << 8);
+		stats->key_g += (stats->key_g << 8);
+		stats->key_b += (stats->key_b << 8);
+	}
+
+cleanup:
+	color_tree_cleanup(&tree);
+	return error;
+}
+
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+/*Adds a single color to the color stats. The stats must already have been inited. The color must be given as 16-bit
+(with 2 bytes repeating for 8-bit and 65535 for opaque alpha channel). This function is expensive, do not call it for
+all pixels of an image but only for a few additional values. */
+static u32 png_color_stats_add(pngColorStats *stats, u32 r, u32 g, u32 b, u32 a)
+{
+	u32 error = 0;
+	u8 image[8];
+	pngColorMode mode;
+	png_color_mode_init(&mode);
+	image[0] = r >> 8;
+	image[1] = r;
+	image[2] = g >> 8;
+	image[3] = g;
+	image[4] = b >> 8;
+	image[5] = b;
+	image[6] = a >> 8;
+	image[7] = a;
+	mode.bitdepth = 16;
+	mode.colortype = LCT_RGBA;
+	error = png_compute_color_stats(stats, image, 1, 1, &mode);
+	png_color_mode_cleanup(&mode);
+	return error;
+}
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+
+/*Computes a minimal PNG color model that can contain all colors as indicated by the stats.
+The stats should be computed with png_compute_color_stats.
+mode_in is raw color profile of the image the stats were computed on, to copy palette order from when relevant.
+Minimal PNG color model means the color type and bit depth that gives smallest amount of bits in the output image,
+e.g. gray if only grayscale pixels, palette if less than 256 colors, color key if only single transparent color, ...
+This is used if auto_convert is enabled (it is by default).
+*/
+static u32 auto_choose_color(pngColorMode *mode_out, const pngColorMode *mode_in,
+			     const pngColorStats *stats)
+{
+	u32 error = 0;
+	u32 palettebits;
+	u32 i, n;
+	u32 numpixels = stats->numpixels;
+	u32 palette_ok, gray_ok;
+
+	u32 alpha = stats->alpha;
+	u32 key = stats->key;
+	u32 bits = stats->bits;
+
+	mode_out->key_defined = 0;
+
+	if (key && numpixels <= 16) {
+		alpha = 1; /*too few pixels to justify tRNS chunk overhead*/
+		key = 0;
+		if (bits < 8)
+			bits = 8; /*PNG has no alphachannel modes with less than 8-bit per channel*/
+	}
+
+	gray_ok = !stats->colored;
+	if (!stats->allow_greyscale)
+		gray_ok = 0;
+	if (!gray_ok && bits < 8)
+		bits = 8;
+
+	n = stats->numcolors;
+	palettebits = n <= 2 ? 1 : (n <= 4 ? 2 : (n <= 16 ? 4 : 8));
+	palette_ok =
+		n <= 256 && bits <= 8 && n != 0; /*n==0 means likely numcolors wasn't computed*/
+	if (numpixels < n * 2)
+		palette_ok = 0; /*don't add palette overhead if image has only a few pixels*/
+	if (gray_ok && !alpha && bits <= palettebits)
+		palette_ok = 0; /*gray is less overhead*/
+	if (!stats->allow_palette)
+		palette_ok = 0;
+
+	if (palette_ok) {
+		const u8 *p = stats->palette;
+		png_palette_clear(mode_out); /*remove potential earlier palette*/
+		for (i = 0; i != stats->numcolors; ++i) {
+			error = png_palette_add(mode_out, p[i * 4 + 0], p[i * 4 + 1], p[i * 4 + 2],
+						p[i * 4 + 3]);
+			if (error)
+				break;
+		}
+
+		mode_out->colortype = LCT_PALETTE;
+		mode_out->bitdepth = palettebits;
+
+		if (mode_in->colortype == LCT_PALETTE &&
+		    mode_in->palettesize >= mode_out->palettesize &&
+		    mode_in->bitdepth == mode_out->bitdepth) {
+			/*If input should have same palette colors, keep original to preserve its order and prevent conversion*/
+			png_color_mode_cleanup(mode_out);
+			png_color_mode_copy(mode_out, mode_in);
+		}
+	} else /*8-bit or 16-bit per channel*/ {
+		mode_out->bitdepth = bits;
+		mode_out->colortype = alpha ? (gray_ok ? LCT_GREY_ALPHA : LCT_RGBA) :
+						    (gray_ok ? LCT_GREY : LCT_RGB);
+		if (key) {
+			u32 mask = (1u << mode_out->bitdepth) -
+				   1u; /*stats always uses 16-bit, mask converts it*/
+			mode_out->key_r = stats->key_r & mask;
+			mode_out->key_g = stats->key_g & mask;
+			mode_out->key_b = stats->key_b & mask;
+			mode_out->key_defined = 1;
+		}
+	}
+
+	return error;
+}
+
+#endif /* #ifdef PNG_COMPILE_ENCODER */
+
+/*
+Paeth predictor, used by PNG filter type 4
+The parameters are of type short, but should come from u8s, the shorts
+are only needed to make the paeth calculation correct.
+*/
+static u8 paethPredictor(short a, short b, short c)
+{
+	short pa = PNG_ABS(b - c);
+	short pb = PNG_ABS(a - c);
+	short pc = PNG_ABS(a + b - c - c);
+	/* return input value associated with smallest of pa, pb, pc (with certain priority if equal) */
+	if (pb < pa) {
+		a = b;
+		pa = pb;
+	}
+	return (pc < pa) ? c : a;
+}
+
+/*shared values used by multiple Adam7 related functions*/
+
+static const u32 ADAM7_IX[7] = { 0, 4, 0, 2, 0, 1, 0 }; /*x start values*/
+static const u32 ADAM7_IY[7] = { 0, 0, 4, 0, 2, 0, 1 }; /*y start values*/
+static const u32 ADAM7_DX[7] = { 8, 8, 4, 4, 2, 2, 1 }; /*x delta values*/
+static const u32 ADAM7_DY[7] = { 8, 8, 8, 4, 4, 2, 2 }; /*y delta values*/
+
+/*
+Outputs various dimensions and positions in the image related to the Adam7 reduced images.
+passw: output containing the width of the 7 passes
+passh: output containing the height of the 7 passes
+filter_passstart: output containing the index of the start and end of each
+ reduced image with filter bytes
+padded_passstart output containing the index of the start and end of each
+ reduced image when without filter bytes but with padded scanlines
+passstart: output containing the index of the start and end of each reduced
+ image without padding between scanlines, but still padding between the images
+w, h: width and height of non-interlaced image
+bpp: bits per pixel
+"padded" is only relevant if bpp is less than 8 and a scanline or image does not
+ end at a full byte
+*/
+static void Adam7_getpassvalues(u32 passw[7], u32 passh[7], u32 filter_passstart[8],
+				u32 padded_passstart[8], u32 passstart[8], u32 w, u32 h, u32 bpp)
+{
+	/*the passstart values have 8 values: the 8th one indicates the byte after the end of the 7th (= last) pass*/
+	u32 i;
+
+	/*calculate width and height in pixels of each pass*/
+	for (i = 0; i != 7; ++i) {
+		passw[i] = (w + ADAM7_DX[i] - ADAM7_IX[i] - 1) / ADAM7_DX[i];
+		passh[i] = (h + ADAM7_DY[i] - ADAM7_IY[i] - 1) / ADAM7_DY[i];
+		if (passw[i] == 0)
+			passh[i] = 0;
+		if (passh[i] == 0)
+			passw[i] = 0;
+	}
+
+	filter_passstart[0] = padded_passstart[0] = passstart[0] = 0;
+	for (i = 0; i != 7; ++i) {
+		/*if passw[i] is 0, it's 0 bytes, not 1 (no filtertype-byte)*/
+		filter_passstart[i + 1] =
+			filter_passstart[i] +
+			((passw[i] && passh[i]) ? passh[i] * (1u + (passw[i] * bpp + 7u) / 8u) : 0);
+		/*bits padded if needed to fill full byte at end of each scanline*/
+		padded_passstart[i + 1] =
+			padded_passstart[i] + passh[i] * ((passw[i] * bpp + 7u) / 8u);
+		/*only padded at end of reduced image*/
+		passstart[i + 1] = passstart[i] + (passh[i] * passw[i] * bpp + 7u) / 8u;
+	}
+}
+
+#ifdef PNG_COMPILE_DECODER
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / PNG Decoder                                                            / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+/*read the information from the header and store it in the pngInfo. return value is error*/
+u32 png_inspect(u32 *w, u32 *h, pngState *state, const u8 *in, u32 insize)
+{
+	u32 width, height;
+	pngInfo *info = &state->info_png;
+	if (insize == 0 || in == 0) {
+		CERROR_RETURN_ERROR(state->error, 48); /*error: the given data is empty*/
+	}
+	if (insize < 33) {
+		CERROR_RETURN_ERROR(
+			state->error,
+			27); /*error: the data length is smaller than the length of a PNG header*/
+	}
+
+	/*when decoding a new PNG image, make sure all parameters created after previous decoding are reset*/
+	/* TODO: remove this. One should use a new pngState for new sessions */
+	png_info_cleanup(info);
+	png_info_init(info);
+
+	if (in[0] != 137 || in[1] != 80 || in[2] != 78 || in[3] != 71 || in[4] != 13 ||
+	    in[5] != 10 || in[6] != 26 || in[7] != 10) {
+		CERROR_RETURN_ERROR(
+			state->error,
+			28); /*error: the first 8 bytes are not the correct PNG signature*/
+	}
+	if (png_chunk_length(in + 8) != 13) {
+		CERROR_RETURN_ERROR(state->error, 94); /*error: header size must be 13 bytes*/
+	}
+	if (!png_chunk_type_equals(in + 8, "IHDR")) {
+		CERROR_RETURN_ERROR(state->error,
+				    29); /*error: it doesn't start with a IHDR chunk!*/
+	}
+
+	/*read the values given in the header*/
+	width = png_read32bitInt(&in[16]);
+	height = png_read32bitInt(&in[20]);
+	/*TODO: remove the undocumented feature that allows to give null pointers to width or height*/
+	if (w)
+		*w = width;
+	if (h)
+		*h = height;
+	info->color.bitdepth = in[24];
+	info->color.colortype = (pngColorType)in[25];
+	info->compression_method = in[26];
+	info->filter_method = in[27];
+	info->interlace_method = in[28];
+
+	/*errors returned only after the parsing so other values are still output*/
+
+	/*error: invalid image size*/
+	if (width == 0 || height == 0)
+		CERROR_RETURN_ERROR(state->error, 93);
+	/*error: invalid colortype or bitdepth combination*/
+	state->error = checkColorValidity(info->color.colortype, info->color.bitdepth);
+	if (state->error)
+		return state->error;
+	/*error: only compression method 0 is allowed in the specification*/
+	if (info->compression_method != 0)
+		CERROR_RETURN_ERROR(state->error, 32);
+	/*error: only filter method 0 is allowed in the specification*/
+	if (info->filter_method != 0)
+		CERROR_RETURN_ERROR(state->error, 33);
+	/*error: only interlace methods 0 and 1 exist in the specification*/
+	if (info->interlace_method > 1)
+		CERROR_RETURN_ERROR(state->error, 34);
+
+	if (!state->decoder.ignore_crc) {
+		u32 CRC = png_read32bitInt(&in[29]);
+		u32 checksum = png_crc32(&in[12], 17);
+		if (CRC != checksum) {
+			CERROR_RETURN_ERROR(state->error, 57); /*invalid CRC*/
+		}
+	}
+
+	return state->error;
+}
+
+static u32 unfilterScanline(u8 *recon, const u8 *scanline, const u8 *precon, u32 bytewidth,
+			    u8 filterType, u32 length)
+{
+	/*
+  For PNG filter method 0
+  unfilter a PNG image scanline by scanline. when the pixels are smaller than 1 byte,
+  the filter works byte per byte (bytewidth = 1)
+  precon is the previous unfiltered scanline, recon the result, scanline the current one
+  the incoming scanlines do NOT include the filtertype byte, that one is given in the parameter filterType instead
+  recon and scanline MAY be the same memory address! precon must be disjoint.
+  */
+
+	u32 i;
+	switch (filterType) {
+	case 0:
+		for (i = 0; i != length; ++i)
+			recon[i] = scanline[i];
+		break;
+	case 1:
+		for (i = 0; i != bytewidth; ++i)
+			recon[i] = scanline[i];
+		for (i = bytewidth; i < length; ++i)
+			recon[i] = scanline[i] + recon[i - bytewidth];
+		break;
+	case 2:
+		if (precon) {
+			for (i = 0; i != length; ++i)
+				recon[i] = scanline[i] + precon[i];
+		} else {
+			for (i = 0; i != length; ++i)
+				recon[i] = scanline[i];
+		}
+		break;
+	case 3:
+		if (precon) {
+			for (i = 0; i != bytewidth; ++i)
+				recon[i] = scanline[i] + (precon[i] >> 1u);
+			for (i = bytewidth; i < length; ++i)
+				recon[i] = scanline[i] + ((recon[i - bytewidth] + precon[i]) >> 1u);
+		} else {
+			for (i = 0; i != bytewidth; ++i)
+				recon[i] = scanline[i];
+			for (i = bytewidth; i < length; ++i)
+				recon[i] = scanline[i] + (recon[i - bytewidth] >> 1u);
+		}
+		break;
+	case 4:
+		if (precon) {
+			for (i = 0; i != bytewidth; ++i) {
+				recon[i] =
+					(scanline[i] +
+					 precon[i]); /*paethPredictor(0, precon[i], 0) is always precon[i]*/
+			}
+
+			/* Unroll independent paths of the paeth predictor. A 6x and 8x version would also be possible but that
+        adds too much code. Whether this actually speeds anything up at all depends on compiler and settings. */
+			if (bytewidth >= 4) {
+				for (; i + 3 < length; i += 4) {
+					u32 j = i - bytewidth;
+					u8 s0 = scanline[i + 0], s1 = scanline[i + 1],
+					   s2 = scanline[i + 2], s3 = scanline[i + 3];
+					u8 r0 = recon[j + 0], r1 = recon[j + 1], r2 = recon[j + 2],
+					   r3 = recon[j + 3];
+					u8 p0 = precon[i + 0], p1 = precon[i + 1],
+					   p2 = precon[i + 2], p3 = precon[i + 3];
+					u8 q0 = precon[j + 0], q1 = precon[j + 1],
+					   q2 = precon[j + 2], q3 = precon[j + 3];
+					recon[i + 0] = s0 + paethPredictor(r0, p0, q0);
+					recon[i + 1] = s1 + paethPredictor(r1, p1, q1);
+					recon[i + 2] = s2 + paethPredictor(r2, p2, q2);
+					recon[i + 3] = s3 + paethPredictor(r3, p3, q3);
+				}
+			} else if (bytewidth >= 3) {
+				for (; i + 2 < length; i += 3) {
+					u32 j = i - bytewidth;
+					u8 s0 = scanline[i + 0], s1 = scanline[i + 1],
+					   s2 = scanline[i + 2];
+					u8 r0 = recon[j + 0], r1 = recon[j + 1], r2 = recon[j + 2];
+					u8 p0 = precon[i + 0], p1 = precon[i + 1],
+					   p2 = precon[i + 2];
+					u8 q0 = precon[j + 0], q1 = precon[j + 1],
+					   q2 = precon[j + 2];
+					recon[i + 0] = s0 + paethPredictor(r0, p0, q0);
+					recon[i + 1] = s1 + paethPredictor(r1, p1, q1);
+					recon[i + 2] = s2 + paethPredictor(r2, p2, q2);
+				}
+			} else if (bytewidth >= 2) {
+				for (; i + 1 < length; i += 2) {
+					u32 j = i - bytewidth;
+					u8 s0 = scanline[i + 0], s1 = scanline[i + 1];
+					u8 r0 = recon[j + 0], r1 = recon[j + 1];
+					u8 p0 = precon[i + 0], p1 = precon[i + 1];
+					u8 q0 = precon[j + 0], q1 = precon[j + 1];
+					recon[i + 0] = s0 + paethPredictor(r0, p0, q0);
+					recon[i + 1] = s1 + paethPredictor(r1, p1, q1);
+				}
+			}
+
+			for (; i != length; ++i) {
+				recon[i] = (scanline[i] + paethPredictor(recon[i - bytewidth],
+									 precon[i],
+									 precon[i - bytewidth]));
+			}
+		} else {
+			for (i = 0; i != bytewidth; ++i) {
+				recon[i] = scanline[i];
+			}
+			for (i = bytewidth; i < length; ++i) {
+				/*paethPredictor(recon[i - bytewidth], 0, 0) is always recon[i - bytewidth]*/
+				recon[i] = (scanline[i] + recon[i - bytewidth]);
+			}
+		}
+		break;
+	default:
+		return 36; /*error: invalid filter type given*/
+	}
+	return 0;
+}
+
+static u32 unfilter(u8 *out, const u8 *in, u32 w, u32 h, u32 bpp)
+{
+	/*
+  For PNG filter method 0
+  this function unfilters a single image (e.g. without interlacing this is called once, with Adam7 seven times)
+  out must have enough bytes allocated already, in must have the scanlines + 1 filtertype byte per scanline
+  w and h are image dimensions or dimensions of reduced image, bpp is bits per pixel
+  in and out are allowed to be the same memory address (but aren't the same size since in has the extra filter bytes)
+  */
+
+	u32 y;
+	u8 *prevline = 0;
+
+	/*bytewidth is used for filtering, is 1 when bpp < 8, number of bytes per pixel otherwise*/
+	u32 bytewidth = (bpp + 7u) / 8u;
+	/*the width of a scanline in bytes, not including the filter type*/
+	u32 linebytes = png_get_raw_size_idat(w, 1, bpp) - 1u;
+
+	for (y = 0; y < h; ++y) {
+		u32 outindex = linebytes * y;
+		u32 inindex = (1 + linebytes) * y; /*the extra filterbyte added to each row*/
+		u8 filterType = in[inindex];
+
+		CERROR_TRY_RETURN(unfilterScanline(&out[outindex], &in[inindex + 1], prevline,
+						   bytewidth, filterType, linebytes));
+
+		prevline = &out[outindex];
+	}
+
+	return 0;
+}
+
+/*
+in: Adam7 interlaced image, with no padding bits between scanlines, but between
+ reduced images so that each reduced image starts at a byte.
+out: the same pixels, but re-ordered so that they're now a non-interlaced image with size w*h
+bpp: bits per pixel
+out has the following size in bits: w * h * bpp.
+in is possibly bigger due to padding bits between reduced images.
+out must be big enough AND must be 0 everywhere if bpp < 8 in the current implementation
+(because that's likely a little bit faster)
+NOTE: comments about padding bits are only relevant if bpp < 8
+*/
+static void Adam7_deinterlace(u8 *out, const u8 *in, u32 w, u32 h, u32 bpp)
+{
+	u32 passw[7], passh[7];
+	u32 filter_passstart[8], padded_passstart[8], passstart[8];
+	u32 i;
+
+	Adam7_getpassvalues(passw, passh, filter_passstart, padded_passstart, passstart, w, h, bpp);
+
+	if (bpp >= 8) {
+		for (i = 0; i != 7; ++i) {
+			u32 x, y, b;
+			u32 bytewidth = bpp / 8u;
+			for (y = 0; y < passh[i]; ++y)
+				for (x = 0; x < passw[i]; ++x) {
+					u32 pixelinstart =
+						passstart[i] + (y * passw[i] + x) * bytewidth;
+					u32 pixeloutstart =
+						((ADAM7_IY[i] + (u32)y * ADAM7_DY[i]) * (u32)w +
+						 ADAM7_IX[i] + (u32)x * ADAM7_DX[i]) *
+						bytewidth;
+					for (b = 0; b < bytewidth; ++b) {
+						out[pixeloutstart + b] = in[pixelinstart + b];
+					}
+				}
+		}
+	} else /*bpp < 8: Adam7 with pixels < 8 bit is a bit trickier: with bit pointers*/ {
+		for (i = 0; i != 7; ++i) {
+			u32 x, y, b;
+			u32 ilinebits = bpp * passw[i];
+			u32 olinebits = bpp * w;
+			u32 obp, ibp; /*bit pointers (for out and in buffer)*/
+			for (y = 0; y < passh[i]; ++y)
+				for (x = 0; x < passw[i]; ++x) {
+					ibp = (8 * passstart[i]) + (y * ilinebits + x * bpp);
+					obp = (ADAM7_IY[i] + (u32)y * ADAM7_DY[i]) * olinebits +
+					      (ADAM7_IX[i] + (u32)x * ADAM7_DX[i]) * bpp;
+					for (b = 0; b < bpp; ++b) {
+						u8 bit = readBitFromReversedStream(&ibp, in);
+						setBitOfReversedStream(&obp, out, bit);
+					}
+				}
+		}
+	}
+}
+
+static void removePaddingBits(u8 *out, const u8 *in, u32 olinebits, u32 ilinebits, u32 h)
+{
+	/*
+  After filtering there are still padding bits if scanlines have non multiple of 8 bit amounts. They need
+  to be removed (except at last scanline of (Adam7-reduced) image) before working with pure image buffers
+  for the Adam7 code, the color convert code and the output to the user.
+  in and out are allowed to be the same buffer, in may also be higher but still overlapping; in must
+  have >= ilinebits*h bits, out must have >= olinebits*h bits, olinebits must be <= ilinebits
+  also used to move bits after earlier such operations happened, e.g. in a sequence of reduced images from Adam7
+  only useful if (ilinebits - olinebits) is a value in the range 1..7
+  */
+	u32 y;
+	u32 diff = ilinebits - olinebits;
+	u32 ibp = 0, obp = 0; /*input and output bit pointers*/
+	for (y = 0; y < h; ++y) {
+		u32 x;
+		for (x = 0; x < olinebits; ++x) {
+			u8 bit = readBitFromReversedStream(&ibp, in);
+			setBitOfReversedStream(&obp, out, bit);
+		}
+		ibp += diff;
+	}
+}
+
+/*out must be buffer big enough to contain full image, and in must contain the full decompressed data from
+the IDAT chunks (with filter index bytes and possible padding bits)
+return value is error*/
+static u32 postProcessScanlines(u8 *out, u8 *in, u32 w, u32 h, const pngInfo *info_png)
+{
+	/*
+  This function converts the filtered-padded-interlaced data into pure 2D image buffer with the PNG's colortype.
+  Steps:
+  *) if no Adam7: 1) unfilter 2) remove padding bits (= possible extra bits per scanline if bpp < 8)
+  *) if adam7: 1) 7x unfilter 2) 7x remove padding bits 3) Adam7_deinterlace
+  NOTE: the in buffer will be overwritten with intermediate data!
+  */
+	u32 bpp = png_get_bpp(&info_png->color);
+	if (bpp == 0)
+		return 31; /*error: invalid colortype*/
+
+	if (info_png->interlace_method == 0) {
+		if (bpp < 8 && w * bpp != ((w * bpp + 7u) / 8u) * 8u) {
+			CERROR_TRY_RETURN(unfilter(in, in, w, h, bpp));
+			removePaddingBits(out, in, w * bpp, ((w * bpp + 7u) / 8u) * 8u, h);
+		}
+		/*we can immediately filter into the out buffer, no other steps needed*/
+		else
+			CERROR_TRY_RETURN(unfilter(out, in, w, h, bpp));
+	} else /*interlace_method is 1 (Adam7)*/ {
+		u32 passw[7], passh[7];
+		u32 filter_passstart[8], padded_passstart[8], passstart[8];
+		u32 i;
+
+		Adam7_getpassvalues(passw, passh, filter_passstart, padded_passstart, passstart, w,
+				    h, bpp);
+
+		for (i = 0; i != 7; ++i) {
+			CERROR_TRY_RETURN(unfilter(&in[padded_passstart[i]],
+						   &in[filter_passstart[i]], passw[i], passh[i],
+						   bpp));
+			/*TODO: possible efficiency improvement: if in this reduced image the bits fit nicely in 1 scanline,
+      move bytes instead of bits or move not at all*/
+			if (bpp < 8) {
+				/*remove padding bits in scanlines; after this there still may be padding
+        bits between the different reduced images: each reduced image still starts nicely at a byte*/
+				removePaddingBits(&in[passstart[i]], &in[padded_passstart[i]],
+						  passw[i] * bpp, ((passw[i] * bpp + 7u) / 8u) * 8u,
+						  passh[i]);
+			}
+		}
+
+		Adam7_deinterlace(out, in, w, h, bpp);
+	}
+
+	return 0;
+}
+
+static u32 readChunk_PLTE(pngColorMode *color, const u8 *data, u32 chunkLength)
+{
+	u32 pos = 0, i;
+	color->palettesize = chunkLength / 3u;
+	if (color->palettesize == 0 || color->palettesize > 256)
+		return 38; /*error: palette too small or big*/
+	png_color_mode_alloc_palette(color);
+	if (!color->palette && color->palettesize) {
+		color->palettesize = 0;
+		return 83; /*alloc fail*/
+	}
+
+	for (i = 0; i != color->palettesize; ++i) {
+		color->palette[4 * i + 0] = data[pos++]; /*R*/
+		color->palette[4 * i + 1] = data[pos++]; /*G*/
+		color->palette[4 * i + 2] = data[pos++]; /*B*/
+		color->palette[4 * i + 3] = 255; /*alpha*/
+	}
+
+	return 0; /* OK */
+}
+
+static u32 readChunk_tRNS(pngColorMode *color, const u8 *data, u32 chunkLength)
+{
+	u32 i;
+	if (color->colortype == LCT_PALETTE) {
+		/*error: more alpha values given than there are palette entries*/
+		if (chunkLength > color->palettesize)
+			return 39;
+
+		for (i = 0; i != chunkLength; ++i)
+			color->palette[4 * i + 3] = data[i];
+	} else if (color->colortype == LCT_GREY) {
+		/*error: this chunk must be 2 bytes for grayscale image*/
+		if (chunkLength != 2)
+			return 30;
+
+		color->key_defined = 1;
+		color->key_r = color->key_g = color->key_b = 256u * data[0] + data[1];
+	} else if (color->colortype == LCT_RGB) {
+		/*error: this chunk must be 6 bytes for RGB image*/
+		if (chunkLength != 6)
+			return 41;
+
+		color->key_defined = 1;
+		color->key_r = 256u * data[0] + data[1];
+		color->key_g = 256u * data[2] + data[3];
+		color->key_b = 256u * data[4] + data[5];
+	} else
+		return 42; /*error: tRNS chunk not allowed for other color models*/
+
+	return 0; /* OK */
+}
+
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+/*background color chunk (bKGD)*/
+static u32 readChunk_bKGD(pngInfo *info, const u8 *data, u32 chunkLength)
+{
+	if (info->color.colortype == LCT_PALETTE) {
+		/*error: this chunk must be 1 byte for indexed color image*/
+		if (chunkLength != 1)
+			return 43;
+
+		/*error: invalid palette index, or maybe this chunk appeared before PLTE*/
+		if (data[0] >= info->color.palettesize)
+			return 103;
+
+		info->background_defined = 1;
+		info->background_r = info->background_g = info->background_b = data[0];
+	} else if (info->color.colortype == LCT_GREY || info->color.colortype == LCT_GREY_ALPHA) {
+		/*error: this chunk must be 2 bytes for grayscale image*/
+		if (chunkLength != 2)
+			return 44;
+
+		/*the values are truncated to bitdepth in the PNG file*/
+		info->background_defined = 1;
+		info->background_r = info->background_g = info->background_b =
+			256u * data[0] + data[1];
+	} else if (info->color.colortype == LCT_RGB || info->color.colortype == LCT_RGBA) {
+		/*error: this chunk must be 6 bytes for grayscale image*/
+		if (chunkLength != 6)
+			return 45;
+
+		/*the values are truncated to bitdepth in the PNG file*/
+		info->background_defined = 1;
+		info->background_r = 256u * data[0] + data[1];
+		info->background_g = 256u * data[2] + data[3];
+		info->background_b = 256u * data[4] + data[5];
+	}
+
+	return 0; /* OK */
+}
+
+/*text chunk (tEXt)*/
+static u32 readChunk_tEXt(pngInfo *info, const u8 *data, u32 chunkLength)
+{
+	u32 error = 0;
+	char *key = 0, *str = 0;
+
+	while (!error) /*not really a while loop, only used to break on error*/ {
+		u32 length, string2_begin;
+
+		length = 0;
+		while (length < chunkLength && data[length] != 0)
+			++length;
+		/*even though it's not allowed by the standard, no error is thrown if
+    there's no null termination char, if the text is empty*/
+		if (length < 1 || length > 79)
+			CERROR_BREAK(error, 89); /*keyword too short or long*/
+
+		key = (char *)png_malloc(length + 1);
+		if (!key)
+			CERROR_BREAK(error, 83); /*alloc fail*/
+
+		png_memcpy(key, data, length);
+		key[length] = 0;
+
+		string2_begin = length + 1; /*skip keyword null terminator*/
+
+		length = (u32)(chunkLength < string2_begin ? 0 : chunkLength - string2_begin);
+		str = (char *)png_malloc(length + 1);
+		if (!str)
+			CERROR_BREAK(error, 83); /*alloc fail*/
+
+		png_memcpy(str, data + string2_begin, length);
+		str[length] = 0;
+
+		error = png_add_text(info, key, str);
+
+		break;
+	}
+
+	png_free(key);
+	png_free(str);
+
+	return error;
+}
+
+/*compressed text chunk (zTXt)*/
+static u32 readChunk_zTXt(pngInfo *info, const pngDecoderSettings *decoder, const u8 *data,
+			  u32 chunkLength)
+{
+	u32 error = 0;
+
+	/*copy the object to change parameters in it*/
+	pngDecompressSettings zlibsettings = decoder->zlibsettings;
+
+	u32 length, string2_begin;
+	char *key = 0;
+	u8 *str = 0;
+	u32 size = 0;
+
+	while (!error) /*not really a while loop, only used to break on error*/ {
+		for (length = 0; length < chunkLength && data[length] != 0; ++length)
+			;
+		if (length + 2 >= chunkLength)
+			CERROR_BREAK(error, 75); /*no null termination, corrupt?*/
+		if (length < 1 || length > 79)
+			CERROR_BREAK(error, 89); /*keyword too short or long*/
+
+		key = (char *)png_malloc(length + 1);
+		if (!key)
+			CERROR_BREAK(error, 83); /*alloc fail*/
+
+		png_memcpy(key, data, length);
+		key[length] = 0;
+
+		if (data[length + 1] != 0)
+			CERROR_BREAK(error, 72); /*the 0 byte indicating compression must be 0*/
+
+		string2_begin = length + 2;
+		if (string2_begin > chunkLength)
+			CERROR_BREAK(error, 75); /*no null termination, corrupt?*/
+
+		length = (u32)chunkLength - string2_begin;
+		zlibsettings.max_output_size = decoder->max_text_size;
+		/*will fail if zlib error, e.g. if length is too small*/
+		error = zlib_decompress(&str, &size, 0, &data[string2_begin], length,
+					&zlibsettings);
+		/*error: compressed text larger than  decoder->max_text_size*/
+		if (error && size > zlibsettings.max_output_size)
+			error = 112;
+		if (error)
+			break;
+		error = png_add_text_sized(info, key, (char *)str, size);
+		break;
+	}
+
+	png_free(key);
+	png_free(str);
+
+	return error;
+}
+
+/*international text chunk (iTXt)*/
+static u32 readChunk_iTXt(pngInfo *info, const pngDecoderSettings *decoder, const u8 *data,
+			  u32 chunkLength)
+{
+	u32 error = 0;
+	u32 i;
+
+	/*copy the object to change parameters in it*/
+	pngDecompressSettings zlibsettings = decoder->zlibsettings;
+
+	u32 length, begin, compressed;
+	char *key = 0, *langtag = 0, *transkey = 0;
+
+	while (!error) /*not really a while loop, only used to break on error*/ {
+		/*Quick check if the chunk length isn't too small. Even without check
+    it'd still fail with other error checks below if it's too short. This just gives a different error code.*/
+		if (chunkLength < 5)
+			CERROR_BREAK(error, 30); /*iTXt chunk too short*/
+
+		/*read the key*/
+		for (length = 0; length < chunkLength && data[length] != 0; ++length)
+			;
+		if (length + 3 >= chunkLength)
+			CERROR_BREAK(error, 75); /*no null termination char, corrupt?*/
+		if (length < 1 || length > 79)
+			CERROR_BREAK(error, 89); /*keyword too short or long*/
+
+		key = (char *)png_malloc(length + 1);
+		if (!key)
+			CERROR_BREAK(error, 83); /*alloc fail*/
+
+		png_memcpy(key, data, length);
+		key[length] = 0;
+
+		/*read the compression method*/
+		compressed = data[length + 1];
+		if (data[length + 2] != 0)
+			CERROR_BREAK(error, 72); /*the 0 byte indicating compression must be 0*/
+
+		/*even though it's not allowed by the standard, no error is thrown if
+    there's no null termination char, if the text is empty for the next 3 texts*/
+
+		/*read the langtag*/
+		begin = length + 3;
+		length = 0;
+		for (i = begin; i < chunkLength && data[i] != 0; ++i)
+			++length;
+
+		langtag = (char *)png_malloc(length + 1);
+		if (!langtag)
+			CERROR_BREAK(error, 83); /*alloc fail*/
+
+		png_memcpy(langtag, data + begin, length);
+		langtag[length] = 0;
+
+		/*read the transkey*/
+		begin += length + 1;
+		length = 0;
+		for (i = begin; i < chunkLength && data[i] != 0; ++i)
+			++length;
+
+		transkey = (char *)png_malloc(length + 1);
+		if (!transkey)
+			CERROR_BREAK(error, 83); /*alloc fail*/
+
+		png_memcpy(transkey, data + begin, length);
+		transkey[length] = 0;
+
+		/*read the actual text*/
+		begin += length + 1;
+
+		length = (u32)chunkLength < begin ? 0 : (u32)chunkLength - begin;
+
+		if (compressed) {
+			u8 *str = 0;
+			u32 size = 0;
+			zlibsettings.max_output_size = decoder->max_text_size;
+			/*will fail if zlib error, e.g. if length is too small*/
+			error = zlib_decompress(&str, &size, 0, &data[begin], length,
+						&zlibsettings);
+			/*error: compressed text larger than  decoder->max_text_size*/
+			if (error && size > zlibsettings.max_output_size)
+				error = 112;
+			if (!error)
+				error = png_add_itext_sized(info, key, langtag, transkey,
+							    (char *)str, size);
+			png_free(str);
+		} else {
+			error = png_add_itext_sized(info, key, langtag, transkey,
+						    (const char *)(data + begin), length);
+		}
+
+		break;
+	}
+
+	png_free(key);
+	png_free(langtag);
+	png_free(transkey);
+
+	return error;
+}
+
+static u32 readChunk_tIME(pngInfo *info, const u8 *data, u32 chunkLength)
+{
+	if (chunkLength != 7)
+		return 73; /*invalid tIME chunk size*/
+
+	info->time_defined = 1;
+	info->time.year = 256u * data[0] + data[1];
+	info->time.month = data[2];
+	info->time.day = data[3];
+	info->time.hour = data[4];
+	info->time.minute = data[5];
+	info->time.second = data[6];
+
+	return 0; /* OK */
+}
+
+static u32 readChunk_pHYs(pngInfo *info, const u8 *data, u32 chunkLength)
+{
+	if (chunkLength != 9)
+		return 74; /*invalid pHYs chunk size*/
+
+	info->phys_defined = 1;
+	info->phys_x = 16777216u * data[0] + 65536u * data[1] + 256u * data[2] + data[3];
+	info->phys_y = 16777216u * data[4] + 65536u * data[5] + 256u * data[6] + data[7];
+	info->phys_unit = data[8];
+
+	return 0; /* OK */
+}
+
+static u32 readChunk_gAMA(pngInfo *info, const u8 *data, u32 chunkLength)
+{
+	if (chunkLength != 4)
+		return 96; /*invalid gAMA chunk size*/
+
+	info->gama_defined = 1;
+	info->gama_gamma = 16777216u * data[0] + 65536u * data[1] + 256u * data[2] + data[3];
+
+	return 0; /* OK */
+}
+
+static u32 readChunk_cHRM(pngInfo *info, const u8 *data, u32 chunkLength)
+{
+	if (chunkLength != 32)
+		return 97; /*invalid cHRM chunk size*/
+
+	info->chrm_defined = 1;
+	info->chrm_white_x = 16777216u * data[0] + 65536u * data[1] + 256u * data[2] + data[3];
+	info->chrm_white_y = 16777216u * data[4] + 65536u * data[5] + 256u * data[6] + data[7];
+	info->chrm_red_x = 16777216u * data[8] + 65536u * data[9] + 256u * data[10] + data[11];
+	info->chrm_red_y = 16777216u * data[12] + 65536u * data[13] + 256u * data[14] + data[15];
+	info->chrm_green_x = 16777216u * data[16] + 65536u * data[17] + 256u * data[18] + data[19];
+	info->chrm_green_y = 16777216u * data[20] + 65536u * data[21] + 256u * data[22] + data[23];
+	info->chrm_blue_x = 16777216u * data[24] + 65536u * data[25] + 256u * data[26] + data[27];
+	info->chrm_blue_y = 16777216u * data[28] + 65536u * data[29] + 256u * data[30] + data[31];
+
+	return 0; /* OK */
+}
+
+static u32 readChunk_sRGB(pngInfo *info, const u8 *data, u32 chunkLength)
+{
+	if (chunkLength != 1)
+		return 98; /*invalid sRGB chunk size (this one is never ignored)*/
+
+	info->srgb_defined = 1;
+	info->srgb_intent = data[0];
+
+	return 0; /* OK */
+}
+
+static u32 readChunk_iCCP(pngInfo *info, const pngDecoderSettings *decoder, const u8 *data,
+			  u32 chunkLength)
+{
+	u32 error = 0;
+	u32 i;
+	u32 size = 0;
+	/*copy the object to change parameters in it*/
+	pngDecompressSettings zlibsettings = decoder->zlibsettings;
+
+	u32 length, string2_begin;
+
+	info->iccp_defined = 1;
+	if (info->iccp_name)
+		png_clear_icc(info);
+
+	for (length = 0; length < chunkLength && data[length] != 0; ++length)
+		;
+	if (length + 2 >= chunkLength)
+		return 75; /*no null termination, corrupt?*/
+	if (length < 1 || length > 79)
+		return 89; /*keyword too short or long*/
+
+	info->iccp_name = (char *)png_malloc(length + 1);
+	if (!info->iccp_name)
+		return 83; /*alloc fail*/
+
+	info->iccp_name[length] = 0;
+	for (i = 0; i != length; ++i)
+		info->iccp_name[i] = (char)data[i];
+
+	if (data[length + 1] != 0)
+		return 72; /*the 0 byte indicating compression must be 0*/
+
+	string2_begin = length + 2;
+	if (string2_begin > chunkLength)
+		return 75; /*no null termination, corrupt?*/
+
+	length = (u32)chunkLength - string2_begin;
+	zlibsettings.max_output_size = decoder->max_icc_size;
+	error = zlib_decompress(&info->iccp_profile, &size, 0, &data[string2_begin], length,
+				&zlibsettings);
+	/*error: ICC profile larger than  decoder->max_icc_size*/
+	if (error && size > zlibsettings.max_output_size)
+		error = 113;
+	info->iccp_profile_size = size;
+	if (!error && !info->iccp_profile_size)
+		error = 100; /*invalid ICC profile size*/
+	return error;
+}
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+
+u32 png_inspect_chunk(pngState *state, u32 pos, const u8 *in, u32 insize)
+{
+	const u8 *chunk = in + pos;
+	u32 chunkLength;
+	const u8 *data;
+	u32 unhandled = 0;
+	u32 error = 0;
+
+	if (pos + 4 > insize)
+		return 30;
+	chunkLength = png_chunk_length(chunk);
+	if (chunkLength > 2147483647)
+		return 63;
+	data = png_chunk_data_const(chunk);
+	if (data + chunkLength + 4 > in + insize)
+		return 30;
+
+	if (png_chunk_type_equals(chunk, "PLTE")) {
+		error = readChunk_PLTE(&state->info_png.color, data, chunkLength);
+	} else if (png_chunk_type_equals(chunk, "tRNS")) {
+		error = readChunk_tRNS(&state->info_png.color, data, chunkLength);
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+	} else if (png_chunk_type_equals(chunk, "bKGD")) {
+		error = readChunk_bKGD(&state->info_png, data, chunkLength);
+	} else if (png_chunk_type_equals(chunk, "tEXt")) {
+		error = readChunk_tEXt(&state->info_png, data, chunkLength);
+	} else if (png_chunk_type_equals(chunk, "zTXt")) {
+		error = readChunk_zTXt(&state->info_png, &state->decoder, data, chunkLength);
+	} else if (png_chunk_type_equals(chunk, "iTXt")) {
+		error = readChunk_iTXt(&state->info_png, &state->decoder, data, chunkLength);
+	} else if (png_chunk_type_equals(chunk, "tIME")) {
+		error = readChunk_tIME(&state->info_png, data, chunkLength);
+	} else if (png_chunk_type_equals(chunk, "pHYs")) {
+		error = readChunk_pHYs(&state->info_png, data, chunkLength);
+	} else if (png_chunk_type_equals(chunk, "gAMA")) {
+		error = readChunk_gAMA(&state->info_png, data, chunkLength);
+	} else if (png_chunk_type_equals(chunk, "cHRM")) {
+		error = readChunk_cHRM(&state->info_png, data, chunkLength);
+	} else if (png_chunk_type_equals(chunk, "sRGB")) {
+		error = readChunk_sRGB(&state->info_png, data, chunkLength);
+	} else if (png_chunk_type_equals(chunk, "iCCP")) {
+		error = readChunk_iCCP(&state->info_png, &state->decoder, data, chunkLength);
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+	} else {
+		/* unhandled chunk is ok (is not an error) */
+		unhandled = 1;
+	}
+
+	if (!error && !unhandled && !state->decoder.ignore_crc) {
+		if (png_chunk_check_crc(chunk))
+			return 57; /*invalid CRC*/
+	}
+
+	return error;
+}
+
+/*read a PNG, the result will be in the same color type as the PNG (hence "generic")*/
+static void decodeGeneric(u8 **out, u32 *w, u32 *h, pngState *state, const u8 *in, u32 insize)
+{
+	u8 IEND = 0;
+	const u8 *chunk;
+	u8 *idat; /*the data from idat chunks, zlib compressed*/
+	u32 idatsize = 0;
+	u8 *scanlines = 0;
+	u32 scanlines_size = 0, expected_size = 0;
+	u32 outsize = 0;
+
+	/*for unknown chunk order*/
+	u32 unknown = 0;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+	u32 critical_pos = 1; /*1 = after IHDR, 2 = after PLTE, 3 = after IDAT*/
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+
+	/* safe output values in case error happens */
+	*out = 0;
+	*w = *h = 0;
+
+	state->error =
+		png_inspect(w, h, state, in,
+			    insize); /*reads header and resets other parameters in state->info_png*/
+	if (state->error)
+		return;
+
+	if (png_pixel_overflow(*w, *h, &state->info_png.color, &state->info_raw)) {
+		CERROR_RETURN(state->error, 92); /*overflow possible due to amount of pixels*/
+	}
+
+	/*the input filesize is a safe upper bound for the sum of idat chunks size*/
+	idat = (u8 *)png_malloc(insize);
+	if (!idat)
+		CERROR_RETURN(state->error, 83); /*alloc fail*/
+
+	chunk = &in[33]; /*first byte of the first chunk after the header*/
+
+	/*loop through the chunks, ignoring unknown chunks and stopping at IEND chunk.
+  IDAT data is put at the start of the in buffer*/
+	while (!IEND && !state->error) {
+		u32 chunkLength;
+		const u8 *data; /*the data in the chunk*/
+
+		/*error: size of the in buffer too small to contain next chunk*/
+		if ((u32)((chunk - in) + 12) > insize || chunk < in) {
+			if (state->decoder.ignore_end)
+				break; /*other errors may still happen though*/
+			CERROR_BREAK(state->error, 30);
+		}
+
+		/*length of the data of the chunk, excluding the length bytes, chunk type and CRC bytes*/
+		chunkLength = png_chunk_length(chunk);
+		/*error: chunk length larger than the max PNG chunk size*/
+		if (chunkLength > 2147483647) {
+			if (state->decoder.ignore_end)
+				break; /*other errors may still happen though*/
+			CERROR_BREAK(state->error, 63);
+		}
+
+		if ((u32)((chunk - in) + chunkLength + 12) > insize ||
+		    (chunk + chunkLength + 12) < in) {
+			CERROR_BREAK(
+				state->error,
+				64); /*error: size of the in buffer too small to contain next chunk*/
+		}
+
+		data = png_chunk_data_const(chunk);
+
+		unknown = 0;
+
+		/*IDAT chunk, containing compressed image data*/
+		if (png_chunk_type_equals(chunk, "IDAT")) {
+			u32 newsize;
+			if (png_addofl(idatsize, chunkLength, &newsize))
+				CERROR_BREAK(state->error, 95);
+			if (newsize > insize)
+				CERROR_BREAK(state->error, 95);
+			png_memcpy(idat + idatsize, data, chunkLength);
+			idatsize += chunkLength;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+			critical_pos = 3;
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+		} else if (png_chunk_type_equals(chunk, "IEND")) {
+			/*IEND chunk*/
+			IEND = 1;
+		} else if (png_chunk_type_equals(chunk, "PLTE")) {
+			/*palette chunk (PLTE)*/
+			state->error = readChunk_PLTE(&state->info_png.color, data, chunkLength);
+			if (state->error)
+				break;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+			critical_pos = 2;
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+		} else if (png_chunk_type_equals(chunk, "tRNS")) {
+			/*palette transparency chunk (tRNS). Even though this one is an ancillary chunk , it is still compiled
+      in without 'PNG_COMPILE_ANCILLARY_CHUNKS' because it contains essential color information that
+      affects the alpha channel of pixels. */
+			state->error = readChunk_tRNS(&state->info_png.color, data, chunkLength);
+			if (state->error)
+				break;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+			/*background color chunk (bKGD)*/
+		} else if (png_chunk_type_equals(chunk, "bKGD")) {
+			state->error = readChunk_bKGD(&state->info_png, data, chunkLength);
+			if (state->error)
+				break;
+		} else if (png_chunk_type_equals(chunk, "tEXt")) {
+			/*text chunk (tEXt)*/
+			if (state->decoder.read_text_chunks) {
+				state->error = readChunk_tEXt(&state->info_png, data, chunkLength);
+				if (state->error)
+					break;
+			}
+		} else if (png_chunk_type_equals(chunk, "zTXt")) {
+			/*compressed text chunk (zTXt)*/
+			if (state->decoder.read_text_chunks) {
+				state->error = readChunk_zTXt(&state->info_png, &state->decoder,
+							      data, chunkLength);
+				if (state->error)
+					break;
+			}
+		} else if (png_chunk_type_equals(chunk, "iTXt")) {
+			/*international text chunk (iTXt)*/
+			if (state->decoder.read_text_chunks) {
+				state->error = readChunk_iTXt(&state->info_png, &state->decoder,
+							      data, chunkLength);
+				if (state->error)
+					break;
+			}
+		} else if (png_chunk_type_equals(chunk, "tIME")) {
+			state->error = readChunk_tIME(&state->info_png, data, chunkLength);
+			if (state->error)
+				break;
+		} else if (png_chunk_type_equals(chunk, "pHYs")) {
+			state->error = readChunk_pHYs(&state->info_png, data, chunkLength);
+			if (state->error)
+				break;
+		} else if (png_chunk_type_equals(chunk, "gAMA")) {
+			state->error = readChunk_gAMA(&state->info_png, data, chunkLength);
+			if (state->error)
+				break;
+		} else if (png_chunk_type_equals(chunk, "cHRM")) {
+			state->error = readChunk_cHRM(&state->info_png, data, chunkLength);
+			if (state->error)
+				break;
+		} else if (png_chunk_type_equals(chunk, "sRGB")) {
+			state->error = readChunk_sRGB(&state->info_png, data, chunkLength);
+			if (state->error)
+				break;
+		} else if (png_chunk_type_equals(chunk, "iCCP")) {
+			state->error = readChunk_iCCP(&state->info_png, &state->decoder, data,
+						      chunkLength);
+			if (state->error)
+				break;
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+		} else /*it's not an implemented chunk type, so ignore it: skip over the data*/ {
+			/*error: unknown critical chunk (5th bit of first byte of chunk type is 0)*/
+			if (!state->decoder.ignore_critical && !png_chunk_ancillary(chunk)) {
+				CERROR_BREAK(state->error, 69);
+			}
+
+			unknown = 1;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+			if (state->decoder.remember_unknown_chunks) {
+				state->error = png_chunk_append(
+					&state->info_png.unknown_chunks_data[critical_pos - 1],
+					&state->info_png.unknown_chunks_size[critical_pos - 1],
+					chunk);
+				if (state->error)
+					break;
+			}
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+		}
+
+		if (!state->decoder.ignore_crc &&
+		    !unknown) /*check CRC if wanted, only on known chunk types*/ {
+			if (png_chunk_check_crc(chunk))
+				CERROR_BREAK(state->error, 57); /*invalid CRC*/
+		}
+
+		if (!IEND)
+			chunk = png_chunk_next_const(chunk, in + insize);
+	}
+
+	if (!state->error && state->info_png.color.colortype == LCT_PALETTE &&
+	    !state->info_png.color.palette) {
+		state->error =
+			106; /* error: PNG file must have PLTE chunk if color type is palette */
+	}
+
+	if (!state->error) {
+		/*predict output size, to allocate exact size for output buffer to avoid more dynamic allocation.
+    If the decompressed size does not match the prediction, the image must be corrupt.*/
+		if (state->info_png.interlace_method == 0) {
+			u32 bpp = png_get_bpp(&state->info_png.color);
+			expected_size = png_get_raw_size_idat(*w, *h, bpp);
+		} else {
+			u32 bpp = png_get_bpp(&state->info_png.color);
+			/*Adam-7 interlaced: expected size is the sum of the 7 sub-images sizes*/
+			expected_size = 0;
+			expected_size += png_get_raw_size_idat((*w + 7) >> 3, (*h + 7) >> 3, bpp);
+			if (*w > 4)
+				expected_size +=
+					png_get_raw_size_idat((*w + 3) >> 3, (*h + 7) >> 3, bpp);
+			expected_size += png_get_raw_size_idat((*w + 3) >> 2, (*h + 3) >> 3, bpp);
+			if (*w > 2)
+				expected_size +=
+					png_get_raw_size_idat((*w + 1) >> 2, (*h + 3) >> 2, bpp);
+			expected_size += png_get_raw_size_idat((*w + 1) >> 1, (*h + 1) >> 2, bpp);
+			if (*w > 1)
+				expected_size +=
+					png_get_raw_size_idat((*w + 0) >> 1, (*h + 1) >> 1, bpp);
+			expected_size += png_get_raw_size_idat((*w + 0), (*h + 0) >> 1, bpp);
+		}
+
+		state->error = zlib_decompress(&scanlines, &scanlines_size, expected_size, idat,
+					       idatsize, &state->decoder.zlibsettings);
+	}
+	if (!state->error && scanlines_size != expected_size)
+		state->error = 91; /*decompressed size doesn't match prediction*/
+	png_free(idat);
+
+	if (!state->error) {
+		outsize = png_get_raw_size(*w, *h, &state->info_png.color);
+		*out = (u8 *)png_malloc(outsize);
+		if (!*out)
+			state->error = 83; /*alloc fail*/
+	}
+	if (!state->error) {
+		png_memset(*out, 0, outsize);
+		state->error = postProcessScanlines(*out, scanlines, *w, *h, &state->info_png);
+	}
+	png_free(scanlines);
+}
+
+u32 png_decode(u8 **out, u32 *w, u32 *h, pngState *state, const u8 *in, u32 insize)
+{
+	*out = 0;
+	decodeGeneric(out, w, h, state, in, insize);
+	if (state->error)
+		return state->error;
+	if (!state->decoder.color_convert ||
+	    png_color_mode_equal(&state->info_raw, &state->info_png.color)) {
+		/*same color type, no copying or converting of data needed*/
+		/*store the info_png color settings on the info_raw so that the info_raw still reflects what colortype
+    the raw image has to the end user*/
+		if (!state->decoder.color_convert) {
+			state->error =
+				png_color_mode_copy(&state->info_raw, &state->info_png.color);
+			if (state->error)
+				return state->error;
+		}
+	} else { /*color conversion needed*/
+		u8 *data = *out;
+		u32 outsize;
+
+		/*TODO: check if this works according to the statement in the documentation: "The converter can convert
+    from grayscale input color type, to 8-bit grayscale or grayscale with alpha"*/
+		if (!(state->info_raw.colortype == LCT_RGB ||
+		      state->info_raw.colortype == LCT_RGBA) &&
+		    !(state->info_raw.bitdepth == 8)) {
+			return 56; /*unsupported color mode conversion*/
+		}
+
+		outsize = png_get_raw_size(*w, *h, &state->info_raw);
+		*out = (u8 *)png_malloc(outsize);
+		if (!(*out)) {
+			state->error = 83; /*alloc fail*/
+		} else
+			state->error = png_convert(*out, data, &state->info_raw,
+						   &state->info_png.color, *w, *h);
+		png_free(data);
+	}
+	return state->error;
+}
+
+u32 png_decode_memory(u8 **out, u32 *w, u32 *h, const u8 *in, u32 insize, pngColorType colortype,
+		      u32 bitdepth)
+{
+	u32 error;
+	pngState state;
+	png_state_init(&state);
+	state.info_raw.colortype = colortype;
+	state.info_raw.bitdepth = bitdepth;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+	/*disable reading things that this function doesn't output*/
+	state.decoder.read_text_chunks = 0;
+	state.decoder.remember_unknown_chunks = 0;
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+	error = png_decode(out, w, h, &state, in, insize);
+	png_state_cleanup(&state);
+	return error;
+}
+
+u32 png_decode32(u8 **out, u32 *w, u32 *h, const u8 *in, u32 insize)
+{
+	return png_decode_memory(out, w, h, in, insize, LCT_RGBA, 8);
+}
+
+u32 png_decode24(u8 **out, u32 *w, u32 *h, const u8 *in, u32 insize)
+{
+	return png_decode_memory(out, w, h, in, insize, LCT_RGB, 8);
+}
+
+#ifdef PNG_COMPILE_DISK
+u32 png_decode_file(u8 **out, u32 *w, u32 *h, const char *filename, pngColorType colortype,
+		    u32 bitdepth)
+{
+	u8 *buffer = 0;
+	u32 buffersize;
+	u32 error;
+	/* safe output values in case error happens */
+	*out = 0;
+	*w = *h = 0;
+	error = png_load_file(&buffer, &buffersize, filename);
+	if (!error)
+		error = png_decode_memory(out, w, h, buffer, buffersize, colortype, bitdepth);
+	png_free(buffer);
+	return error;
+}
+
+u32 png_decode32_file(u8 **out, u32 *w, u32 *h, const char *filename)
+{
+	return png_decode_file(out, w, h, filename, LCT_RGBA, 8);
+}
+
+u32 png_decode24_file(u8 **out, u32 *w, u32 *h, const char *filename)
+{
+	return png_decode_file(out, w, h, filename, LCT_RGB, 8);
+}
+#endif /*PNG_COMPILE_DISK*/
+
+void png_decoder_settings_init(pngDecoderSettings *settings)
+{
+	settings->color_convert = 1;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+	settings->read_text_chunks = 1;
+	settings->remember_unknown_chunks = 0;
+	settings->max_text_size = 16777216;
+	settings->max_icc_size =
+		16777216; /* 16MB is much more than enough for any reasonable ICC profile */
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+	settings->ignore_crc = 0;
+	settings->ignore_critical = 0;
+	settings->ignore_end = 0;
+	png_decompress_settings_init(&settings->zlibsettings);
+}
+
+#endif /*PNG_COMPILE_DECODER*/
+
+#if defined(PNG_COMPILE_DECODER) || defined(PNG_COMPILE_ENCODER)
+
+void png_state_init(pngState *state)
+{
+#ifdef PNG_COMPILE_DECODER
+	png_decoder_settings_init(&state->decoder);
+#endif /*PNG_COMPILE_DECODER*/
+#ifdef PNG_COMPILE_ENCODER
+	png_encoder_settings_init(&state->encoder);
+#endif /*PNG_COMPILE_ENCODER*/
+	png_color_mode_init(&state->info_raw);
+	png_info_init(&state->info_png);
+	state->error = 1;
+}
+
+void png_state_cleanup(pngState *state)
+{
+	png_color_mode_cleanup(&state->info_raw);
+	png_info_cleanup(&state->info_png);
+}
+
+void png_state_copy(pngState *dest, const pngState *source)
+{
+	png_state_cleanup(dest);
+	*dest = *source;
+	png_color_mode_init(&dest->info_raw);
+	png_info_init(&dest->info_png);
+	dest->error = png_color_mode_copy(&dest->info_raw, &source->info_raw);
+	if (dest->error)
+		return;
+	dest->error = png_info_copy(&dest->info_png, &source->info_png);
+	if (dest->error)
+		return;
+}
+
+#endif /* defined(PNG_COMPILE_DECODER) || defined(PNG_COMPILE_ENCODER) */
+
+#ifdef PNG_COMPILE_ENCODER
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* / PNG Encoder                                                            / */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+static u32 writeSignature(ucvector *out)
+{
+	u32 pos = out->size;
+	const u8 signature[] = { 137, 80, 78, 71, 13, 10, 26, 10 };
+	/*8 bytes PNG signature, aka the magic bytes*/
+	if (!ucvector_resize(out, out->size + 8))
+		return 83; /*alloc fail*/
+	png_memcpy(out->data + pos, signature, 8);
+	return 0;
+}
+
+static u32 addChunk_IHDR(ucvector *out, u32 w, u32 h, pngColorType colortype, u32 bitdepth,
+			 u32 interlace_method)
+{
+	u8 *chunk, *data;
+	CERROR_TRY_RETURN(png_chunk_init(&chunk, out, 13, "IHDR"));
+	data = chunk + 8;
+
+	png_set32bitInt(data + 0, w); /*width*/
+	png_set32bitInt(data + 4, h); /*height*/
+	data[8] = (u8)bitdepth; /*bit depth*/
+	data[9] = (u8)colortype; /*color type*/
+	data[10] = 0; /*compression method*/
+	data[11] = 0; /*filter method*/
+	data[12] = interlace_method; /*interlace method*/
+
+	png_chunk_generate_crc(chunk);
+	return 0;
+}
+
+/* only adds the chunk if needed (there is a key or palette with alpha) */
+static u32 addChunk_PLTE(ucvector *out, const pngColorMode *info)
+{
+	u8 *chunk;
+	u32 i, j = 8;
+
+	CERROR_TRY_RETURN(png_chunk_init(&chunk, out, info->palettesize * 3, "PLTE"));
+
+	for (i = 0; i != info->palettesize; ++i) {
+		/*add all channels except alpha channel*/
+		chunk[j++] = info->palette[i * 4 + 0];
+		chunk[j++] = info->palette[i * 4 + 1];
+		chunk[j++] = info->palette[i * 4 + 2];
+	}
+
+	png_chunk_generate_crc(chunk);
+	return 0;
+}
+
+static u32 addChunk_tRNS(ucvector *out, const pngColorMode *info)
+{
+	u8 *chunk = 0;
+
+	if (info->colortype == LCT_PALETTE) {
+		u32 i, amount = info->palettesize;
+		/*the tail of palette values that all have 255 as alpha, does not have to be encoded*/
+		for (i = info->palettesize; i != 0; --i) {
+			if (info->palette[4 * (i - 1) + 3] != 255)
+				break;
+			--amount;
+		}
+		if (amount) {
+			CERROR_TRY_RETURN(png_chunk_init(&chunk, out, amount, "tRNS"));
+			/*add the alpha channel values from the palette*/
+			for (i = 0; i != amount; ++i)
+				chunk[8 + i] = info->palette[4 * i + 3];
+		}
+	} else if (info->colortype == LCT_GREY) {
+		if (info->key_defined) {
+			CERROR_TRY_RETURN(png_chunk_init(&chunk, out, 2, "tRNS"));
+			chunk[8] = (u8)(info->key_r >> 8);
+			chunk[9] = (u8)(info->key_r & 255);
+		}
+	} else if (info->colortype == LCT_RGB) {
+		if (info->key_defined) {
+			CERROR_TRY_RETURN(png_chunk_init(&chunk, out, 6, "tRNS"));
+			chunk[8] = (u8)(info->key_r >> 8);
+			chunk[9] = (u8)(info->key_r & 255);
+			chunk[10] = (u8)(info->key_g >> 8);
+			chunk[11] = (u8)(info->key_g & 255);
+			chunk[12] = (u8)(info->key_b >> 8);
+			chunk[13] = (u8)(info->key_b & 255);
+		}
+	}
+
+	if (chunk)
+		png_chunk_generate_crc(chunk);
+	return 0;
+}
+
+static u32 addChunk_IDAT(ucvector *out, const u8 *data, u32 datasize,
+			 pngCompressSettings *zlibsettings)
+{
+	u32 error = 0;
+	u8 *zlib = 0;
+	u32 zlibsize = 0;
+
+	error = zlib_compress(&zlib, &zlibsize, data, datasize, zlibsettings);
+	if (!error) {
+		error = png_chunk_createv(out, zlibsize, "IDAT", zlib);
+	}
+	png_free(zlib);
+	return error;
+}
+
+static u32 addChunk_IEND(ucvector *out)
+{
+	return png_chunk_createv(out, 0, "IEND", 0);
+}
+
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+
+static u32 addChunk_tEXt(ucvector *out, const char *keyword, const char *textstring)
+{
+	u8 *chunk = 0;
+	u32 keysize = png_strlen(keyword), textsize = png_strlen(textstring);
+	u32 size = keysize + 1 + textsize;
+	if (keysize < 1 || keysize > 79)
+		return 89; /*error: invalid keyword size*/
+	CERROR_TRY_RETURN(png_chunk_init(&chunk, out, size, "tEXt"));
+	png_memcpy(chunk + 8, keyword, keysize);
+	chunk[8 + keysize] = 0; /*null termination char*/
+	png_memcpy(chunk + 9 + keysize, textstring, textsize);
+	png_chunk_generate_crc(chunk);
+	return 0;
+}
+
+static u32 addChunk_zTXt(ucvector *out, const char *keyword, const char *textstring,
+			 pngCompressSettings *zlibsettings)
+{
+	u32 error = 0;
+	u8 *chunk = 0;
+	u8 *compressed = 0;
+	u32 compressedsize = 0;
+	u32 textsize = png_strlen(textstring);
+	u32 keysize = png_strlen(keyword);
+	if (keysize < 1 || keysize > 79)
+		return 89; /*error: invalid keyword size*/
+
+	error = zlib_compress(&compressed, &compressedsize, (const u8 *)textstring, textsize,
+			      zlibsettings);
+	if (!error) {
+		u32 size = keysize + 2 + compressedsize;
+		error = png_chunk_init(&chunk, out, size, "zTXt");
+	}
+	if (!error) {
+		png_memcpy(chunk + 8, keyword, keysize);
+		chunk[8 + keysize] = 0; /*null termination char*/
+		chunk[9 + keysize] = 0; /*compression method: 0*/
+		png_memcpy(chunk + 10 + keysize, compressed, compressedsize);
+		png_chunk_generate_crc(chunk);
+	}
+
+	png_free(compressed);
+	return error;
+}
+
+static u32 addChunk_iTXt(ucvector *out, u32 compress, const char *keyword, const char *langtag,
+			 const char *transkey, const char *textstring,
+			 pngCompressSettings *zlibsettings)
+{
+	u32 error = 0;
+	u8 *chunk = 0;
+	u8 *compressed = 0;
+	u32 compressedsize = 0;
+	u32 textsize = png_strlen(textstring);
+	u32 keysize = png_strlen(keyword), langsize = png_strlen(langtag),
+	    transsize = png_strlen(transkey);
+
+	if (keysize < 1 || keysize > 79)
+		return 89; /*error: invalid keyword size*/
+
+	if (compress) {
+		error = zlib_compress(&compressed, &compressedsize, (const u8 *)textstring,
+				      textsize, zlibsettings);
+	}
+	if (!error) {
+		u32 size = keysize + 3 + langsize + 1 + transsize + 1 +
+			   (compress ? compressedsize : textsize);
+		error = png_chunk_init(&chunk, out, size, "iTXt");
+	}
+	if (!error) {
+		u32 pos = 8;
+		png_memcpy(chunk + pos, keyword, keysize);
+		pos += keysize;
+		chunk[pos++] = 0; /*null termination char*/
+		chunk[pos++] = (compress ? 1 : 0); /*compression flag*/
+		chunk[pos++] = 0; /*compression method: 0*/
+		png_memcpy(chunk + pos, langtag, langsize);
+		pos += langsize;
+		chunk[pos++] = 0; /*null termination char*/
+		png_memcpy(chunk + pos, transkey, transsize);
+		pos += transsize;
+		chunk[pos++] = 0; /*null termination char*/
+		if (compress) {
+			png_memcpy(chunk + pos, compressed, compressedsize);
+		} else {
+			png_memcpy(chunk + pos, textstring, textsize);
+		}
+		png_chunk_generate_crc(chunk);
+	}
+
+	png_free(compressed);
+	return error;
+}
+
+static u32 addChunk_bKGD(ucvector *out, const pngInfo *info)
+{
+	u8 *chunk = 0;
+	if (info->color.colortype == LCT_GREY || info->color.colortype == LCT_GREY_ALPHA) {
+		CERROR_TRY_RETURN(png_chunk_init(&chunk, out, 2, "bKGD"));
+		chunk[8] = (u8)(info->background_r >> 8);
+		chunk[9] = (u8)(info->background_r & 255);
+	} else if (info->color.colortype == LCT_RGB || info->color.colortype == LCT_RGBA) {
+		CERROR_TRY_RETURN(png_chunk_init(&chunk, out, 6, "bKGD"));
+		chunk[8] = (u8)(info->background_r >> 8);
+		chunk[9] = (u8)(info->background_r & 255);
+		chunk[10] = (u8)(info->background_g >> 8);
+		chunk[11] = (u8)(info->background_g & 255);
+		chunk[12] = (u8)(info->background_b >> 8);
+		chunk[13] = (u8)(info->background_b & 255);
+	} else if (info->color.colortype == LCT_PALETTE) {
+		CERROR_TRY_RETURN(png_chunk_init(&chunk, out, 1, "bKGD"));
+		chunk[8] = (u8)(info->background_r & 255); /*palette index*/
+	}
+	if (chunk)
+		png_chunk_generate_crc(chunk);
+	return 0;
+}
+
+static u32 addChunk_tIME(ucvector *out, const pngTime *time)
+{
+	u8 *chunk;
+	CERROR_TRY_RETURN(png_chunk_init(&chunk, out, 7, "tIME"));
+	chunk[8] = (u8)(time->year >> 8);
+	chunk[9] = (u8)(time->year & 255);
+	chunk[10] = (u8)time->month;
+	chunk[11] = (u8)time->day;
+	chunk[12] = (u8)time->hour;
+	chunk[13] = (u8)time->minute;
+	chunk[14] = (u8)time->second;
+	png_chunk_generate_crc(chunk);
+	return 0;
+}
+
+static u32 addChunk_pHYs(ucvector *out, const pngInfo *info)
+{
+	u8 *chunk;
+	CERROR_TRY_RETURN(png_chunk_init(&chunk, out, 9, "pHYs"));
+	png_set32bitInt(chunk + 8, info->phys_x);
+	png_set32bitInt(chunk + 12, info->phys_y);
+	chunk[16] = info->phys_unit;
+	png_chunk_generate_crc(chunk);
+	return 0;
+}
+
+static u32 addChunk_gAMA(ucvector *out, const pngInfo *info)
+{
+	u8 *chunk;
+	CERROR_TRY_RETURN(png_chunk_init(&chunk, out, 4, "gAMA"));
+	png_set32bitInt(chunk + 8, info->gama_gamma);
+	png_chunk_generate_crc(chunk);
+	return 0;
+}
+
+static u32 addChunk_cHRM(ucvector *out, const pngInfo *info)
+{
+	u8 *chunk;
+	CERROR_TRY_RETURN(png_chunk_init(&chunk, out, 32, "cHRM"));
+	png_set32bitInt(chunk + 8, info->chrm_white_x);
+	png_set32bitInt(chunk + 12, info->chrm_white_y);
+	png_set32bitInt(chunk + 16, info->chrm_red_x);
+	png_set32bitInt(chunk + 20, info->chrm_red_y);
+	png_set32bitInt(chunk + 24, info->chrm_green_x);
+	png_set32bitInt(chunk + 28, info->chrm_green_y);
+	png_set32bitInt(chunk + 32, info->chrm_blue_x);
+	png_set32bitInt(chunk + 36, info->chrm_blue_y);
+	png_chunk_generate_crc(chunk);
+	return 0;
+}
+
+static u32 addChunk_sRGB(ucvector *out, const pngInfo *info)
+{
+	u8 data = info->srgb_intent;
+	return png_chunk_createv(out, 1, "sRGB", &data);
+}
+
+static u32 addChunk_iCCP(ucvector *out, const pngInfo *info, pngCompressSettings *zlibsettings)
+{
+	u32 error = 0;
+	u8 *chunk = 0;
+	u8 *compressed = 0;
+	u32 compressedsize = 0;
+	u32 keysize = png_strlen(info->iccp_name);
+
+	if (keysize < 1 || keysize > 79)
+		return 89; /*error: invalid keyword size*/
+	error = zlib_compress(&compressed, &compressedsize, info->iccp_profile,
+			      info->iccp_profile_size, zlibsettings);
+	if (!error) {
+		u32 size = keysize + 2 + compressedsize;
+		error = png_chunk_init(&chunk, out, size, "iCCP");
+	}
+	if (!error) {
+		png_memcpy(chunk + 8, info->iccp_name, keysize);
+		chunk[8 + keysize] = 0; /*null termination char*/
+		chunk[9 + keysize] = 0; /*compression method: 0*/
+		png_memcpy(chunk + 10 + keysize, compressed, compressedsize);
+		png_chunk_generate_crc(chunk);
+	}
+
+	png_free(compressed);
+	return error;
+}
+
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+
+static void filterScanline(u8 *out, const u8 *scanline, const u8 *prevline, u32 length,
+			   u32 bytewidth, u8 filterType)
+{
+	u32 i;
+	switch (filterType) {
+	case 0: /*None*/
+		for (i = 0; i != length; ++i)
+			out[i] = scanline[i];
+		break;
+	case 1: /*Sub*/
+		for (i = 0; i != bytewidth; ++i)
+			out[i] = scanline[i];
+		for (i = bytewidth; i < length; ++i)
+			out[i] = scanline[i] - scanline[i - bytewidth];
+		break;
+	case 2: /*Up*/
+		if (prevline) {
+			for (i = 0; i != length; ++i)
+				out[i] = scanline[i] - prevline[i];
+		} else {
+			for (i = 0; i != length; ++i)
+				out[i] = scanline[i];
+		}
+		break;
+	case 3: /*Average*/
+		if (prevline) {
+			for (i = 0; i != bytewidth; ++i)
+				out[i] = scanline[i] - (prevline[i] >> 1);
+			for (i = bytewidth; i < length; ++i)
+				out[i] = scanline[i] -
+					 ((scanline[i - bytewidth] + prevline[i]) >> 1);
+		} else {
+			for (i = 0; i != bytewidth; ++i)
+				out[i] = scanline[i];
+			for (i = bytewidth; i < length; ++i)
+				out[i] = scanline[i] - (scanline[i - bytewidth] >> 1);
+		}
+		break;
+	case 4: /*Paeth*/
+		if (prevline) {
+			/*paethPredictor(0, prevline[i], 0) is always prevline[i]*/
+			for (i = 0; i != bytewidth; ++i)
+				out[i] = (scanline[i] - prevline[i]);
+			for (i = bytewidth; i < length; ++i) {
+				out[i] = (scanline[i] - paethPredictor(scanline[i - bytewidth],
+								       prevline[i],
+								       prevline[i - bytewidth]));
+			}
+		} else {
+			for (i = 0; i != bytewidth; ++i)
+				out[i] = scanline[i];
+			/*paethPredictor(scanline[i - bytewidth], 0, 0) is always scanline[i - bytewidth]*/
+			for (i = bytewidth; i < length; ++i)
+				out[i] = (scanline[i] - scanline[i - bytewidth]);
+		}
+		break;
+	default:
+		return; /*invalid filter type given*/
+	}
+}
+
+/* integer binary logarithm, max return value is 31 */
+static u32 ilog2(u32 i)
+{
+	u32 result = 0;
+	if (i >= 65536) {
+		result += 16;
+		i >>= 16;
+	}
+	if (i >= 256) {
+		result += 8;
+		i >>= 8;
+	}
+	if (i >= 16) {
+		result += 4;
+		i >>= 4;
+	}
+	if (i >= 4) {
+		result += 2;
+		i >>= 2;
+	}
+	if (i >= 2) {
+		result += 1; /*i >>= 1;*/
+	}
+	return result;
+}
+
+/* integer approximation for i * log2(i), helper function for LFS_ENTROPY */
+static u32 ilog2i(u32 i)
+{
+	u32 l;
+	if (i == 0)
+		return 0;
+	l = ilog2(i);
+	/* approximate i*log2(i): l is integer logarithm, ((i - (1u << l)) << 1u)
+  linearly approximates the missing fractional part multiplied by i */
+	return i * l + ((i - (1u << l)) << 1u);
+}
+
+static u32 filter(u8 *out, const u8 *in, u32 w, u32 h, const pngColorMode *color,
+		  const pngEncoderSettings *settings)
+{
+	/*
+  For PNG filter method 0
+  out must be a buffer with as size: h + (w * h * bpp + 7u) / 8u, because there are
+  the scanlines with 1 extra byte per scanline
+  */
+
+	u32 bpp = png_get_bpp(color);
+	/*the width of a scanline in bytes, not including the filter type*/
+	u32 linebytes = png_get_raw_size_idat(w, 1, bpp) - 1u;
+
+	/*bytewidth is used for filtering, is 1 when bpp < 8, number of bytes per pixel otherwise*/
+	u32 bytewidth = (bpp + 7u) / 8u;
+	const u8 *prevline = 0;
+	u32 x, y;
+	u32 error = 0;
+	pngFilterStrategy strategy = settings->filter_strategy;
+
+	/*
+  There is a heuristic called the minimum sum of absolute differences heuristic, suggested by the PNG standard:
+   *  If the image type is Palette, or the bit depth is smaller than 8, then do not filter the image (i.e.
+      use fixed filtering, with the filter None).
+   * (The other case) If the image type is Grayscale or RGB (with or without Alpha), and the bit depth is
+     not smaller than 8, then use adaptive filtering heuristic as follows: independently for each row, apply
+     all five filters and select the filter that produces the smallest sum of absolute values per row.
+  This heuristic is used if filter strategy is LFS_MINSUM and filter_palette_zero is true.
+
+  If filter_palette_zero is true and filter_strategy is not LFS_MINSUM, the above heuristic is followed,
+  but for "the other case", whatever strategy filter_strategy is set to instead of the minimum sum
+  heuristic is used.
+  */
+	if (settings->filter_palette_zero &&
+	    (color->colortype == LCT_PALETTE || color->bitdepth < 8))
+		strategy = LFS_ZERO;
+
+	if (bpp == 0)
+		return 31; /*error: invalid color type*/
+
+	if (strategy >= LFS_ZERO && strategy <= LFS_FOUR) {
+		u8 type = (u8)strategy;
+		for (y = 0; y != h; ++y) {
+			u32 outindex =
+				(1 + linebytes) * y; /*the extra filterbyte added to each row*/
+			u32 inindex = linebytes * y;
+			out[outindex] = type; /*filter type byte*/
+			filterScanline(&out[outindex + 1], &in[inindex], prevline, linebytes,
+				       bytewidth, type);
+			prevline = &in[inindex];
+		}
+	} else if (strategy == LFS_MINSUM) {
+		/*adaptive filtering*/
+		u8 *attempt[5]; /*five filtering attempts, one for each filter type*/
+		u32 smallest = 0;
+		u8 type, bestType = 0;
+
+		for (type = 0; type != 5; ++type) {
+			attempt[type] = (u8 *)png_malloc(linebytes);
+			if (!attempt[type])
+				error = 83; /*alloc fail*/
+		}
+
+		if (!error) {
+			for (y = 0; y != h; ++y) {
+				/*try the 5 filter types*/
+				for (type = 0; type != 5; ++type) {
+					u32 sum = 0;
+					filterScanline(attempt[type], &in[y * linebytes], prevline,
+						       linebytes, bytewidth, type);
+
+					/*calculate the sum of the result*/
+					if (type == 0) {
+						for (x = 0; x != linebytes; ++x)
+							sum += (u8)(attempt[type][x]);
+					} else {
+						for (x = 0; x != linebytes; ++x) {
+							/*For differences, each byte should be treated as signed, values above 127 are negative
+              (converted to signed char). Filtertype 0 isn't a difference though, so use u32 there.
+              This means filtertype 0 is almost never chosen, but that is justified.*/
+							u8 s = attempt[type][x];
+							sum += s < 128 ? s : (255U - s);
+						}
+					}
+
+					/*check if this is smallest sum (or if type == 0 it's the first case so always store the values)*/
+					if (type == 0 || sum < smallest) {
+						bestType = type;
+						smallest = sum;
+					}
+				}
+
+				prevline = &in[y * linebytes];
+
+				/*now fill the out values*/
+				out[y * (linebytes + 1)] =
+					bestType; /*the first byte of a scanline will be the filter type*/
+				for (x = 0; x != linebytes; ++x)
+					out[y * (linebytes + 1) + 1 + x] = attempt[bestType][x];
+			}
+		}
+
+		for (type = 0; type != 5; ++type)
+			png_free(attempt[type]);
+	} else if (strategy == LFS_ENTROPY) {
+		u8 *attempt[5]; /*five filtering attempts, one for each filter type*/
+		u32 bestSum = 0;
+		u32 type, bestType = 0;
+		u32 count[256];
+
+		for (type = 0; type != 5; ++type) {
+			attempt[type] = (u8 *)png_malloc(linebytes);
+			if (!attempt[type])
+				error = 83; /*alloc fail*/
+		}
+
+		if (!error) {
+			for (y = 0; y != h; ++y) {
+				/*try the 5 filter types*/
+				for (type = 0; type != 5; ++type) {
+					u32 sum = 0;
+					filterScanline(attempt[type], &in[y * linebytes], prevline,
+						       linebytes, bytewidth, type);
+					png_memset(count, 0, 256 * sizeof(*count));
+					for (x = 0; x != linebytes; ++x)
+						++count[attempt[type][x]];
+					++count[type]; /*the filter type itself is part of the scanline*/
+					for (x = 0; x != 256; ++x) {
+						sum += ilog2i(count[x]);
+					}
+					/*check if this is smallest sum (or if type == 0 it's the first case so always store the values)*/
+					if (type == 0 || sum > bestSum) {
+						bestType = type;
+						bestSum = sum;
+					}
+				}
+
+				prevline = &in[y * linebytes];
+
+				/*now fill the out values*/
+				out[y * (linebytes + 1)] =
+					bestType; /*the first byte of a scanline will be the filter type*/
+				for (x = 0; x != linebytes; ++x)
+					out[y * (linebytes + 1) + 1 + x] = attempt[bestType][x];
+			}
+		}
+
+		for (type = 0; type != 5; ++type)
+			png_free(attempt[type]);
+	} else if (strategy == LFS_PREDEFINED) {
+		for (y = 0; y != h; ++y) {
+			u32 outindex =
+				(1 + linebytes) * y; /*the extra filterbyte added to each row*/
+			u32 inindex = linebytes * y;
+			u8 type = settings->predefined_filters[y];
+			out[outindex] = type; /*filter type byte*/
+			filterScanline(&out[outindex + 1], &in[inindex], prevline, linebytes,
+				       bytewidth, type);
+			prevline = &in[inindex];
+		}
+	} else if (strategy == LFS_BRUTE_FORCE) {
+		/*brute force filter chooser.
+    deflate the scanline after every filter attempt to see which one deflates best.
+    This is very slow and gives only slightly smaller, sometimes even larger, result*/
+		u32 size[5];
+		u8 *attempt[5]; /*five filtering attempts, one for each filter type*/
+		u32 smallest = 0;
+		u32 type = 0, bestType = 0;
+		u8 *dummy;
+		pngCompressSettings zlibsettings;
+		png_memcpy(&zlibsettings, &settings->zlibsettings, sizeof(pngCompressSettings));
+		/*use fixed tree on the attempts so that the tree is not adapted to the filtertype on purpose,
+    to simulate the true case where the tree is the same for the whole image. Sometimes it gives
+    better result with dynamic tree anyway. Using the fixed tree sometimes gives worse, but in rare
+    cases better compression. It does make this a bit less slow, so it's worth doing this.*/
+		zlibsettings.btype = 1;
+		/*a custom encoder likely doesn't read the btype setting and is optimized for complete PNG
+    images only, so disable it*/
+		zlibsettings.custom_zlib = 0;
+		zlibsettings.custom_deflate = 0;
+		for (type = 0; type != 5; ++type) {
+			attempt[type] = (u8 *)png_malloc(linebytes);
+			if (!attempt[type])
+				error = 83; /*alloc fail*/
+		}
+		if (!error) {
+			for (y = 0; y != h; ++y) /*try the 5 filter types*/ {
+				for (type = 0; type != 5; ++type) {
+					u32 testsize = (u32)linebytes;
+					/*if(testsize > 8) testsize /= 8;*/ /*it already works good enough by testing a part of the row*/
+
+					filterScanline(attempt[type], &in[y * linebytes], prevline,
+						       linebytes, bytewidth, type);
+					size[type] = 0;
+					dummy = 0;
+					zlib_compress(&dummy, &size[type], attempt[type], testsize,
+						      &zlibsettings);
+					png_free(dummy);
+					/*check if this is smallest size (or if type == 0 it's the first case so always store the values)*/
+					if (type == 0 || size[type] < smallest) {
+						bestType = type;
+						smallest = size[type];
+					}
+				}
+				prevline = &in[y * linebytes];
+				out[y * (linebytes + 1)] =
+					bestType; /*the first byte of a scanline will be the filter type*/
+				for (x = 0; x != linebytes; ++x)
+					out[y * (linebytes + 1) + 1 + x] = attempt[bestType][x];
+			}
+		}
+		for (type = 0; type != 5; ++type)
+			png_free(attempt[type]);
+	} else
+		return 88; /* unknown filter strategy */
+
+	return error;
+}
+
+static void addPaddingBits(u8 *out, const u8 *in, u32 olinebits, u32 ilinebits, u32 h)
+{
+	/*The opposite of the removePaddingBits function
+  olinebits must be >= ilinebits*/
+	u32 y;
+	u32 diff = olinebits - ilinebits;
+	u32 obp = 0, ibp = 0; /*bit pointers*/
+	for (y = 0; y != h; ++y) {
+		u32 x;
+		for (x = 0; x < ilinebits; ++x) {
+			u8 bit = readBitFromReversedStream(&ibp, in);
+			setBitOfReversedStream(&obp, out, bit);
+		}
+		/*obp += diff; --> no, fill in some value in the padding bits too, to avoid
+    "Use of uninitialised value of size ###" warning from valgrind*/
+		for (x = 0; x != diff; ++x)
+			setBitOfReversedStream(&obp, out, 0);
+	}
+}
+
+/*
+in: non-interlaced image with size w*h
+out: the same pixels, but re-ordered according to PNG's Adam7 interlacing, with
+ no padding bits between scanlines, but between reduced images so that each
+ reduced image starts at a byte.
+bpp: bits per pixel
+there are no padding bits, not between scanlines, not between reduced images
+in has the following size in bits: w * h * bpp.
+out is possibly bigger due to padding bits between reduced images
+NOTE: comments about padding bits are only relevant if bpp < 8
+*/
+static void Adam7_interlace(u8 *out, const u8 *in, u32 w, u32 h, u32 bpp)
+{
+	u32 passw[7], passh[7];
+	u32 filter_passstart[8], padded_passstart[8], passstart[8];
+	u32 i;
+
+	Adam7_getpassvalues(passw, passh, filter_passstart, padded_passstart, passstart, w, h, bpp);
+
+	if (bpp >= 8) {
+		for (i = 0; i != 7; ++i) {
+			u32 x, y, b;
+			u32 bytewidth = bpp / 8u;
+			for (y = 0; y < passh[i]; ++y)
+				for (x = 0; x < passw[i]; ++x) {
+					u32 pixelinstart = ((ADAM7_IY[i] + y * ADAM7_DY[i]) * w +
+							    ADAM7_IX[i] + x * ADAM7_DX[i]) *
+							   bytewidth;
+					u32 pixeloutstart =
+						passstart[i] + (y * passw[i] + x) * bytewidth;
+					for (b = 0; b < bytewidth; ++b) {
+						out[pixeloutstart + b] = in[pixelinstart + b];
+					}
+				}
+		}
+	} else /*bpp < 8: Adam7 with pixels < 8 bit is a bit trickier: with bit pointers*/ {
+		for (i = 0; i != 7; ++i) {
+			u32 x, y, b;
+			u32 ilinebits = bpp * passw[i];
+			u32 olinebits = bpp * w;
+			u32 obp, ibp; /*bit pointers (for out and in buffer)*/
+			for (y = 0; y < passh[i]; ++y)
+				for (x = 0; x < passw[i]; ++x) {
+					ibp = (ADAM7_IY[i] + y * ADAM7_DY[i]) * olinebits +
+					      (ADAM7_IX[i] + x * ADAM7_DX[i]) * bpp;
+					obp = (8 * passstart[i]) + (y * ilinebits + x * bpp);
+					for (b = 0; b < bpp; ++b) {
+						u8 bit = readBitFromReversedStream(&ibp, in);
+						setBitOfReversedStream(&obp, out, bit);
+					}
+				}
+		}
+	}
+}
+
+/*out must be buffer big enough to contain uncompressed IDAT chunk data, and in must contain the full image.
+return value is error**/
+static u32 preProcessScanlines(u8 **out, u32 *outsize, const u8 *in, u32 w, u32 h,
+			       const pngInfo *info_png, const pngEncoderSettings *settings)
+{
+	/*
+  This function converts the pure 2D image with the PNG's colortype, into filtered-padded-interlaced data. Steps:
+  *) if no Adam7: 1) add padding bits (= possible extra bits per scanline if bpp < 8) 2) filter
+  *) if adam7: 1) Adam7_interlace 2) 7x add padding bits 3) 7x filter
+  */
+	u32 bpp = png_get_bpp(&info_png->color);
+	u32 error = 0;
+
+	if (info_png->interlace_method == 0) {
+		*outsize =
+			h +
+			(h *
+			 ((w * bpp + 7u) /
+			  8u)); /*image size plus an extra byte per scanline + possible padding bits*/
+		*out = (u8 *)png_malloc(*outsize);
+		if (!(*out) && (*outsize))
+			error = 83; /*alloc fail*/
+
+		if (!error) {
+			/*non multiple of 8 bits per scanline, padding bits needed per scanline*/
+			if (bpp < 8 && w * bpp != ((w * bpp + 7u) / 8u) * 8u) {
+				u8 *padded = (u8 *)png_malloc(h * ((w * bpp + 7u) / 8u));
+				if (!padded)
+					error = 83; /*alloc fail*/
+				if (!error) {
+					addPaddingBits(padded, in, ((w * bpp + 7u) / 8u) * 8u,
+						       w * bpp, h);
+					error = filter(*out, padded, w, h, &info_png->color,
+						       settings);
+				}
+				png_free(padded);
+			} else {
+				/*we can immediately filter into the out buffer, no other steps needed*/
+				error = filter(*out, in, w, h, &info_png->color, settings);
+			}
+		}
+	} else /*interlace_method is 1 (Adam7)*/ {
+		u32 passw[7], passh[7];
+		u32 filter_passstart[8], padded_passstart[8], passstart[8];
+		u8 *adam7;
+
+		Adam7_getpassvalues(passw, passh, filter_passstart, padded_passstart, passstart, w,
+				    h, bpp);
+
+		*outsize = filter_passstart
+			[7]; /*image size plus an extra byte per scanline + possible padding bits*/
+		*out = (u8 *)png_malloc(*outsize);
+		if (!(*out))
+			error = 83; /*alloc fail*/
+
+		adam7 = (u8 *)png_malloc(passstart[7]);
+		if (!adam7 && passstart[7])
+			error = 83; /*alloc fail*/
+
+		if (!error) {
+			u32 i;
+
+			Adam7_interlace(adam7, in, w, h, bpp);
+			for (i = 0; i != 7; ++i) {
+				if (bpp < 8) {
+					u8 *padded = (u8 *)png_malloc(padded_passstart[i + 1] -
+								      padded_passstart[i]);
+					if (!padded)
+						ERROR_BREAK(83); /*alloc fail*/
+					addPaddingBits(padded, &adam7[passstart[i]],
+						       ((passw[i] * bpp + 7u) / 8u) * 8u,
+						       passw[i] * bpp, passh[i]);
+					error = filter(&(*out)[filter_passstart[i]], padded,
+						       passw[i], passh[i], &info_png->color,
+						       settings);
+					png_free(padded);
+				} else {
+					error = filter(&(*out)[filter_passstart[i]],
+						       &adam7[padded_passstart[i]], passw[i],
+						       passh[i], &info_png->color, settings);
+				}
+
+				if (error)
+					break;
+			}
+		}
+
+		png_free(adam7);
+	}
+
+	return error;
+}
+
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+static u32 addUnknownChunks(ucvector *out, u8 *data, u32 datasize)
+{
+	u8 *inchunk = data;
+	while ((u32)(inchunk - data) < datasize) {
+		CERROR_TRY_RETURN(png_chunk_append(&out->data, &out->size, inchunk));
+		out->allocsize = out->size; /*fix the allocsize again*/
+		inchunk = png_chunk_next(inchunk, data + datasize);
+	}
+	return 0;
+}
+
+static u32 isGrayICCProfile(const u8 *profile, u32 size)
+{
+	/*
+  It is a gray profile if bytes 16-19 are "GRAY", rgb profile if bytes 16-19
+  are "RGB ". We do not perform any full parsing of the ICC profile here, other
+  than check those 4 bytes to grayscale profile. Other than that, validity of
+  the profile is not checked. This is needed only because the PNG specification
+  requires using a non-gray color model if there is an ICC profile with "RGB "
+  (sadly limiting compression opportunities if the input data is grayscale RGB
+  data), and requires using a gray color model if it is "GRAY".
+  */
+	if (size < 20)
+		return 0;
+	return profile[16] == 'G' && profile[17] == 'R' && profile[18] == 'A' && profile[19] == 'Y';
+}
+
+static u32 isRGBICCProfile(const u8 *profile, u32 size)
+{
+	/* See comment in isGrayICCProfile*/
+	if (size < 20)
+		return 0;
+	return profile[16] == 'R' && profile[17] == 'G' && profile[18] == 'B' && profile[19] == ' ';
+}
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+
+u32 png_encode(u8 **out, u32 *outsize, const u8 *image, u32 w, u32 h, pngState *state)
+{
+	u8 *data = 0; /*uncompressed version of the IDAT chunk data*/
+	u32 datasize = 0;
+	ucvector outv = ucvector_init(NULL, 0);
+	pngInfo info;
+	const pngInfo *info_png = &state->info_png;
+
+	png_info_init(&info);
+
+	/*provide some proper output values if error will happen*/
+	*out = 0;
+	*outsize = 0;
+	state->error = 0;
+
+	/*check input values validity*/
+	if ((info_png->color.colortype == LCT_PALETTE || state->encoder.force_palette) &&
+	    (info_png->color.palettesize == 0 || info_png->color.palettesize > 256)) {
+		state->error = 68; /*invalid palette size, it is only allowed to be 1-256*/
+		goto cleanup;
+	}
+	if (state->encoder.zlibsettings.btype > 2) {
+		state->error = 61; /*error: invalid btype*/
+		goto cleanup;
+	}
+	if (info_png->interlace_method > 1) {
+		state->error = 71; /*error: invalid interlace mode*/
+		goto cleanup;
+	}
+	state->error = checkColorValidity(info_png->color.colortype, info_png->color.bitdepth);
+	if (state->error)
+		goto cleanup; /*error: invalid color type given*/
+	state->error = checkColorValidity(state->info_raw.colortype, state->info_raw.bitdepth);
+	if (state->error)
+		goto cleanup; /*error: invalid color type given*/
+
+	/* color convert and compute scanline filter types */
+	png_info_copy(&info, &state->info_png);
+	if (state->encoder.auto_convert) {
+		pngColorStats stats;
+		png_color_stats_init(&stats);
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+		if (info_png->iccp_defined &&
+		    isGrayICCProfile(info_png->iccp_profile, info_png->iccp_profile_size)) {
+			/*the PNG specification does not allow to use palette with a GRAY ICC profile, even
+      if the palette has only gray colors, so disallow it.*/
+			stats.allow_palette = 0;
+		}
+		if (info_png->iccp_defined &&
+		    isRGBICCProfile(info_png->iccp_profile, info_png->iccp_profile_size)) {
+			/*the PNG specification does not allow to use grayscale color with RGB ICC profile, so disallow gray.*/
+			stats.allow_greyscale = 0;
+		}
+#endif /* PNG_COMPILE_ANCILLARY_CHUNKS */
+		state->error = png_compute_color_stats(&stats, image, w, h, &state->info_raw);
+		if (state->error)
+			goto cleanup;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+		if (info_png->background_defined) {
+			/*the background chunk's color must be taken into account as well*/
+			u32 r = 0, g = 0, b = 0;
+			pngColorMode mode16 = png_color_mode_make(LCT_RGB, 16);
+			png_convert_rgb(&r, &g, &b, info_png->background_r, info_png->background_g,
+					info_png->background_b, &mode16, &info_png->color);
+			state->error = png_color_stats_add(&stats, r, g, b, 65535);
+			if (state->error)
+				goto cleanup;
+		}
+#endif /* PNG_COMPILE_ANCILLARY_CHUNKS */
+		state->error = auto_choose_color(&info.color, &state->info_raw, &stats);
+		if (state->error)
+			goto cleanup;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+		/*also convert the background chunk*/
+		if (info_png->background_defined) {
+			if (png_convert_rgb(&info.background_r, &info.background_g,
+					    &info.background_b, info_png->background_r,
+					    info_png->background_g, info_png->background_b,
+					    &info.color, &info_png->color)) {
+				state->error = 104;
+				goto cleanup;
+			}
+		}
+#endif /* PNG_COMPILE_ANCILLARY_CHUNKS */
+	}
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+	if (info_png->iccp_defined) {
+		u32 gray_icc =
+			isGrayICCProfile(info_png->iccp_profile, info_png->iccp_profile_size);
+		u32 rgb_icc = isRGBICCProfile(info_png->iccp_profile, info_png->iccp_profile_size);
+		u32 gray_png =
+			info.color.colortype == LCT_GREY || info.color.colortype == LCT_GREY_ALPHA;
+		if (!gray_icc && !rgb_icc) {
+			state->error = 100; /* Disallowed profile color type for PNG */
+			goto cleanup;
+		}
+		if (gray_icc != gray_png) {
+			/*Not allowed to use RGB/RGBA/palette with GRAY ICC profile or vice versa,
+      or in case of auto_convert, it wasn't possible to find appropriate model*/
+			state->error = state->encoder.auto_convert ? 102 : 101;
+			goto cleanup;
+		}
+	}
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+	if (!png_color_mode_equal(&state->info_raw, &info.color)) {
+		u8 *converted;
+		u32 size = ((u32)w * (u32)h * (u32)png_get_bpp(&info.color) + 7u) / 8u;
+
+		converted = (u8 *)png_malloc(size);
+		if (!converted && size)
+			state->error = 83; /*alloc fail*/
+		if (!state->error) {
+			state->error =
+				png_convert(converted, image, &info.color, &state->info_raw, w, h);
+		}
+		if (!state->error) {
+			state->error = preProcessScanlines(&data, &datasize, converted, w, h, &info,
+							   &state->encoder);
+		}
+		png_free(converted);
+		if (state->error)
+			goto cleanup;
+	} else {
+		state->error =
+			preProcessScanlines(&data, &datasize, image, w, h, &info, &state->encoder);
+		if (state->error)
+			goto cleanup;
+	}
+
+	/* output all PNG chunks */ {
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+		u32 i;
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+		/*write signature and chunks*/
+		state->error = writeSignature(&outv);
+		if (state->error)
+			goto cleanup;
+		/*IHDR*/
+		state->error = addChunk_IHDR(&outv, w, h, info.color.colortype, info.color.bitdepth,
+					     info.interlace_method);
+		if (state->error)
+			goto cleanup;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+		/*unknown chunks between IHDR and PLTE*/
+		if (info.unknown_chunks_data[0]) {
+			state->error = addUnknownChunks(&outv, info.unknown_chunks_data[0],
+							info.unknown_chunks_size[0]);
+			if (state->error)
+				goto cleanup;
+		}
+		/*color profile chunks must come before PLTE */
+		if (info.iccp_defined) {
+			state->error = addChunk_iCCP(&outv, &info, &state->encoder.zlibsettings);
+			if (state->error)
+				goto cleanup;
+		}
+		if (info.srgb_defined) {
+			state->error = addChunk_sRGB(&outv, &info);
+			if (state->error)
+				goto cleanup;
+		}
+		if (info.gama_defined) {
+			state->error = addChunk_gAMA(&outv, &info);
+			if (state->error)
+				goto cleanup;
+		}
+		if (info.chrm_defined) {
+			state->error = addChunk_cHRM(&outv, &info);
+			if (state->error)
+				goto cleanup;
+		}
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+		/*PLTE*/
+		if (info.color.colortype == LCT_PALETTE) {
+			state->error = addChunk_PLTE(&outv, &info.color);
+			if (state->error)
+				goto cleanup;
+		}
+		if (state->encoder.force_palette &&
+		    (info.color.colortype == LCT_RGB || info.color.colortype == LCT_RGBA)) {
+			/*force_palette means: write suggested palette for truecolor in PLTE chunk*/
+			state->error = addChunk_PLTE(&outv, &info.color);
+			if (state->error)
+				goto cleanup;
+		}
+		/*tRNS (this will only add if when necessary) */
+		state->error = addChunk_tRNS(&outv, &info.color);
+		if (state->error)
+			goto cleanup;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+		/*bKGD (must come between PLTE and the IDAt chunks*/
+		if (info.background_defined) {
+			state->error = addChunk_bKGD(&outv, &info);
+			if (state->error)
+				goto cleanup;
+		}
+		/*pHYs (must come before the IDAT chunks)*/
+		if (info.phys_defined) {
+			state->error = addChunk_pHYs(&outv, &info);
+			if (state->error)
+				goto cleanup;
+		}
+
+		/*unknown chunks between PLTE and IDAT*/
+		if (info.unknown_chunks_data[1]) {
+			state->error = addUnknownChunks(&outv, info.unknown_chunks_data[1],
+							info.unknown_chunks_size[1]);
+			if (state->error)
+				goto cleanup;
+		}
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+		/*IDAT (multiple IDAT chunks must be consecutive)*/
+		state->error = addChunk_IDAT(&outv, data, datasize, &state->encoder.zlibsettings);
+		if (state->error)
+			goto cleanup;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+		/*tIME*/
+		if (info.time_defined) {
+			state->error = addChunk_tIME(&outv, &info.time);
+			if (state->error)
+				goto cleanup;
+		}
+		/*tEXt and/or zTXt*/
+		for (i = 0; i != info.text_num; ++i) {
+			if (png_strlen(info.text_keys[i]) > 79) {
+				state->error = 66; /*text chunk too large*/
+				goto cleanup;
+			}
+			if (png_strlen(info.text_keys[i]) < 1) {
+				state->error = 67; /*text chunk too small*/
+				goto cleanup;
+			}
+			if (state->encoder.text_compression) {
+				state->error = addChunk_zTXt(&outv, info.text_keys[i],
+							     info.text_strings[i],
+							     &state->encoder.zlibsettings);
+				if (state->error)
+					goto cleanup;
+			} else {
+				state->error = addChunk_tEXt(&outv, info.text_keys[i],
+							     info.text_strings[i]);
+				if (state->error)
+					goto cleanup;
+			}
+		}
+		/*png version id in text chunk*/
+		if (state->encoder.add_id) {
+			u32 already_added_id_text = 0;
+			for (i = 0; i != info.text_num; ++i) {
+				const char *k = info.text_keys[i];
+				/* Could use strcmp, but we're not calling or reimplementing this C library function for this use only */
+				if (k[0] == 'L' && k[1] == 'o' && k[2] == 'd' && k[3] == 'e' &&
+				    k[4] == 'P' && k[5] == 'N' && k[6] == 'G' && k[7] == '\0') {
+					already_added_id_text = 1;
+					break;
+				}
+			}
+			if (already_added_id_text == 0) {
+				state->error = addChunk_tEXt(
+					&outv, "png",
+					PNG_VERSION_STRING); /*it's shorter as tEXt than as zTXt chunk*/
+				if (state->error)
+					goto cleanup;
+			}
+		}
+		/*iTXt*/
+		for (i = 0; i != info.itext_num; ++i) {
+			if (png_strlen(info.itext_keys[i]) > 79) {
+				state->error = 66; /*text chunk too large*/
+				goto cleanup;
+			}
+			if (png_strlen(info.itext_keys[i]) < 1) {
+				state->error = 67; /*text chunk too small*/
+				goto cleanup;
+			}
+			state->error = addChunk_iTXt(&outv, state->encoder.text_compression,
+						     info.itext_keys[i], info.itext_langtags[i],
+						     info.itext_transkeys[i], info.itext_strings[i],
+						     &state->encoder.zlibsettings);
+			if (state->error)
+				goto cleanup;
+		}
+
+		/*unknown chunks between IDAT and IEND*/
+		if (info.unknown_chunks_data[2]) {
+			state->error = addUnknownChunks(&outv, info.unknown_chunks_data[2],
+							info.unknown_chunks_size[2]);
+			if (state->error)
+				goto cleanup;
+		}
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+		state->error = addChunk_IEND(&outv);
+		if (state->error)
+			goto cleanup;
+	}
+
+cleanup:
+	png_info_cleanup(&info);
+	png_free(data);
+
+	/*instead of cleaning the vector up, give it to the output*/
+	*out = outv.data;
+	*outsize = outv.size;
+
+	return state->error;
+}
+
+u32 png_encode_memory(u8 **out, u32 *outsize, const u8 *image, u32 w, u32 h, pngColorType colortype,
+		      u32 bitdepth)
+{
+	u32 error;
+	pngState state;
+	png_state_init(&state);
+	state.info_raw.colortype = colortype;
+	state.info_raw.bitdepth = bitdepth;
+	state.info_png.color.colortype = colortype;
+	state.info_png.color.bitdepth = bitdepth;
+	png_encode(out, outsize, image, w, h, &state);
+	error = state.error;
+	png_state_cleanup(&state);
+	return error;
+}
+
+u32 png_encode32(u8 **out, u32 *outsize, const u8 *image, u32 w, u32 h)
+{
+	return png_encode_memory(out, outsize, image, w, h, LCT_RGBA, 8);
+}
+
+u32 png_encode24(u8 **out, u32 *outsize, const u8 *image, u32 w, u32 h)
+{
+	return png_encode_memory(out, outsize, image, w, h, LCT_RGB, 8);
+}
+
+#ifdef PNG_COMPILE_DISK
+u32 png_encode_file(const char *filename, const u8 *image, u32 w, u32 h, pngColorType colortype,
+		    u32 bitdepth)
+{
+	u8 *buffer;
+	u32 buffersize;
+	u32 error = png_encode_memory(&buffer, &buffersize, image, w, h, colortype, bitdepth);
+	if (!error)
+		error = png_save_file(buffer, buffersize, filename);
+	png_free(buffer);
+	return error;
+}
+
+u32 png_encode32_file(const char *filename, const u8 *image, u32 w, u32 h)
+{
+	return png_encode_file(filename, image, w, h, LCT_RGBA, 8);
+}
+
+u32 png_encode24_file(const char *filename, const u8 *image, u32 w, u32 h)
+{
+	return png_encode_file(filename, image, w, h, LCT_RGB, 8);
+}
+#endif /*PNG_COMPILE_DISK*/
+
+void png_encoder_settings_init(pngEncoderSettings *settings)
+{
+	png_compress_settings_init(&settings->zlibsettings);
+	settings->filter_palette_zero = 1;
+	settings->filter_strategy = LFS_MINSUM;
+	settings->auto_convert = 1;
+	settings->force_palette = 0;
+	settings->predefined_filters = 0;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+	settings->add_id = 0;
+	settings->text_compression = 1;
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+}
+
+#endif /*PNG_COMPILE_ENCODER*/
+#endif /*PNG_COMPILE_PNG*/
+
+#ifdef PNG_COMPILE_ERROR_TEXT
+/*
+This returns the description of a numerical error code in English. This is also
+the documentation of all the error codes.
+*/
+const char *png_error_text(u32 code)
+{
+	switch (code) {
+	case 0:
+		return "no error, everything went ok";
+	case 1:
+		return "nothing done yet"; /*the Encoder/Decoder has done nothing yet, error checking makes no sense yet*/
+	case 10:
+		return "end of input memory reached without huffman end code"; /*while huffman decoding*/
+	case 11:
+		return "error in code tree made it jump outside of huffman tree"; /*while huffman decoding*/
+	case 13:
+		return "problem while processing dynamic deflate block";
+	case 14:
+		return "problem while processing dynamic deflate block";
+	case 15:
+		return "problem while processing dynamic deflate block";
+	/*this error could happen if there are only 0 or 1 symbols present in the huffman code:*/
+	case 16:
+		return "invalid code while processing dynamic deflate block";
+	case 17:
+		return "end of out buffer memory reached while inflating";
+	case 18:
+		return "invalid distance code while inflating";
+	case 19:
+		return "end of out buffer memory reached while inflating";
+	case 20:
+		return "invalid deflate block BTYPE encountered while decoding";
+	case 21:
+		return "NLEN is not ones complement of LEN in a deflate block";
+
+	/*end of out buffer memory reached while inflating:
+    This can happen if the inflated deflate data is longer than the amount of bytes required to fill up
+    all the pixels of the image, given the color depth and image dimensions. Something that doesn't
+    happen in a normal, well encoded, PNG image.*/
+	case 22:
+		return "end of out buffer memory reached while inflating";
+	case 23:
+		return "end of in buffer memory reached while inflating";
+	case 24:
+		return "invalid FCHECK in zlib header";
+	case 25:
+		return "invalid compression method in zlib header";
+	case 26:
+		return "FDICT encountered in zlib header while it's not used for PNG";
+	case 27:
+		return "PNG file is smaller than a PNG header";
+	/*Checks the magic file header, the first 8 bytes of the PNG file*/
+	case 28:
+		return "incorrect PNG signature, it's no PNG or corrupted";
+	case 29:
+		return "first chunk is not the header chunk";
+	case 30:
+		return "chunk length too large, chunk broken off at end of file";
+	case 31:
+		return "illegal PNG color type or bpp";
+	case 32:
+		return "illegal PNG compression method";
+	case 33:
+		return "illegal PNG filter method";
+	case 34:
+		return "illegal PNG interlace method";
+	case 35:
+		return "chunk length of a chunk is too large or the chunk too small";
+	case 36:
+		return "illegal PNG filter type encountered";
+	case 37:
+		return "illegal bit depth for this color type given";
+	case 38:
+		return "the palette is too small or too big"; /*0, or more than 256 colors*/
+	case 39:
+		return "tRNS chunk before PLTE or has more entries than palette size";
+	case 40:
+		return "tRNS chunk has wrong size for grayscale image";
+	case 41:
+		return "tRNS chunk has wrong size for RGB image";
+	case 42:
+		return "tRNS chunk appeared while it was not allowed for this color type";
+	case 43:
+		return "bKGD chunk has wrong size for palette image";
+	case 44:
+		return "bKGD chunk has wrong size for grayscale image";
+	case 45:
+		return "bKGD chunk has wrong size for RGB image";
+	case 48:
+		return "empty input buffer given to decoder. Maybe caused by non-existing file?";
+	case 49:
+		return "jumped past memory while generating dynamic huffman tree";
+	case 50:
+		return "jumped past memory while generating dynamic huffman tree";
+	case 51:
+		return "jumped past memory while inflating huffman block";
+	case 52:
+		return "jumped past memory while inflating";
+	case 53:
+		return "size of zlib data too small";
+	case 54:
+		return "repeat symbol in tree while there was no value symbol yet";
+	/*jumped past tree while generating huffman tree, this could be when the
+    tree will have more leaves than symbols after generating it out of the
+    given lengths. They call this an oversubscribed dynamic bit lengths tree in zlib.*/
+	case 55:
+		return "jumped past tree while generating huffman tree";
+	case 56:
+		return "given output image colortype or bitdepth not supported for color conversion";
+	case 57:
+		return "invalid CRC encountered (checking CRC can be disabled)";
+	case 58:
+		return "invalid ADLER32 encountered (checking ADLER32 can be disabled)";
+	case 59:
+		return "requested color conversion not supported";
+	case 60:
+		return "invalid window size given in the settings of the encoder (must be 0-32768)";
+	case 61:
+		return "invalid BTYPE given in the settings of the encoder (only 0, 1 and 2 are allowed)";
+	/*png leaves the choice of RGB to grayscale conversion formula to the user.*/
+	case 62:
+		return "conversion from color to grayscale not supported";
+	/*(2^31-1)*/
+	case 63:
+		return "length of a chunk too long, max allowed for PNG is 2147483647 bytes per chunk";
+	/*this would result in the inability of a deflated block to ever contain an end code. It must be at least 1.*/
+	case 64:
+		return "the length of the END symbol 256 in the Huffman tree is 0";
+	case 66:
+		return "the length of a text chunk keyword given to the encoder is longer than the maximum of 79 bytes";
+	case 67:
+		return "the length of a text chunk keyword given to the encoder is smaller than the minimum of 1 byte";
+	case 68:
+		return "tried to encode a PLTE chunk with a palette that has less than 1 or more than 256 colors";
+	case 69:
+		return "unknown chunk type with 'critical' flag encountered by the decoder";
+	case 71:
+		return "invalid interlace mode given to encoder (must be 0 or 1)";
+	case 72:
+		return "while decoding, invalid compression method encountering in zTXt or iTXt chunk (it must be 0)";
+	case 73:
+		return "invalid tIME chunk size";
+	case 74:
+		return "invalid pHYs chunk size";
+	/*length could be wrong, or data chopped off*/
+	case 75:
+		return "no null termination char found while decoding text chunk";
+	case 76:
+		return "iTXt chunk too short to contain required bytes";
+	case 77:
+		return "integer overflow in buffer size";
+	case 78:
+		return "failed to open file for reading"; /*file doesn't exist or couldn't be opened for reading*/
+	case 79:
+		return "failed to open file for writing";
+	case 80:
+		return "tried creating a tree of 0 symbols";
+	case 81:
+		return "lazy matching at pos 0 is impossible";
+	case 82:
+		return "color conversion to palette requested while a color isn't in palette, or index out of bounds";
+	case 83:
+		return "memory allocation failed";
+	case 84:
+		return "given image too small to contain all pixels to be encoded";
+	case 86:
+		return "impossible offset in lz77 encoding (internal bug)";
+	case 87:
+		return "must provide custom zlib function pointer if PNG_COMPILE_ZLIB is not defined";
+	case 88:
+		return "invalid filter strategy given for pngEncoderSettings.filter_strategy";
+	case 89:
+		return "text chunk keyword too short or long: must have size 1-79";
+	/*the windowsize in the pngCompressSettings. Requiring POT(==> & instead of %) makes encoding 12% faster.*/
+	case 90:
+		return "windowsize must be a power of two";
+	case 91:
+		return "invalid decompressed idat size";
+	case 92:
+		return "integer overflow due to too many pixels";
+	case 93:
+		return "zero width or height is invalid";
+	case 94:
+		return "header chunk must have a size of 13 bytes";
+	case 95:
+		return "integer overflow with combined idat chunk size";
+	case 96:
+		return "invalid gAMA chunk size";
+	case 97:
+		return "invalid cHRM chunk size";
+	case 98:
+		return "invalid sRGB chunk size";
+	case 99:
+		return "invalid sRGB rendering intent";
+	case 100:
+		return "invalid ICC profile color type, the PNG specification only allows RGB or GRAY";
+	case 101:
+		return "PNG specification does not allow RGB ICC profile on gray color types and vice versa";
+	case 102:
+		return "not allowed to set grayscale ICC profile with colored pixels by PNG specification";
+	case 103:
+		return "invalid palette index in bKGD chunk. Maybe it came before PLTE chunk?";
+	case 104:
+		return "invalid bKGD color while encoding (e.g. palette index out of range)";
+	case 105:
+		return "integer overflow of bitsize";
+	case 106:
+		return "PNG file must have PLTE chunk if color type is palette";
+	case 107:
+		return "color convert from palette mode requested without setting the palette data in it";
+	case 108:
+		return "tried to add more than 256 values to a palette";
+	/*this limit can be configured in pngDecompressSettings*/
+	case 109:
+		return "tried to decompress zlib or deflate data larger than desired max_output_size";
+	case 110:
+		return "custom zlib or inflate decompression failed";
+	case 111:
+		return "custom zlib or deflate compression failed";
+	/*max text size limit can be configured in pngDecoderSettings. This error prevents
+    unreasonable memory consumption when decoding due to impossibly large text sizes.*/
+	case 112:
+		return "compressed text unreasonably large";
+	/*max ICC size limit can be configured in pngDecoderSettings. This error prevents
+    unreasonable memory consumption when decoding due to impossibly large ICC profile*/
+	case 113:
+		return "ICC profile unreasonably large";
+	default:
+		return "unknown error code";
+	}
+}
+#endif /*PNG_COMPILE_ERROR_TEXT*/
+
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* // C++ Wrapper                                                          // */
+/* ////////////////////////////////////////////////////////////////////////// */
+/* ////////////////////////////////////////////////////////////////////////// */
+
+#ifdef PNG_COMPILE_CPP
+namespace png
+{
+#ifdef PNG_COMPILE_DISK
+u32 load_file(std::vector<u8> &buffer, const std::string &filename)
+{
+	long size = png_filesize(filename.c_str());
+	if (size < 0)
+		return 78;
+	buffer.resize((u32)size);
+	return size == 0 ? 0 : png_buffer_file(&buffer[0], (u32)size, filename.c_str());
+}
+
+/*write given buffer to the file, overwriting the file, it doesn't append to it.*/
+u32 save_file(const std::vector<u8> &buffer, const std::string &filename)
+{
+	return png_save_file(buffer.empty() ? 0 : &buffer[0], buffer.size(), filename.c_str());
+}
+#endif /* PNG_COMPILE_DISK */
+
+#ifdef PNG_COMPILE_ZLIB
+#ifdef PNG_COMPILE_DECODER
+u32 decompress(std::vector<u8> &out, const u8 *in, u32 insize,
+	       const pngDecompressSettings &settings)
+{
+	u8 *buffer = 0;
+	u32 buffersize = 0;
+	u32 error = zlib_decompress(&buffer, &buffersize, 0, in, insize, &settings);
+	if (buffer) {
+		out.insert(out.end(), &buffer[0], &buffer[buffersize]);
+		png_free(buffer);
+	}
+	return error;
+}
+
+u32 decompress(std::vector<u8> &out, const std::vector<u8> &in,
+	       const pngDecompressSettings &settings)
+{
+	return decompress(out, in.empty() ? 0 : &in[0], in.size(), settings);
+}
+#endif /* PNG_COMPILE_DECODER */
+
+#ifdef PNG_COMPILE_ENCODER
+u32 compress(std::vector<u8> &out, const u8 *in, u32 insize, const pngCompressSettings &settings)
+{
+	u8 *buffer = 0;
+	u32 buffersize = 0;
+	u32 error = zlib_compress(&buffer, &buffersize, in, insize, &settings);
+	if (buffer) {
+		out.insert(out.end(), &buffer[0], &buffer[buffersize]);
+		png_free(buffer);
+	}
+	return error;
+}
+
+u32 compress(std::vector<u8> &out, const std::vector<u8> &in, const pngCompressSettings &settings)
+{
+	return compress(out, in.empty() ? 0 : &in[0], in.size(), settings);
+}
+#endif /* PNG_COMPILE_ENCODER */
+#endif /* PNG_COMPILE_ZLIB */
+
+#ifdef PNG_COMPILE_PNG
+
+State::State()
+{
+	png_state_init(this);
+}
+
+State::State(const State &other)
+{
+	png_state_init(this);
+	png_state_copy(this, &other);
+}
+
+State::~State()
+{
+	png_state_cleanup(this);
+}
+
+State &State::operator=(const State &other)
+{
+	png_state_copy(this, &other);
+	return *this;
+}
+
+#ifdef PNG_COMPILE_DECODER
+
+u32 decode(std::vector<u8> &out, u32 &w, u32 &h, const u8 *in, u32 insize, pngColorType colortype,
+	   u32 bitdepth)
+{
+	u8 *buffer = 0;
+	u32 error = png_decode_memory(&buffer, &w, &h, in, insize, colortype, bitdepth);
+	if (buffer && !error) {
+		State state;
+		state.info_raw.colortype = colortype;
+		state.info_raw.bitdepth = bitdepth;
+		u32 buffersize = png_get_raw_size(w, h, &state.info_raw);
+		out.insert(out.end(), &buffer[0], &buffer[buffersize]);
+	}
+	png_free(buffer);
+	return error;
+}
+
+u32 decode(std::vector<u8> &out, u32 &w, u32 &h, const std::vector<u8> &in, pngColorType colortype,
+	   u32 bitdepth)
+{
+	return decode(out, w, h, in.empty() ? 0 : &in[0], (u32)in.size(), colortype, bitdepth);
+}
+
+u32 decode(std::vector<u8> &out, u32 &w, u32 &h, State &state, const u8 *in, u32 insize)
+{
+	u8 *buffer = NULL;
+	u32 error = png_decode(&buffer, &w, &h, &state, in, insize);
+	if (buffer && !error) {
+		u32 buffersize = png_get_raw_size(w, h, &state.info_raw);
+		out.insert(out.end(), &buffer[0], &buffer[buffersize]);
+	}
+	png_free(buffer);
+	return error;
+}
+
+u32 decode(std::vector<u8> &out, u32 &w, u32 &h, State &state, const std::vector<u8> &in)
+{
+	return decode(out, w, h, state, in.empty() ? 0 : &in[0], in.size());
+}
+
+#ifdef PNG_COMPILE_DISK
+u32 decode(std::vector<u8> &out, u32 &w, u32 &h, const std::string &filename,
+	   pngColorType colortype, u32 bitdepth)
+{
+	std::vector<u8> buffer;
+	/* safe output values in case error happens */
+	w = h = 0;
+	u32 error = load_file(buffer, filename);
+	if (error)
+		return error;
+	return decode(out, w, h, buffer, colortype, bitdepth);
+}
+#endif /* PNG_COMPILE_DECODER */
+#endif /* PNG_COMPILE_DISK */
+
+#ifdef PNG_COMPILE_ENCODER
+u32 encode(std::vector<u8> &out, const u8 *in, u32 w, u32 h, pngColorType colortype, u32 bitdepth)
+{
+	u8 *buffer;
+	u32 buffersize;
+	u32 error = png_encode_memory(&buffer, &buffersize, in, w, h, colortype, bitdepth);
+	if (buffer) {
+		out.insert(out.end(), &buffer[0], &buffer[buffersize]);
+		png_free(buffer);
+	}
+	return error;
+}
+
+u32 encode(std::vector<u8> &out, const std::vector<u8> &in, u32 w, u32 h, pngColorType colortype,
+	   u32 bitdepth)
+{
+	if (png_get_raw_size_lct(w, h, colortype, bitdepth) > in.size())
+		return 84;
+	return encode(out, in.empty() ? 0 : &in[0], w, h, colortype, bitdepth);
+}
+
+u32 encode(std::vector<u8> &out, const u8 *in, u32 w, u32 h, State &state)
+{
+	u8 *buffer;
+	u32 buffersize;
+	u32 error = png_encode(&buffer, &buffersize, in, w, h, &state);
+	if (buffer) {
+		out.insert(out.end(), &buffer[0], &buffer[buffersize]);
+		png_free(buffer);
+	}
+	return error;
+}
+
+u32 encode(std::vector<u8> &out, const std::vector<u8> &in, u32 w, u32 h, State &state)
+{
+	if (png_get_raw_size(w, h, &state.info_raw) > in.size())
+		return 84;
+	return encode(out, in.empty() ? 0 : &in[0], w, h, state);
+}
+
+#ifdef PNG_COMPILE_DISK
+u32 encode(const std::string &filename, const u8 *in, u32 w, u32 h, pngColorType colortype,
+	   u32 bitdepth)
+{
+	std::vector<u8> buffer;
+	u32 error = encode(buffer, in, w, h, colortype, bitdepth);
+	if (!error)
+		error = save_file(buffer, filename);
+	return error;
+}
+
+u32 encode(const std::string &filename, const std::vector<u8> &in, u32 w, u32 h,
+	   pngColorType colortype, u32 bitdepth)
+{
+	if (png_get_raw_size_lct(w, h, colortype, bitdepth) > in.size())
+		return 84;
+	return encode(filename, in.empty() ? 0 : &in[0], w, h, colortype, bitdepth);
+}
+#endif /* PNG_COMPILE_DISK */
+#endif /* PNG_COMPILE_ENCODER */
+#endif /* PNG_COMPILE_PNG */
+} /* namespace png */
+#endif /*PNG_COMPILE_CPP*/
diff --git a/libs/libgui/png.h b/libs/libgui/png.h
new file mode 100644
index 0000000..dba0947
--- /dev/null
+++ b/libs/libgui/png.h
@@ -0,0 +1,1036 @@
+/*
+LodePNG version 20201017
+
+Copyright (c) 2005-2020 Lode Vandevenne
+Copyright (c) 2021 Marvin Borner
+
+This software is provided 'as-is', without any express or implied
+warranty. In no event will the authors be held liable for any damages
+arising from the use of this software.
+
+Permission is granted to anyone to use this software for any purpose,
+including commercial applications, and to alter it and redistribute it
+freely, subject to the following restrictions:
+
+    1. The origin of this software must not be misrepresented; you must not
+    claim that you wrote the original software. If you use this software
+    in a product, an acknowledgment in the product documentation would be
+    appreciated but is not required.
+
+    2. Altered source versions must be plainly marked as such, and must not be
+    misrepresented as being the original software.
+
+    3. This notice may not be removed or altered from any source
+    distribution.
+*/
+
+#ifndef PNG_H
+#define PNG_H
+
+/* For now! TODO: Use PNG encoding */
+#define PNG_NO_COMPILE_ENCODER
+
+#include <def.h>
+extern const char *PNG_VERSION_STRING;
+
+/*
+The following #defines are used to create code sections. They can be disabled
+to disable code sections, which can give faster compile time and smaller binary.
+The "NO_COMPILE" defines are designed to be used to pass as defines to the
+compiler command to disable them without modifying this header, e.g.
+-DPNG_NO_COMPILE_ZLIB for gcc.
+In addition to those below, you can also define PNG_NO_COMPILE_CRC to
+allow implementing a custom png_crc32.
+*/
+/*deflate & zlib. If disabled, you must specify alternative zlib functions in
+the custom_zlib field of the compress and decompress settings*/
+#ifndef PNG_NO_COMPILE_ZLIB
+#define PNG_COMPILE_ZLIB
+#endif
+
+/*png encoder and png decoder*/
+#ifndef PNG_NO_COMPILE_PNG
+#define PNG_COMPILE_PNG
+#endif
+
+/*deflate&zlib decoder and png decoder*/
+#ifndef PNG_NO_COMPILE_DECODER
+#define PNG_COMPILE_DECODER
+#endif
+
+/*deflate&zlib encoder and png encoder*/
+#ifndef PNG_NO_COMPILE_ENCODER
+#define PNG_COMPILE_ENCODER
+#endif
+
+/*the optional built in harddisk file loading and saving functions*/
+#ifndef PNG_NO_COMPILE_DISK
+#define PNG_COMPILE_DISK
+#endif
+
+/*support for chunks other than IHDR, IDAT, PLTE, tRNS, IEND: ancillary and unknown chunks*/
+#ifndef PNG_NO_COMPILE_ANCILLARY_CHUNKS
+#define PNG_COMPILE_ANCILLARY_CHUNKS
+#endif
+
+/*ability to convert error numerical codes to English text string*/
+#ifndef PNG_NO_COMPILE_ERROR_TEXT
+#define PNG_COMPILE_ERROR_TEXT
+#endif
+
+/*Compile the default allocators (C's free, malloc and realloc). If you disable this,
+you can define the functions png_free, png_malloc and png_realloc in your
+source files with custom allocators.*/
+#ifndef PNG_NO_COMPILE_ALLOCATORS
+#define PNG_COMPILE_ALLOCATORS
+#endif
+
+/*compile the C++ version (you can disable the C++ wrapper here even when compiling for C++)*/
+#ifdef __cplusplus
+#ifndef PNG_NO_COMPILE_CPP
+#define PNG_COMPILE_CPP
+#endif
+#endif
+
+#ifdef PNG_COMPILE_CPP
+#include <string>
+#include <vector>
+#endif /*PNG_COMPILE_CPP*/
+
+#ifdef PNG_COMPILE_PNG
+/*The PNG color types (also used for raw image).*/
+typedef enum pngColorType {
+	LCT_GREY = 0, /*grayscale: 1,2,4,8,16 bit*/
+	LCT_RGB = 2, /*RGB: 8,16 bit*/
+	LCT_PALETTE = 3, /*palette: 1,2,4,8 bit*/
+	LCT_GREY_ALPHA = 4, /*grayscale with alpha: 8,16 bit*/
+	LCT_RGBA = 6, /*RGB with alpha: 8,16 bit*/
+	/*LCT_MAX_OCTET_VALUE lets the compiler allow this enum to represent any invalid
+  byte value from 0 to 255 that could be present in an invalid PNG file header. Do
+  not use, compare with or set the name LCT_MAX_OCTET_VALUE, instead either use
+  the valid color type names above, or numeric values like 1 or 7 when checking for
+  particular disallowed color type byte values, or cast to integer to print it.*/
+	LCT_MAX_OCTET_VALUE = 255
+} pngColorType;
+
+#ifdef PNG_COMPILE_DECODER
+/*
+Converts PNG data in memory to raw pixel data.
+out: Output parameter. Pointer to buffer that will contain the raw pixel data.
+     After decoding, its size is w * h * (bytes per pixel) bytes larger than
+     initially. Bytes per pixel depends on colortype and bitdepth.
+     Must be freed after usage with free(*out).
+     Note: for 16-bit per channel colors, uses big endian format like PNG does.
+w: Output parameter. Pointer to width of pixel data.
+h: Output parameter. Pointer to height of pixel data.
+in: Memory buffer with the PNG file.
+insize: size of the in buffer.
+colortype: the desired color type for the raw output image. See explanation on PNG color types.
+bitdepth: the desired bit depth for the raw output image. See explanation on PNG color types.
+Return value: png error code (0 means no error).
+*/
+u32 png_decode_memory(u8 **out, u32 *w, u32 *h, const u8 *in, u32 insize, pngColorType colortype,
+		      u32 bitdepth);
+
+/*Same as png_decode_memory, but always decodes to 32-bit RGBA raw image*/
+u32 png_decode32(u8 **out, u32 *w, u32 *h, const u8 *in, u32 insize);
+
+/*Same as png_decode_memory, but always decodes to 24-bit RGB raw image*/
+u32 png_decode24(u8 **out, u32 *w, u32 *h, const u8 *in, u32 insize);
+
+#ifdef PNG_COMPILE_DISK
+/*
+Load PNG from disk, from file with given name.
+Same as the other decode functions, but instead takes a filename as input.
+*/
+u32 png_decode_file(u8 **out, u32 *w, u32 *h, const char *filename, pngColorType colortype,
+		    u32 bitdepth);
+
+/*Same as png_decode_file, but always decodes to 32-bit RGBA raw image.*/
+u32 png_decode32_file(u8 **out, u32 *w, u32 *h, const char *filename);
+
+/*Same as png_decode_file, but always decodes to 24-bit RGB raw image.*/
+u32 png_decode24_file(u8 **out, u32 *w, u32 *h, const char *filename);
+#endif /*PNG_COMPILE_DISK*/
+#endif /*PNG_COMPILE_DECODER*/
+
+#ifdef PNG_COMPILE_ENCODER
+/*
+Converts raw pixel data into a PNG image in memory. The colortype and bitdepth
+  of the output PNG image cannot be chosen, they are automatically determined
+  by the colortype, bitdepth and content of the input pixel data.
+  Note: for 16-bit per channel colors, needs big endian format like PNG does.
+out: Output parameter. Pointer to buffer that will contain the PNG image data.
+     Must be freed after usage with free(*out).
+outsize: Output parameter. Pointer to the size in bytes of the out buffer.
+image: The raw pixel data to encode. The size of this buffer should be
+       w * h * (bytes per pixel), bytes per pixel depends on colortype and bitdepth.
+w: width of the raw pixel data in pixels.
+h: height of the raw pixel data in pixels.
+colortype: the color type of the raw input image. See explanation on PNG color types.
+bitdepth: the bit depth of the raw input image. See explanation on PNG color types.
+Return value: png error code (0 means no error).
+*/
+u32 png_encode_memory(u8 **out, u32 *outsize, const u8 *image, u32 w, u32 h, pngColorType colortype,
+		      u32 bitdepth);
+
+/*Same as png_encode_memory, but always encodes from 32-bit RGBA raw image.*/
+u32 png_encode32(u8 **out, u32 *outsize, const u8 *image, u32 w, u32 h);
+
+/*Same as png_encode_memory, but always encodes from 24-bit RGB raw image.*/
+u32 png_encode24(u8 **out, u32 *outsize, const u8 *image, u32 w, u32 h);
+
+#ifdef PNG_COMPILE_DISK
+/*
+Converts raw pixel data into a PNG file on disk.
+Same as the other encode functions, but instead takes a filename as output.
+NOTE: This overwrites existing files without warning!
+*/
+u32 png_encode_file(const char *filename, const u8 *image, u32 w, u32 h, pngColorType colortype,
+		    u32 bitdepth);
+
+/*Same as png_encode_file, but always encodes from 32-bit RGBA raw image.*/
+u32 png_encode32_file(const char *filename, const u8 *image, u32 w, u32 h);
+
+/*Same as png_encode_file, but always encodes from 24-bit RGB raw image.*/
+u32 png_encode24_file(const char *filename, const u8 *image, u32 w, u32 h);
+#endif /*PNG_COMPILE_DISK*/
+#endif /*PNG_COMPILE_ENCODER*/
+
+#ifdef PNG_COMPILE_CPP
+namespace png
+{
+#ifdef PNG_COMPILE_DECODER
+/*Same as png_decode_memory, but decodes to an std::vector. The colortype
+is the format to output the pixels to. Default is RGBA 8-bit per channel.*/
+u32 decode(std::vector<u8> &out, u32 &w, u32 &h, const u8 *in, u32 insize,
+	   pngColorType colortype = LCT_RGBA, u32 bitdepth = 8);
+u32 decode(std::vector<u8> &out, u32 &w, u32 &h, const std::vector<u8> &in,
+	   pngColorType colortype = LCT_RGBA, u32 bitdepth = 8);
+#ifdef PNG_COMPILE_DISK
+/*
+Converts PNG file from disk to raw pixel data in memory.
+Same as the other decode functions, but instead takes a filename as input.
+*/
+u32 decode(std::vector<u8> &out, u32 &w, u32 &h, const std::string &filename,
+	   pngColorType colortype = LCT_RGBA, u32 bitdepth = 8);
+#endif /* PNG_COMPILE_DISK */
+#endif /* PNG_COMPILE_DECODER */
+
+#ifdef PNG_COMPILE_ENCODER
+/*Same as png_encode_memory, but encodes to an std::vector. colortype
+is that of the raw input data. The output PNG color type will be auto chosen.*/
+u32 encode(std::vector<u8> &out, const u8 *in, u32 w, u32 h, pngColorType colortype = LCT_RGBA,
+	   u32 bitdepth = 8);
+u32 encode(std::vector<u8> &out, const std::vector<u8> &in, u32 w, u32 h,
+	   pngColorType colortype = LCT_RGBA, u32 bitdepth = 8);
+#ifdef PNG_COMPILE_DISK
+/*
+Converts 32-bit RGBA raw pixel data into a PNG file on disk.
+Same as the other encode functions, but instead takes a filename as output.
+NOTE: This overwrites existing files without warning!
+*/
+u32 encode(const std::string &filename, const u8 *in, u32 w, u32 h,
+	   pngColorType colortype = LCT_RGBA, u32 bitdepth = 8);
+u32 encode(const std::string &filename, const std::vector<u8> &in, u32 w, u32 h,
+	   pngColorType colortype = LCT_RGBA, u32 bitdepth = 8);
+#endif /* PNG_COMPILE_DISK */
+#endif /* PNG_COMPILE_ENCODER */
+} /* namespace png */
+#endif /*PNG_COMPILE_CPP*/
+#endif /*PNG_COMPILE_PNG*/
+
+#ifdef PNG_COMPILE_ERROR_TEXT
+/*Returns an English description of the numerical error code.*/
+const char *png_error_text(u32 code);
+#endif /*PNG_COMPILE_ERROR_TEXT*/
+
+#ifdef PNG_COMPILE_DECODER
+/*Settings for zlib decompression*/
+typedef struct pngDecompressSettings pngDecompressSettings;
+struct pngDecompressSettings {
+	/* Check pngDecoderSettings for more ignorable errors such as ignore_crc */
+	u32 ignore_adler32; /*if 1, continue and don't give an error message if the Adler32 checksum is corrupted*/
+	u32 ignore_nlen; /*ignore complement of len checksum in uncompressed blocks*/
+
+	/*Maximum decompressed size, beyond this the decoder may (and is encouraged to) stop decoding,
+  return an error, output a data size > max_output_size and all the data up to that point. This is
+  not hard limit nor a guarantee, but can prevent excessive memory usage. This setting is
+  ignored by the PNG decoder, but is used by the deflate/zlib decoder and can be used by custom ones.
+  Set to 0 to impose no limit (the default).*/
+	u32 max_output_size;
+
+	/*use custom zlib decoder instead of built in one (default: null).
+  Should return 0 if success, any non-0 if error (numeric value not exposed).*/
+	u32 (*custom_zlib)(u8 **, u32 *, const u8 *, u32, const pngDecompressSettings *);
+	/*use custom deflate decoder instead of built in one (default: null)
+  if custom_zlib is not null, custom_inflate is ignored (the zlib format uses deflate).
+  Should return 0 if success, any non-0 if error (numeric value not exposed).*/
+	u32 (*custom_inflate)(u8 **, u32 *, const u8 *, u32, const pngDecompressSettings *);
+
+	const void *custom_context; /*optional custom settings for custom functions*/
+};
+
+extern const pngDecompressSettings png_default_decompress_settings;
+void png_decompress_settings_init(pngDecompressSettings *settings);
+#endif /*PNG_COMPILE_DECODER*/
+
+#ifdef PNG_COMPILE_ENCODER
+/*
+Settings for zlib compression. Tweaking these settings tweaks the balance
+between speed and compression ratio.
+*/
+typedef struct pngCompressSettings pngCompressSettings;
+struct pngCompressSettings /*deflate = compress*/ {
+	/*LZ77 related settings*/
+	u32 btype; /*the block type for LZ (0, 1, 2 or 3, see zlib standard). Should be 2 for proper compression.*/
+	u32 use_lz77; /*whether or not to use LZ77. Should be 1 for proper compression.*/
+	u32 windowsize; /*must be a power of two <= 32768. higher compresses more but is slower. Default value: 2048.*/
+	u32 minmatch; /*minimum lz77 length. 3 is normally best, 6 can be better for some PNGs. Default: 0*/
+	u32 nicematch; /*stop searching if >= this length found. Set to 258 for best compression. Default: 128*/
+	u32 lazymatching; /*use lazy matching: better compression but a bit slower. Default: true*/
+
+	/*use custom zlib encoder instead of built in one (default: null)*/
+	u32 (*custom_zlib)(u8 **, u32 *, const u8 *, u32, const pngCompressSettings *);
+	/*use custom deflate encoder instead of built in one (default: null)
+  if custom_zlib is used, custom_deflate is ignored since only the built in
+  zlib function will call custom_deflate*/
+	u32 (*custom_deflate)(u8 **, u32 *, const u8 *, u32, const pngCompressSettings *);
+
+	const void *custom_context; /*optional custom settings for custom functions*/
+};
+
+extern const pngCompressSettings png_default_compress_settings;
+void png_compress_settings_init(pngCompressSettings *settings);
+#endif /*PNG_COMPILE_ENCODER*/
+
+#ifdef PNG_COMPILE_PNG
+/*
+Color mode of an image. Contains all information required to decode the pixel
+bits to RGBA colors. This information is the same as used in the PNG file
+format, and is used both for PNG and raw image data in png.
+*/
+typedef struct pngColorMode {
+	/*header (IHDR)*/
+	pngColorType
+		colortype; /*color type, see PNG standard or documentation further in this header file*/
+	u32 bitdepth; /*bits per sample, see PNG standard or documentation further in this header file*/
+
+	/*
+  palette (PLTE and tRNS)
+
+  Dynamically allocated with the colors of the palette, including alpha.
+  This field may not be allocated directly, use png_color_mode_init first,
+  then png_palette_add per color to correctly initialize it (to ensure size
+  of exactly 1024 bytes).
+
+  The alpha channels must be set as well, set them to 255 for opaque images.
+
+  When decoding, by default you can ignore this palette, since png already
+  fills the palette colors in the pixels of the raw RGBA output.
+
+  The palette is only supported for color type 3.
+  */
+	u8 *palette; /*palette in RGBARGBA... order. Must be either 0, or when allocated must have 1024 bytes*/
+	u32 palettesize; /*palette size in number of colors (amount of used bytes is 4 * palettesize)*/
+
+	/*
+  transparent color key (tRNS)
+
+  This color uses the same bit depth as the bitdepth value in this struct, which can be 1-bit to 16-bit.
+  For grayscale PNGs, r, g and b will all 3 be set to the same.
+
+  When decoding, by default you can ignore this information, since png sets
+  pixels with this key to transparent already in the raw RGBA output.
+
+  The color key is only supported for color types 0 and 2.
+  */
+	u32 key_defined; /*is a transparent color key given? 0 = false, 1 = true*/
+	u32 key_r; /*red/grayscale component of color key*/
+	u32 key_g; /*green component of color key*/
+	u32 key_b; /*blue component of color key*/
+} pngColorMode;
+
+/*init, cleanup and copy functions to use with this struct*/
+void png_color_mode_init(pngColorMode *info);
+void png_color_mode_cleanup(pngColorMode *info);
+/*return value is error code (0 means no error)*/
+u32 png_color_mode_copy(pngColorMode *dest, const pngColorMode *source);
+/* Makes a temporary pngColorMode that does not need cleanup (no palette) */
+pngColorMode png_color_mode_make(pngColorType colortype, u32 bitdepth);
+
+void png_palette_clear(pngColorMode *info);
+/*add 1 color to the palette*/
+u32 png_palette_add(pngColorMode *info, u8 r, u8 g, u8 b, u8 a);
+
+/*get the total amount of bits per pixel, based on colortype and bitdepth in the struct*/
+u32 png_get_bpp(const pngColorMode *info);
+/*get the amount of color channels used, based on colortype in the struct.
+If a palette is used, it counts as 1 channel.*/
+u32 png_get_channels(const pngColorMode *info);
+/*is it a grayscale type? (only colortype 0 or 4)*/
+u32 png_is_greyscale_type(const pngColorMode *info);
+/*has it got an alpha channel? (only colortype 2 or 6)*/
+u32 png_is_alpha_type(const pngColorMode *info);
+/*has it got a palette? (only colortype 3)*/
+u32 png_is_palette_type(const pngColorMode *info);
+/*only returns true if there is a palette and there is a value in the palette with alpha < 255.
+Loops through the palette to check this.*/
+u32 png_has_palette_alpha(const pngColorMode *info);
+/*
+Check if the given color info indicates the possibility of having non-opaque pixels in the PNG image.
+Returns true if the image can have translucent or invisible pixels (it still be opaque if it doesn't use such pixels).
+Returns false if the image can only have opaque pixels.
+In detail, it returns true only if it's a color type with alpha, or has a palette with non-opaque values,
+or if "key_defined" is true.
+*/
+u32 png_can_have_alpha(const pngColorMode *info);
+/*Returns the byte size of a raw image buffer with given width, height and color mode*/
+u32 png_get_raw_size(u32 w, u32 h, const pngColorMode *color);
+
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+/*The information of a Time chunk in PNG.*/
+typedef struct pngTime {
+	u32 year; /*2 bytes used (0-65535)*/
+	u32 month; /*1-12*/
+	u32 day; /*1-31*/
+	u32 hour; /*0-23*/
+	u32 minute; /*0-59*/
+	u32 second; /*0-60 (to allow for leap seconds)*/
+} pngTime;
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+
+/*Information about the PNG image, except pixels, width and height.*/
+typedef struct pngInfo {
+	/*header (IHDR), palette (PLTE) and transparency (tRNS) chunks*/
+	u32 compression_method; /*compression method of the original file. Always 0.*/
+	u32 filter_method; /*filter method of the original file*/
+	u32 interlace_method; /*interlace method of the original file: 0=none, 1=Adam7*/
+	pngColorMode color; /*color type and bits, palette and transparency of the PNG file*/
+
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+	/*
+  Suggested background color chunk (bKGD)
+
+  This uses the same color mode and bit depth as the PNG (except no alpha channel),
+  with values truncated to the bit depth in the u32 integer.
+
+  For grayscale and palette PNGs, the value is stored in background_r. The values
+  in background_g and background_b are then unused.
+
+  So when decoding, you may get these in a different color mode than the one you requested
+  for the raw pixels.
+
+  When encoding with auto_convert, you must use the color model defined in info_png.color for
+  these values. The encoder normally ignores info_png.color when auto_convert is on, but will
+  use it to interpret these values (and convert copies of them to its chosen color model).
+
+  When encoding, avoid setting this to an expensive color, such as a non-gray value
+  when the image is gray, or the compression will be worse since it will be forced to
+  write the PNG with a more expensive color mode (when auto_convert is on).
+
+  The decoder does not use this background color to edit the color of pixels. This is a
+  completely optional metadata feature.
+  */
+	u32 background_defined; /*is a suggested background color given?*/
+	u32 background_r; /*red/gray/palette component of suggested background color*/
+	u32 background_g; /*green component of suggested background color*/
+	u32 background_b; /*blue component of suggested background color*/
+
+	/*
+  Non-international text chunks (tEXt and zTXt)
+
+  The char** arrays each contain num strings. The actual messages are in
+  text_strings, while text_keys are keywords that give a short description what
+  the actual text represents, e.g. Title, Author, Description, or anything else.
+
+  All the string fields below including strings, keys, names and language tags are null terminated.
+  The PNG specification uses null characters for the keys, names and tags, and forbids null
+  characters to appear in the main text which is why we can use null termination everywhere here.
+
+  A keyword is minimum 1 character and maximum 79 characters long (plus the
+  additional null terminator). It's discouraged to use a single line length
+  longer than 79 characters for texts.
+
+  Don't allocate these text buffers yourself. Use the init/cleanup functions
+  correctly and use png_add_text and png_clear_text.
+
+  Standard text chunk keywords and strings are encoded using Latin-1.
+  */
+	u32 text_num; /*the amount of texts in these char** buffers (there may be more texts in itext)*/
+	char **text_keys; /*the keyword of a text chunk (e.g. "Comment")*/
+	char **text_strings; /*the actual text*/
+
+	/*
+  International text chunks (iTXt)
+  Similar to the non-international text chunks, but with additional strings
+  "langtags" and "transkeys", and the following text encodings are used:
+  keys: Latin-1, langtags: ASCII, transkeys and strings: UTF-8.
+  keys must be 1-79 characters (plus the additional null terminator), the other
+  strings are any length.
+  */
+	u32 itext_num; /*the amount of international texts in this PNG*/
+	char **itext_keys; /*the English keyword of the text chunk (e.g. "Comment")*/
+	char **itext_langtags; /*language tag for this text's language, ISO/IEC 646 string, e.g. ISO 639 language tag*/
+	char **itext_transkeys; /*keyword translated to the international language - UTF-8 string*/
+	char **itext_strings; /*the actual international text - UTF-8 string*/
+
+	/*time chunk (tIME)*/
+	u32 time_defined; /*set to 1 to make the encoder generate a tIME chunk*/
+	pngTime time;
+
+	/*phys chunk (pHYs)*/
+	u32 phys_defined; /*if 0, there is no pHYs chunk and the values below are undefined, if 1 else there is one*/
+	u32 phys_x; /*pixels per unit in x direction*/
+	u32 phys_y; /*pixels per unit in y direction*/
+	u32 phys_unit; /*may be 0 (unknown unit) or 1 (metre)*/
+
+	/*
+  Color profile related chunks: gAMA, cHRM, sRGB, iCPP
+
+  png does not apply any color conversions on pixels in the encoder or decoder and does not interpret these color
+  profile values. It merely passes on the information. If you wish to use color profiles and convert colors, please
+  use these values with a color management library.
+
+  See the PNG, ICC and sRGB specifications for more information about the meaning of these values.
+  */
+
+	/* gAMA chunk: optional, overridden by sRGB or iCCP if those are present. */
+	u32 gama_defined; /* Whether a gAMA chunk is present (0 = not present, 1 = present). */
+	u32 gama_gamma; /* Gamma exponent times 100000 */
+
+	/* cHRM chunk: optional, overridden by sRGB or iCCP if those are present. */
+	u32 chrm_defined; /* Whether a cHRM chunk is present (0 = not present, 1 = present). */
+	u32 chrm_white_x; /* White Point x times 100000 */
+	u32 chrm_white_y; /* White Point y times 100000 */
+	u32 chrm_red_x; /* Red x times 100000 */
+	u32 chrm_red_y; /* Red y times 100000 */
+	u32 chrm_green_x; /* Green x times 100000 */
+	u32 chrm_green_y; /* Green y times 100000 */
+	u32 chrm_blue_x; /* Blue x times 100000 */
+	u32 chrm_blue_y; /* Blue y times 100000 */
+
+	/*
+  sRGB chunk: optional. May not appear at the same time as iCCP.
+  If gAMA is also present gAMA must contain value 45455.
+  If cHRM is also present cHRM must contain respectively 31270,32900,64000,33000,30000,60000,15000,6000.
+  */
+	u32 srgb_defined; /* Whether an sRGB chunk is present (0 = not present, 1 = present). */
+	u32 srgb_intent; /* Rendering intent: 0=perceptual, 1=rel. colorimetric, 2=saturation, 3=abs. colorimetric */
+
+	/*
+  iCCP chunk: optional. May not appear at the same time as sRGB.
+
+  png does not parse or use the ICC profile (except its color space header field for an edge case), a
+  separate library to handle the ICC data (not included in png) format is needed to use it for color
+  management and conversions.
+
+  For encoding, if iCCP is present, gAMA and cHRM are recommended to be added as well with values that match the ICC
+  profile as closely as possible, if you wish to do this you should provide the correct values for gAMA and cHRM and
+  enable their '_defined' flags since png will not automatically compute them from the ICC profile.
+
+  For encoding, the ICC profile is required by the PNG specification to be an "RGB" profile for non-gray
+  PNG color types and a "GRAY" profile for gray PNG color types. If you disable auto_convert, you must ensure
+  the ICC profile type matches your requested color type, else the encoder gives an error. If auto_convert is
+  enabled (the default), and the ICC profile is not a good match for the pixel data, this will result in an encoder
+  error if the pixel data has non-gray pixels for a GRAY profile, or a silent less-optimal compression of the pixel
+  data if the pixels could be encoded as grayscale but the ICC profile is RGB.
+
+  To avoid this do not set an ICC profile in the image unless there is a good reason for it, and when doing so
+  make sure you compute it carefully to avoid the above problems.
+  */
+	u32 iccp_defined; /* Whether an iCCP chunk is present (0 = not present, 1 = present). */
+	char *iccp_name; /* Null terminated string with profile name, 1-79 bytes */
+	/*
+  The ICC profile in iccp_profile_size bytes.
+  Don't allocate this buffer yourself. Use the init/cleanup functions
+  correctly and use png_set_icc and png_clear_icc.
+  */
+	u8 *iccp_profile;
+	u32 iccp_profile_size; /* The size of iccp_profile in bytes */
+
+	/* End of color profile related chunks */
+
+	/*
+  unknown chunks: chunks not known by png, passed on byte for byte.
+
+  There are 3 buffers, one for each position in the PNG where unknown chunks can appear.
+  Each buffer contains all unknown chunks for that position consecutively.
+  The 3 positions are:
+  0: between IHDR and PLTE, 1: between PLTE and IDAT, 2: between IDAT and IEND.
+
+  For encoding, do not store critical chunks or known chunks that are enabled with a "_defined" flag
+  above in here, since the encoder will blindly follow this and could then encode an invalid PNG file
+  (such as one with two IHDR chunks or the disallowed combination of sRGB with iCCP). But do use
+  this if you wish to store an ancillary chunk that is not supported by png (such as sPLT or hIST),
+  or any non-standard PNG chunk.
+
+  Do not allocate or traverse this data yourself. Use the chunk traversing functions declared
+  later, such as png_chunk_next and png_chunk_append, to read/write this struct.
+  */
+	u8 *unknown_chunks_data[3];
+	u32 unknown_chunks_size[3]; /*size in bytes of the unknown chunks, given for protection*/
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+} pngInfo;
+
+/*init, cleanup and copy functions to use with this struct*/
+void png_info_init(pngInfo *info);
+void png_info_cleanup(pngInfo *info);
+/*return value is error code (0 means no error)*/
+u32 png_info_copy(pngInfo *dest, const pngInfo *source);
+
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+u32 png_add_text(pngInfo *info, const char *key, const char *str); /*push back both texts at once*/
+void png_clear_text(pngInfo *info); /*use this to clear the texts again after you filled them in*/
+
+u32 png_add_itext(pngInfo *info, const char *key, const char *langtag, const char *transkey,
+		  const char *str); /*push back the 4 texts of 1 chunk at once*/
+void png_clear_itext(pngInfo *info); /*use this to clear the itexts again after you filled them in*/
+
+/*replaces if exists*/
+u32 png_set_icc(pngInfo *info, const char *name, const u8 *profile, u32 profile_size);
+void png_clear_icc(pngInfo *info); /*use this to clear the texts again after you filled them in*/
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+
+/*
+Converts raw buffer from one color type to another color type, based on
+pngColorMode structs to describe the input and output color type.
+See the reference manual at the end of this header file to see which color conversions are supported.
+return value = png error code (0 if all went ok, an error if the conversion isn't supported)
+The out buffer must have size (w * h * bpp + 7) / 8, where bpp is the bits per pixel
+of the output color type (png_get_bpp).
+For < 8 bpp images, there should not be padding bits at the end of scanlines.
+For 16-bit per channel colors, uses big endian format like PNG does.
+Return value is png error code
+*/
+u32 png_convert(u8 *out, const u8 *in, const pngColorMode *mode_out, const pngColorMode *mode_in,
+		u32 w, u32 h);
+
+#ifdef PNG_COMPILE_DECODER
+/*
+Settings for the decoder. This contains settings for the PNG and the Zlib
+decoder, but not the Info settings from the Info structs.
+*/
+typedef struct pngDecoderSettings {
+	pngDecompressSettings zlibsettings; /*in here is the setting to ignore Adler32 checksums*/
+
+	/* Check pngDecompressSettings for more ignorable errors such as ignore_adler32 */
+	u32 ignore_crc; /*ignore CRC checksums*/
+	u32 ignore_critical; /*ignore unknown critical chunks*/
+	u32 ignore_end; /*ignore issues at end of file if possible (missing IEND chunk, too large chunk, ...)*/
+	/* TODO: make a system involving warnings with levels and a strict mode instead. Other potentially recoverable
+     errors: srgb rendering intent value, size of content of ancillary chunks, more than 79 characters for some
+     strings, placement/combination rules for ancillary chunks, crc of unknown chunks, allowed characters
+     in string keys, etc... */
+
+	u32 color_convert; /*whether to convert the PNG to the color type you want. Default: yes*/
+
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+	u32 read_text_chunks; /*if false but remember_unknown_chunks is true, they're stored in the unknown chunks*/
+
+	/*store all bytes from unknown chunks in the pngInfo (off by default, useful for a png editor)*/
+	u32 remember_unknown_chunks;
+
+	/* maximum size for decompressed text chunks. If a text chunk's text is larger than this, an error is returned,
+  unless reading text chunks is disabled or this limit is set higher or disabled. Set to 0 to allow any size.
+  By default it is a value that prevents unreasonably large strings from hogging memory. */
+	u32 max_text_size;
+
+	/* maximum size for compressed ICC chunks. If the ICC profile is larger than this, an error will be returned. Set to
+  0 to allow any size. By default this is a value that prevents ICC profiles that would be much larger than any
+  legitimate profile could be to hog memory. */
+	u32 max_icc_size;
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+} pngDecoderSettings;
+
+void png_decoder_settings_init(pngDecoderSettings *settings);
+#endif /*PNG_COMPILE_DECODER*/
+
+#ifdef PNG_COMPILE_ENCODER
+/*automatically use color type with less bits per pixel if losslessly possible. Default: AUTO*/
+typedef enum pngFilterStrategy {
+	/*every filter at zero*/
+	LFS_ZERO = 0,
+	/*every filter at 1, 2, 3 or 4 (paeth), unlike LFS_ZERO not a good choice, but for testing*/
+	LFS_ONE = 1,
+	LFS_TWO = 2,
+	LFS_THREE = 3,
+	LFS_FOUR = 4,
+	/*Use filter that gives minimum sum, as described in the official PNG filter heuristic.*/
+	LFS_MINSUM,
+	/*Use the filter type that gives smallest Shannon entropy for this scanline. Depending
+  on the image, this is better or worse than minsum.*/
+	LFS_ENTROPY,
+	/*
+  Brute-force-search PNG filters by compressing each filter for each scanline.
+  Experimental, very slow, and only rarely gives better compression than MINSUM.
+  */
+	LFS_BRUTE_FORCE,
+	/*use predefined_filters buffer: you specify the filter type for each scanline*/
+	LFS_PREDEFINED
+} pngFilterStrategy;
+
+/*Gives characteristics about the integer RGBA colors of the image (count, alpha channel usage, bit depth, ...),
+which helps decide which color model to use for encoding.
+Used internally by default if "auto_convert" is enabled. Public because it's useful for custom algorithms.*/
+typedef struct pngColorStats {
+	u32 colored; /*not grayscale*/
+	u32 key; /*image is not opaque and color key is possible instead of full alpha*/
+	u16 key_r; /*key values, always as 16-bit, in 8-bit case the byte is duplicated, e.g. 65535 means 255*/
+	u16 key_g;
+	u16 key_b;
+	u32 alpha; /*image is not opaque and alpha channel or alpha palette required*/
+	u32 numcolors; /*amount of colors, up to 257. Not valid if bits == 16 or allow_palette is disabled.*/
+	u8 palette[1024]; /*Remembers up to the first 256 RGBA colors, in no particular order, only valid when numcolors is valid*/
+	u32 bits; /*bits per channel (not for palette). 1,2 or 4 for grayscale only. 16 if 16-bit per channel required.*/
+	u32 numpixels;
+
+	/*user settings for computing/using the stats*/
+	u32 allow_palette; /*default 1. if 0, disallow choosing palette colortype in auto_choose_color, and don't count numcolors*/
+	u32 allow_greyscale; /*default 1. if 0, choose RGB or RGBA even if the image only has gray colors*/
+} pngColorStats;
+
+void png_color_stats_init(pngColorStats *stats);
+
+/*Get a pngColorStats of the image. The stats must already have been inited.
+Returns error code (e.g. alloc fail) or 0 if ok.*/
+u32 png_compute_color_stats(pngColorStats *stats, const u8 *image, u32 w, u32 h,
+			    const pngColorMode *mode_in);
+
+/*Settings for the encoder.*/
+typedef struct pngEncoderSettings {
+	pngCompressSettings zlibsettings; /*settings for the zlib encoder, such as window size, ...*/
+
+	u32 auto_convert; /*automatically choose output PNG color type. Default: true*/
+
+	/*If true, follows the official PNG heuristic: if the PNG uses a palette or lower than
+  8 bit depth, set all filters to zero. Otherwise use the filter_strategy. Note that to
+  completely follow the official PNG heuristic, filter_palette_zero must be true and
+  filter_strategy must be LFS_MINSUM*/
+	u32 filter_palette_zero;
+	/*Which filter strategy to use when not using zeroes due to filter_palette_zero.
+  Set filter_palette_zero to 0 to ensure always using your chosen strategy. Default: LFS_MINSUM*/
+	pngFilterStrategy filter_strategy;
+	/*used if filter_strategy is LFS_PREDEFINED. In that case, this must point to a buffer with
+  the same length as the amount of scanlines in the image, and each value must <= 5. You
+  have to cleanup this buffer, png will never free it. Don't forget that filter_palette_zero
+  must be set to 0 to ensure this is also used on palette or low bitdepth images.*/
+	const u8 *predefined_filters;
+
+	/*force creating a PLTE chunk if colortype is 2 or 6 (= a suggested palette).
+  If colortype is 3, PLTE is _always_ created.*/
+	u32 force_palette;
+#ifdef PNG_COMPILE_ANCILLARY_CHUNKS
+	/*add png identifier and version as a text chunk, for debugging*/
+	u32 add_id;
+	/*encode text chunks as zTXt chunks instead of tEXt chunks, and use compression in iTXt chunks*/
+	u32 text_compression;
+#endif /*PNG_COMPILE_ANCILLARY_CHUNKS*/
+} pngEncoderSettings;
+
+void png_encoder_settings_init(pngEncoderSettings *settings);
+#endif /*PNG_COMPILE_ENCODER*/
+
+#if defined(PNG_COMPILE_DECODER) || defined(PNG_COMPILE_ENCODER)
+/*The settings, state and information for extended encoding and decoding.*/
+typedef struct pngState {
+#ifdef PNG_COMPILE_DECODER
+	pngDecoderSettings decoder; /*the decoding settings*/
+#endif /*PNG_COMPILE_DECODER*/
+#ifdef PNG_COMPILE_ENCODER
+	pngEncoderSettings encoder; /*the encoding settings*/
+#endif /*PNG_COMPILE_ENCODER*/
+	pngColorMode
+		info_raw; /*specifies the format in which you would like to get the raw pixel buffer*/
+	pngInfo info_png; /*info of the PNG image obtained after decoding*/
+	u32 error;
+} pngState;
+
+/*init, cleanup and copy functions to use with this struct*/
+void png_state_init(pngState *state);
+void png_state_cleanup(pngState *state);
+void png_state_copy(pngState *dest, const pngState *source);
+#endif /* defined(PNG_COMPILE_DECODER) || defined(PNG_COMPILE_ENCODER) */
+
+#ifdef PNG_COMPILE_DECODER
+/*
+Same as png_decode_memory, but uses a pngState to allow custom settings and
+getting much more information about the PNG image and color mode.
+*/
+u32 png_decode(u8 **out, u32 *w, u32 *h, pngState *state, const u8 *in, u32 insize);
+
+/*
+Read the PNG header, but not the actual data. This returns only the information
+that is in the IHDR chunk of the PNG, such as width, height and color type. The
+information is placed in the info_png field of the pngState.
+*/
+u32 png_inspect(u32 *w, u32 *h, pngState *state, const u8 *in, u32 insize);
+#endif /*PNG_COMPILE_DECODER*/
+
+/*
+Reads one metadata chunk (other than IHDR) of the PNG file and outputs what it
+read in the state. Returns error code on failure.
+Use png_inspect first with a new state, then e.g. png_chunk_find_const
+to find the desired chunk type, and if non null use png_inspect_chunk (with
+chunk_pointer - start_of_file as pos).
+Supports most metadata chunks from the PNG standard (gAMA, bKGD, tEXt, ...).
+Ignores unsupported, unknown, non-metadata or IHDR chunks (without error).
+Requirements: &in[pos] must point to start of a chunk, must use regular
+png_inspect first since format of most other chunks depends on IHDR, and if
+there is a PLTE chunk, that one must be inspected before tRNS or bKGD.
+*/
+u32 png_inspect_chunk(pngState *state, u32 pos, const u8 *in, u32 insize);
+
+#ifdef PNG_COMPILE_ENCODER
+/*This function allocates the out buffer with standard malloc and stores the size in *outsize.*/
+u32 png_encode(u8 **out, u32 *outsize, const u8 *image, u32 w, u32 h, pngState *state);
+#endif /*PNG_COMPILE_ENCODER*/
+
+/*
+The png_chunk functions are normally not needed, except to traverse the
+unknown chunks stored in the pngInfo struct, or add new ones to it.
+It also allows traversing the chunks of an encoded PNG file yourself.
+
+The chunk pointer always points to the beginning of the chunk itself, that is
+the first byte of the 4 length bytes.
+
+In the PNG file format, chunks have the following format:
+-4 bytes length: length of the data of the chunk in bytes (chunk itself is 12 bytes longer)
+-4 bytes chunk type (ASCII a-z,A-Z only, see below)
+-length bytes of data (may be 0 bytes if length was 0)
+-4 bytes of CRC, computed on chunk name + data
+
+The first chunk starts at the 8th byte of the PNG file, the entire rest of the file
+exists out of concatenated chunks with the above format.
+
+PNG standard chunk ASCII naming conventions:
+-First byte: uppercase = critical, lowercase = ancillary
+-Second byte: uppercase = public, lowercase = private
+-Third byte: must be uppercase
+-Fourth byte: uppercase = unsafe to copy, lowercase = safe to copy
+*/
+
+/*
+Gets the length of the data of the chunk. Total chunk length has 12 bytes more.
+There must be at least 4 bytes to read from. If the result value is too large,
+it may be corrupt data.
+*/
+u32 png_chunk_length(const u8 *chunk);
+
+/*puts the 4-byte type in null terminated string*/
+void png_chunk_type(char type[5], const u8 *chunk);
+
+/*check if the type is the given type*/
+u8 png_chunk_type_equals(const u8 *chunk, const char *type);
+
+/*0: it's one of the critical chunk types, 1: it's an ancillary chunk (see PNG standard)*/
+u8 png_chunk_ancillary(const u8 *chunk);
+
+/*0: public, 1: private (see PNG standard)*/
+u8 png_chunk_private(const u8 *chunk);
+
+/*0: the chunk is unsafe to copy, 1: the chunk is safe to copy (see PNG standard)*/
+u8 png_chunk_safetocopy(const u8 *chunk);
+
+/*get pointer to the data of the chunk, where the input points to the header of the chunk*/
+u8 *png_chunk_data(u8 *chunk);
+const u8 *png_chunk_data_const(const u8 *chunk);
+
+/*returns 0 if the crc is correct, 1 if it's incorrect (0 for OK as usual!)*/
+u32 png_chunk_check_crc(const u8 *chunk);
+
+/*generates the correct CRC from the data and puts it in the last 4 bytes of the chunk*/
+void png_chunk_generate_crc(u8 *chunk);
+
+/*
+Iterate to next chunks, allows iterating through all chunks of the PNG file.
+Input must be at the beginning of a chunk (result of a previous png_chunk_next call,
+or the 8th byte of a PNG file which always has the first chunk), or alternatively may
+point to the first byte of the PNG file (which is not a chunk but the magic header, the
+function will then skip over it and return the first real chunk).
+Will output pointer to the start of the next chunk, or at or beyond end of the file if there
+is no more chunk after this or possibly if the chunk is corrupt.
+Start this process at the 8th byte of the PNG file.
+In a non-corrupt PNG file, the last chunk should have name "IEND".
+*/
+u8 *png_chunk_next(u8 *chunk, u8 *end);
+const u8 *png_chunk_next_const(const u8 *chunk, const u8 *end);
+
+/*Finds the first chunk with the given type in the range [chunk, end), or returns NULL if not found.*/
+u8 *png_chunk_find(u8 *chunk, u8 *end, const char type[5]);
+const u8 *png_chunk_find_const(const u8 *chunk, const u8 *end, const char type[5]);
+
+/*
+Appends chunk to the data in out. The given chunk should already have its chunk header.
+The out variable and outsize are updated to reflect the new reallocated buffer.
+Returns error code (0 if it went ok)
+*/
+u32 png_chunk_append(u8 **out, u32 *outsize, const u8 *chunk);
+
+/*
+Appends new chunk to out. The chunk to append is given by giving its length, type
+and data separately. The type is a 4-letter string.
+The out variable and outsize are updated to reflect the new reallocated buffer.
+Returne error code (0 if it went ok)
+*/
+u32 png_chunk_create(u8 **out, u32 *outsize, u32 length, const char *type, const u8 *data);
+
+/*Calculate CRC32 of buffer*/
+u32 png_crc32(const u8 *buf, u32 len);
+#endif /*PNG_COMPILE_PNG*/
+
+#ifdef PNG_COMPILE_ZLIB
+/*
+This zlib part can be used independently to zlib compress and decompress a
+buffer. It cannot be used to create gzip files however, and it only supports the
+part of zlib that is required for PNG, it does not support dictionaries.
+*/
+
+#ifdef PNG_COMPILE_DECODER
+/*Inflate a buffer. Inflate is the decompression step of deflate. Out buffer must be freed after use.*/
+u32 png_inflate(u8 **out, u32 *outsize, const u8 *in, u32 insize,
+		const pngDecompressSettings *settings);
+
+/*
+Decompresses Zlib data. Reallocates the out buffer and appends the data. The
+data must be according to the zlib specification.
+Either, *out must be NULL and *outsize must be 0, or, *out must be a valid
+buffer and *outsize its size in bytes. out must be freed by user after usage.
+*/
+u32 png_zlib_decompress(u8 **out, u32 *outsize, const u8 *in, u32 insize,
+			const pngDecompressSettings *settings);
+#endif /*PNG_COMPILE_DECODER*/
+
+#ifdef PNG_COMPILE_ENCODER
+/*
+Compresses data with Zlib. Reallocates the out buffer and appends the data.
+Zlib adds a small header and trailer around the deflate data.
+The data is output in the format of the zlib specification.
+Either, *out must be NULL and *outsize must be 0, or, *out must be a valid
+buffer and *outsize its size in bytes. out must be freed by user after usage.
+*/
+u32 png_zlib_compress(u8 **out, u32 *outsize, const u8 *in, u32 insize,
+		      const pngCompressSettings *settings);
+
+/*
+Find length-limited Huffman code for given frequencies. This function is in the
+public interface only for tests, it's used internally by png_deflate.
+*/
+u32 png_huffman_code_lengths(u32 *lengths, const u32 *frequencies, u32 numcodes, u32 maxbitlen);
+
+/*Compress a buffer with deflate. See RFC 1951. Out buffer must be freed after use.*/
+u32 png_deflate(u8 **out, u32 *outsize, const u8 *in, u32 insize,
+		const pngCompressSettings *settings);
+
+#endif /*PNG_COMPILE_ENCODER*/
+#endif /*PNG_COMPILE_ZLIB*/
+
+#ifdef PNG_COMPILE_DISK
+/*
+Load a file from disk into buffer. The function allocates the out buffer, and
+after usage you should free it.
+out: output parameter, contains pointer to loaded buffer.
+outsize: output parameter, size of the allocated out buffer
+filename: the path to the file to load
+return value: error code (0 means ok)
+*/
+u32 png_load_file(u8 **out, u32 *outsize, const char *filename);
+
+/*
+Save a file from buffer to disk. Warning, if it exists, this function overwrites
+the file without warning!
+buffer: the buffer to write
+buffersize: size of the buffer to write
+filename: the path to the file to save to
+return value: error code (0 means ok)
+*/
+u32 png_save_file(const u8 *buffer, u32 buffersize, const char *filename);
+#endif /*PNG_COMPILE_DISK*/
+
+#ifdef PNG_COMPILE_CPP
+/* The png C++ wrapper uses std::vectors instead of manually allocated memory buffers. */
+namespace png
+{
+#ifdef PNG_COMPILE_PNG
+class State : public pngState {
+    public:
+	State();
+	State(const State &other);
+	~State();
+	State &operator=(const State &other);
+};
+
+#ifdef PNG_COMPILE_DECODER
+/* Same as other png::decode, but using a State for more settings and information. */
+u32 decode(std::vector<u8> &out, u32 &w, u32 &h, State &state, const u8 *in, u32 insize);
+u32 decode(std::vector<u8> &out, u32 &w, u32 &h, State &state, const std::vector<u8> &in);
+#endif /*PNG_COMPILE_DECODER*/
+
+#ifdef PNG_COMPILE_ENCODER
+/* Same as other png::encode, but using a State for more settings and information. */
+u32 encode(std::vector<u8> &out, const u8 *in, u32 w, u32 h, State &state);
+u32 encode(std::vector<u8> &out, const std::vector<u8> &in, u32 w, u32 h, State &state);
+#endif /*PNG_COMPILE_ENCODER*/
+
+#ifdef PNG_COMPILE_DISK
+/*
+Load a file from disk into an std::vector.
+return value: error code (0 means ok)
+*/
+u32 load_file(std::vector<u8> &buffer, const std::string &filename);
+
+/*
+Save the binary data in an std::vector to a file on disk. The file is overwritten
+without warning.
+*/
+u32 save_file(const std::vector<u8> &buffer, const std::string &filename);
+#endif /* PNG_COMPILE_DISK */
+#endif /* PNG_COMPILE_PNG */
+
+#ifdef PNG_COMPILE_ZLIB
+#ifdef PNG_COMPILE_DECODER
+/* Zlib-decompress an u8 buffer */
+u32 decompress(std::vector<u8> &out, const u8 *in, u32 insize,
+	       const pngDecompressSettings &settings = png_default_decompress_settings);
+
+/* Zlib-decompress an std::vector */
+u32 decompress(std::vector<u8> &out, const std::vector<u8> &in,
+	       const pngDecompressSettings &settings = png_default_decompress_settings);
+#endif /* PNG_COMPILE_DECODER */
+
+#ifdef PNG_COMPILE_ENCODER
+/* Zlib-compress an u8 buffer */
+u32 compress(std::vector<u8> &out, const u8 *in, u32 insize,
+	     const pngCompressSettings &settings = png_default_compress_settings);
+
+/* Zlib-compress an std::vector */
+u32 compress(std::vector<u8> &out, const std::vector<u8> &in,
+	     const pngCompressSettings &settings = png_default_compress_settings);
+#endif /* PNG_COMPILE_ENCODER */
+#endif /* PNG_COMPILE_ZLIB */
+} /* namespace png */
+#endif /*PNG_COMPILE_CPP*/
+
+/*
+TODO:
+[.] test if there are no memory leaks or security exploits - done a lot but needs to be checked often
+[.] check compatibility with various compilers  - done but needs to be redone for every newer version
+[X] converting color to 16-bit per channel types
+[X] support color profile chunk types (but never let them touch RGB values by default)
+[ ] support all public PNG chunk types (almost done except sBIT, sPLT and hIST)
+[ ] make sure encoder generates no chunks with size > (2^31)-1
+[ ] partial decoding (stream processing)
+[X] let the "isFullyOpaque" function check color keys and transparent palettes too
+[X] better name for the variables "codes", "codesD", "codelengthcodes", "clcl" and "lldl"
+[ ] allow treating some errors like warnings, when image is recoverable (e.g. 69, 57, 58)
+[ ] make warnings like: oob palette, checksum fail, data after iend, wrong/unknown crit chunk, no null terminator in text, ...
+[ ] error messages with line numbers (and version)
+[ ] errors in state instead of as return code?
+[ ] new errors/warnings like suspiciously big decompressed ztxt or iccp chunk
+[ ] let the C++ wrapper catch exceptions coming from the standard library and return png error codes
+[ ] allow user to provide custom color conversion functions, e.g. for premultiplied alpha, padding bits or not, ...
+[ ] allow user to give data (void*) to custom allocator
+[X] provide alternatives for C library functions not present on some platforms (memcpy, ...)
+*/
+
+#endif /*PNG_H inclusion guard*/
diff --git a/libs/libgui/psf.c b/libs/libgui/psf.c
new file mode 100644
index 0000000..e28c2d7
--- /dev/null
+++ b/libs/libgui/psf.c
@@ -0,0 +1,60 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+// PSF parser
+
+#include <def.h>
+#include <libgui/gfx.h>
+#include <libgui/psf.h>
+#include <mem.h>
+#include <print.h>
+
+// Verifies the PSF magics
+// Returns the PSF version or 0
+static int psf_verify(char *data)
+{
+	struct psf1_header *header1 = (struct psf1_header *)data;
+	struct psf2_header *header2 = (struct psf2_header *)data;
+
+	if (header1->magic[0] == PSF1_MAGIC_0 && header1->magic[1] == PSF1_MAGIC_1)
+		return 1;
+	else if (header2->magic[0] == PSF2_MAGIC_0 && header2->magic[1] == PSF2_MAGIC_1 &&
+		 header2->magic[2] == PSF2_MAGIC_2 && header2->magic[3] == PSF2_MAGIC_3)
+		return 2;
+	else
+		return 0;
+}
+
+struct font *psf_parse(char *data)
+{
+	if (!data)
+		return NULL;
+
+	int version = psf_verify(data);
+
+	char *chars;
+	int height;
+	int width;
+	int char_size;
+
+	if (version == 1) {
+		chars = data + sizeof(struct psf1_header);
+		height = ((struct psf1_header *)data)->char_size;
+		width = 8;
+		char_size = ((struct psf1_header *)data)->char_size;
+	} else if (version == 2) {
+		chars = data + ((struct psf2_header *)data)->size;
+		height = ((struct psf2_header *)data)->height;
+		width = ((struct psf2_header *)data)->width;
+		char_size = ((struct psf2_header *)data)->char_size;
+	} else {
+		print("Unknown font!\n");
+		return NULL;
+	}
+
+	struct font *font = malloc(sizeof(*font));
+	font->chars = chars;
+	font->size.x = width;
+	font->size.y = height;
+	font->char_size = char_size;
+
+	return font;
+}
diff --git a/libs/libgui/psf.h b/libs/libgui/psf.h
new file mode 100644
index 0000000..63a3d1e
--- /dev/null
+++ b/libs/libgui/psf.h
@@ -0,0 +1,48 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+// PSF parser
+
+#ifndef PSF_H
+#define PSF_H
+
+#include <def.h>
+
+/**
+ * PSF version 1
+ */
+
+#define PSF1_MAGIC_0 0x36
+#define PSF1_MAGIC_1 0x04
+#define PSF1_MODE_256 0
+#define PSF1_MODE_512 1
+#define PSF1_MODE_256_UNICODE 2
+#define PSF1_MODE_512_UNICODE 3
+
+struct psf1_header {
+	u8 magic[2];
+	u8 mode;
+	u8 char_size;
+};
+
+/**
+ * PSF version 2
+ */
+
+#define PSF2_MAGIC_0 0x72
+#define PSF2_MAGIC_1 0xb5
+#define PSF2_MAGIC_2 0x4a
+#define PSF2_MAGIC_3 0x86
+
+struct psf2_header {
+	u8 magic[4];
+	u32 version;
+	u32 size;
+	u32 flags;
+	u32 glyph_count;
+	u32 char_size;
+	u32 height;
+	u32 width;
+};
+
+struct font *psf_parse(char *data);
+
+#endif
diff --git a/libs/libgui/vesa.h b/libs/libgui/vesa.h
new file mode 100644
index 0000000..892bd89
--- /dev/null
+++ b/libs/libgui/vesa.h
@@ -0,0 +1,46 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef VBE_H
+#define VBE_H
+
+#include <def.h>
+
+struct vbe {
+	u16 attributes;
+	u8 window_a;
+	u8 window_b;
+	u16 granularity;
+	u16 window_size;
+	u16 segment_a;
+	u16 segment_b;
+	u32 win_func_ptr;
+	u16 pitch;
+	u16 width;
+	u16 height;
+	u8 w_char;
+	u8 y_char;
+	u8 planes;
+	u8 bpp;
+	u8 banks;
+	u8 memory_model;
+	u8 bank_size;
+	u8 image_pages;
+	u8 reserved0;
+
+	u8 red_mask;
+	u8 red_position;
+	u8 green_mask;
+	u8 green_position;
+	u8 blue_mask;
+	u8 blue_position;
+	u8 reserved_mask;
+	u8 reserved_position;
+	u8 direct_color_attributes;
+
+	u8 *fb;
+	u32 off_screen_mem_off;
+	u16 off_screen_mem_size;
+	u8 reserved1[206];
+};
+
+#endif
diff --git a/libs/libnet/Makefile b/libs/libnet/Makefile
new file mode 100644
index 0000000..85c8fab
--- /dev/null
+++ b/libs/libnet/Makefile
@@ -0,0 +1,20 @@
+# MIT License, Copyright (c) 2020 Marvin Borner
+
+COBJS = dns.o http.o ip.o
+CC = ccache ../../cross/opt/bin/i686-elf-gcc
+LD = ccache ../../cross/opt/bin/i686-elf-ld
+AR = ccache ../../cross/opt/bin/i686-elf-ar
+
+CFLAGS = $(CFLAGS_DEFAULT) -I../ -I../libc/inc/ -Duserspace
+
+all: libtxt
+
+%.o: %.c
+	@$(CC) -c $(CFLAGS) $< -o $@
+
+libtxt: $(COBJS)
+	@mkdir -p ../../build/
+	@$(AR) rcs ../../build/libnet.a $+
+
+clean:
+	@find . -name "*.o" -type f -delete
diff --git a/libs/libnet/dns.c b/libs/libnet/dns.c
new file mode 100644
index 0000000..f20f33a
--- /dev/null
+++ b/libs/libnet/dns.c
@@ -0,0 +1,116 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+// TODO: Less magic, auto xld splitting
+// TODO: DNS cache
+
+#include <def.h>
+#include <libnet/net.h>
+#include <libnet/socket.h>
+#include <mem.h>
+#include <print.h>
+#include <random.h>
+#include <str.h>
+
+static u32 dns_ip_addr = ip(1, 1, 1, 1);
+
+struct dns_packet {
+	u16 qid;
+	u16 flags;
+	u16 questions;
+	u16 answers;
+	u16 authorities;
+	u16 additional;
+	u8 data[];
+} __attribute__((packed));
+
+static u32 part_count(const char *name)
+{
+	u32 cnt = 0;
+	for (u32 i = 0; i < strlen(name); i++) {
+		if (name[i] == '.')
+			cnt++;
+	}
+	return cnt + 1;
+}
+
+static u32 part_len(const char *name, u32 index)
+{
+	const char *data = name;
+
+	u32 cnt = 0;
+	for (u32 i = 0; i < strlen(name); i++) {
+		if (cnt == index) {
+			data += i;
+			break;
+		}
+
+		if (name[i] == '.')
+			cnt++;
+	}
+
+	for (cnt = 0; cnt < strlen(data); cnt++) {
+		if (data[cnt] == '.' || data[cnt] == '\0')
+			break;
+	}
+
+	return cnt;
+}
+
+static void dns_make_packet(struct dns_packet *packet, const char *name)
+{
+	packet->qid = htons(rand());
+	packet->flags = htons(0x0100); // Standard query
+	packet->questions = htons(1);
+	packet->answers = htons(0);
+	packet->authorities = htons(0);
+	packet->additional = htons(0);
+
+	u8 *data = packet->data;
+	u32 cnt = 0;
+	for (u32 i = 0; i < part_count(name) * 2; i += 2) {
+		data[cnt] = part_len(name, i / 2);
+		memcpy(&data[cnt + 1], &name[cnt], data[cnt]);
+		cnt += data[cnt] + 1;
+	}
+
+	packet->data[cnt + 0] = 0x00; // Name end
+	packet->data[cnt + 2] = 0x01; // A
+	packet->data[cnt + 4] = 0x01; // IN
+}
+
+static u32 dns_handle_packet(struct dns_packet *packet)
+{
+	u16 flags = htons(packet->flags);
+	u8 reply_code = flags & 0xf;
+	if (reply_code != DNS_NOERROR) {
+		printf("DNS error: %d\n", reply_code);
+		return 0;
+	}
+
+	u8 *start = &packet->data[1] + strlen((char *)&packet->data[1]);
+	printf("TTL of %s: %ds\n", &packet->data[1], (u32)start[14]);
+	u8 *ip = &start[17];
+	printf("IP: %d.%d.%d.%d\n", ip[0], ip[1], ip[2], ip[3]);
+	return ip(ip[0], ip[1], ip[2], ip[3]);
+}
+
+u32 dns_request(const char *name)
+{
+	struct socket *socket = net_open(S_UDP);
+	if (!socket || !net_connect(socket, dns_ip_addr, 53, NET_TIMEOUT) || part_count(name) == 1)
+		return 0;
+
+	u32 length = sizeof(struct dns_packet) + strlen(name) + part_count(name) + 4;
+	struct dns_packet *packet = malloc(length);
+	memset(packet, 0, length);
+	dns_make_packet(packet, name);
+	net_send(socket, packet, length);
+	free(packet);
+
+	u8 buf[1024] = { 0 };
+	int l = net_receive(socket, buf, 1024, NET_TIMEOUT);
+	net_close(socket);
+	if (l > 0)
+		return dns_handle_packet((void *)buf);
+	else
+		return 0;
+}
diff --git a/libs/libnet/dns.h b/libs/libnet/dns.h
new file mode 100644
index 0000000..d6673e6
--- /dev/null
+++ b/libs/libnet/dns.h
@@ -0,0 +1,21 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef DNS_H
+#define DNS_H
+
+#include <def.h>
+
+#define DNS_NOERROR 0
+#define DNS_FORMERR 1
+#define DNS_SERVFAIL 2
+#define DNS_NXDOMAIN 3
+#define DNS_NOTIMP 4
+#define DNS_REFUSED 5
+#define DNS_YXDOMAIN 6
+#define DNS_XRRSET 7
+#define DNS_NOTAUTH 8
+#define DNS_NOTZONE 9
+
+u32 dns_request(const char *name);
+
+#endif
diff --git a/libs/libnet/http.c b/libs/libnet/http.c
new file mode 100644
index 0000000..808d4c2
--- /dev/null
+++ b/libs/libnet/http.c
@@ -0,0 +1,134 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#include <assert.h>
+#include <conv.h>
+#include <def.h>
+#include <libnet/http.h>
+#include <libnet/net.h>
+#include <libnet/socket.h>
+#include <mem.h>
+#include <print.h>
+#include <str.h>
+
+char *http_data(char *r)
+{
+	char *h = NULL;
+	for (u32 i = 0; i < strlen(r); ++i) {
+		if (r[i] == '\r' && r[i + 1] == '\n' && r[i + 2] == '\r' && r[i + 3] == '\n') {
+			h = &r[i + 4];
+			break;
+		}
+	}
+	return h;
+}
+
+char *http_header_key(char *r, const char *key)
+{
+	char *res = NULL;
+	for (char *p = r; *p; p++) {
+		/* printf("'%c%c%c' vs '%c%c%c'\n", p[0], p[1], p[2], key[0], key[1], key[2]); */
+		if (strlen(p) >= strlen(key) && !memcmp(p, key, strlen(key))) {
+			char *start = p + strlen(key) + 2;
+			char *end = start;
+			for (; *end != '\n'; end++)
+				;
+			res = malloc(end - start);
+			memcpy(res, start, end - start - 1);
+			res[end - start] = '\0';
+			break;
+		}
+	}
+	return res;
+}
+
+u32 http_content_length(char *r)
+{
+	char *value = http_header_key(r, "Content-Length");
+	int length = value ? atoi(value) : 0;
+	free(value);
+	return length;
+}
+
+char *http_code(char *r)
+{
+	char *code = malloc(4);
+	char tmp = r[12];
+	r[12] = '\0';
+	memcpy(code, r + 9, 3);
+	code[3] = '\0';
+	r[12] = tmp;
+	return code;
+}
+
+u32 http_response(const char *http_code, u32 content_length, const char *data, char *resp)
+{
+	char buf[16] = { 0 };
+
+	resp[0] = '\0';
+	strcat(resp, "HTTP/1.1 ");
+	strcat(resp, buf);
+	strcat(resp, http_code);
+	strcat(resp, "\r\n");
+	strcat(resp, "Content-Length: ");
+	strcat(resp, conv_base(content_length, buf, 10, 0));
+	strcat(resp, "\r\n");
+	strcat(resp, "Server: Melvix\r\n");
+	strcat(resp, "Content-Type: text/html\r\n");
+	strcat(resp, "Connection: close\r\n\r\n");
+	u32 len = strlen(resp);
+	memcpy(&resp[len], data, content_length);
+
+	return len + content_length;
+}
+
+char *http_query_get(const char *url, const char *path)
+{
+	char *query = malloc(27 + strlen(url)); // TODO: Dynamic http length etc
+	query[0] = '\0';
+	strcat(query, "GET ");
+	if (path[0] != '/')
+		strcat(query, "/");
+	strcat(query, path);
+	strcat(query, " HTTP/1.1\r\nHost: ");
+	strcat(query, url);
+	strcat(query, "\r\n\r\n");
+	return query;
+}
+
+char *http_query_path(const char *query, char *path)
+{
+	u8 b = 0;
+	u32 s = 0;
+	u32 e = 0;
+
+	while (1) {
+		if (!b && query[e] == ' ' && query[++e]) {
+			s = e;
+			b = 1;
+		} else if (b && query[e] == ' ') {
+			strncat(path, &query[s], e - s);
+			break;
+		} else if (query[e] == '\0') {
+			return NULL;
+		}
+		e++;
+	}
+
+	return path;
+}
+
+char *http_receive(struct socket *socket)
+{
+	char buf[4096] = { 0 };
+	if (!net_receive(socket, buf, 4096, NET_TIMEOUT))
+		return NULL;
+
+	u32 length = http_content_length(buf);
+	char *data = malloc(strlen(buf) + length);
+	memcpy(data, buf, strlen(buf));
+	while (strlen(http_data(data)) != length) {
+		if (!net_receive(socket, data, length, NET_TIMEOUT))
+			break;
+	}
+	return data;
+}
diff --git a/libs/libnet/http.h b/libs/libnet/http.h
new file mode 100644
index 0000000..b9160ad
--- /dev/null
+++ b/libs/libnet/http.h
@@ -0,0 +1,82 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef HTTP_H
+#define HTTP_H
+
+#include <def.h>
+#include <libnet/socket.h>
+
+char *http_data(char *response);
+char *http_header_key(char *r, const char *key);
+u32 http_content_length(char *r);
+char *http_code(char *r);
+u32 http_response(const char *http_code, u32 content_length, const char *data, char *resp);
+char *http_query_get(const char *url, const char *path);
+char *http_query_path(const char *query, char *path);
+char *http_receive(struct socket *socket);
+
+#define HTTP_100 "100 Continue"
+#define HTTP_101 "101 Switching Protocol"
+#define HTTP_102 "102 Processing"
+#define HTTP_103 "103 Early Hints"
+#define HTTP_200 "200 OK"
+#define HTTP_201 "201 Created"
+#define HTTP_202 "202 Accepted"
+#define HTTP_203 "203 Non-Authoritative Information"
+#define HTTP_204 "204 No Content"
+#define HTTP_205 "205 Reset Content"
+#define HTTP_206 "206 Partial Content"
+#define HTTP_207 "207 Multi-Status"
+#define HTTP_208 "208 Already Reported"
+#define HTTP_226 "226 IM Used"
+#define HTTP_300 "300 Multiple Choice"
+#define HTTP_301 "301 Moved Permanently"
+#define HTTP_302 "302 Found"
+#define HTTP_303 "303 See Other"
+#define HTTP_304 "304 Not Modified"
+#define HTTP_305 "305 Use Proxy"
+#define HTTP_306 "306 Unused"
+#define HTTP_307 "307 Temporary Redirect"
+#define HTTP_308 "308 Permanent Redirect"
+#define HTTP_400 "400 Bad Request"
+#define HTTP_401 "401 Unauthorized"
+#define HTTP_402 "402 Payment Required"
+#define HTTP_403 "403 Forbidden"
+#define HTTP_404 "404 Not Found"
+#define HTTP_405 "405 Method Not Allowed"
+#define HTTP_406 "406 Not Acceptable"
+#define HTTP_407 "407 Proxy Authentication Required"
+#define HTTP_408 "408 Request Timeout"
+#define HTTP_409 "409 Conflict"
+#define HTTP_410 "410 Gone"
+#define HTTP_411 "411 Length Required"
+#define HTTP_412 "412 Precondition Failed"
+#define HTTP_413 "413 Payload Too Large"
+#define HTTP_414 "414 URI Too Long"
+#define HTTP_415 "415 Unsupported Media Type"
+#define HTTP_416 "416 Range Not Satisfiable"
+#define HTTP_417 "417 Expectation Failed"
+#define HTTP_418 "418 I'm a teapot"
+#define HTTP_421 "421 Misdirected Request"
+#define HTTP_422 "422 Unprocessable Entity"
+#define HTTP_423 "423 Locked"
+#define HTTP_424 "424 Failed Dependency"
+#define HTTP_425 "425 Too Early"
+#define HTTP_426 "426 Upgrade Required"
+#define HTTP_428 "428 Precondition Required"
+#define HTTP_429 "429 Too Many Request"
+#define HTTP_431 "431 Request Header Fields Too Large"
+#define HTTP_451 "451 Unavailable For Legal Reasons"
+#define HTTP_500 "500 Internal Server Error"
+#define HTTP_501 "501 Not Implemented"
+#define HTTP_502 "502 Bad Gateway"
+#define HTTP_503 "503 Service Unavailable"
+#define HTTP_504 "504 Gateway Timeout"
+#define HTTP_505 "505 HTTP Version Not Supported"
+#define HTTP_506 "506 Variant Also Negotiates"
+#define HTTP_507 "507 Insufficient Storage"
+#define HTTP_508 "508 Loop Detected"
+#define HTTP_510 "510 Not Extended"
+#define HTTP_511 "511 Network Authentication Required"
+
+#endif
diff --git a/libs/libnet/ip.c b/libs/libnet/ip.c
new file mode 100644
index 0000000..20e1a38
--- /dev/null
+++ b/libs/libnet/ip.c
@@ -0,0 +1,49 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+// Most net/ip handlers are in the kernel space
+// This is a userspace wrapper for some things
+
+#include <def.h>
+#include <libnet/net.h>
+#include <mem.h>
+#include <str.h>
+
+// Inspired by Paul Vixie, 1996
+int ip_pton(const char *src, u32 *dst)
+{
+	const char *end = src + strlen(src);
+	u8 tmp[4], *tp;
+	int ch = 0;
+	int saw_digit = 0;
+	int octets = 0;
+	*(tp = tmp) = 0;
+
+	while (src < end) {
+		ch = *src++;
+		if (ch >= '0' && ch <= '9') {
+			u32 new = *tp * 10 + (ch - '0');
+
+			if ((saw_digit && *tp == 0) || new > 255)
+				return 0;
+
+			*tp = new;
+			if (!saw_digit) {
+				if (++octets > 4)
+					return 0;
+				saw_digit = 1;
+			}
+		} else if (ch == '.' && saw_digit) {
+			if (octets == 4)
+				return 0;
+			*++tp = 0;
+			saw_digit = 0;
+		} else {
+			return 0;
+		}
+	}
+
+	if (octets < 4)
+		return 0;
+
+	*dst = htonl(*(u32 *)tmp);
+	return 1;
+}
diff --git a/libs/libnet/ip.h b/libs/libnet/ip.h
new file mode 100644
index 0000000..e06aba2
--- /dev/null
+++ b/libs/libnet/ip.h
@@ -0,0 +1,12 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+// Most net/ip handlers are in the kernel space
+// This is a userspace wrapper for some things
+
+#ifndef IP_H
+#define IP_H
+
+#include <def.h>
+
+int ip_pton(const char *src, u32 *dst);
+
+#endif
diff --git a/libs/libnet/net.h b/libs/libnet/net.h
new file mode 100644
index 0000000..4bfda2b
--- /dev/null
+++ b/libs/libnet/net.h
@@ -0,0 +1,73 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef NET_H
+#define NET_H
+
+#include <libnet/dns.h>
+#include <libnet/ip.h>
+#include <libnet/socket.h>
+#include <print.h>
+#include <sys.h>
+
+#define htonl(l)                                                                                   \
+	((((l)&0xff) << 24) | (((l)&0xff00) << 8) | (((l)&0xff0000) >> 8) |                        \
+	 (((l)&0xff000000) >> 24))
+#define htons(s) ((((s)&0xff) << 8) | (((s)&0xff00) >> 8))
+#define ntohl(l) htonl((l))
+#define ntohs(s) htons((s))
+#define ip(a, b, c, d)                                                                             \
+	((((a)&0xff) << 24) | (((b)&0xff) << 16) | (((c)&0xff) << 8) | (((d)&0xff) << 0))
+
+#define NET_TIMEOUT 2000
+#define NET_NO_TIMEOUT 0
+
+static inline int net_data_available(struct socket *socket)
+{
+	return (socket && socket->packets && socket->packets->head && socket->packets->head->data &&
+		((struct socket_data *)socket->packets->head->data)->length > 0);
+}
+
+#define net_open(type) (void *)sys1(SYS_NET_OPEN, (int)(type))
+#define net_send(socket, data, len) (void)sys3(SYS_NET_SEND, (int)(socket), (int)(data), (int)(len))
+
+static inline int net_connect(struct socket *socket, u32 ip_addr, u16 dst_port, u32 timeout)
+{
+	if (!socket || !ip_addr || !dst_port)
+		return 0;
+	sys3(SYS_NET_CONNECT, (int)(socket), (int)(ip_addr), (int)(dst_port));
+	int time = time();
+	while (socket->state != S_CONNECTED) {
+		if (socket->state == S_FAILED || (timeout && time() - time >= timeout))
+			return 0;
+		yield();
+	}
+	return 1;
+}
+
+static inline int net_close(struct socket *socket)
+{
+	if (!socket)
+		return 0;
+	int res = 0;
+	while (socket->state == S_CLOSING || !(res = (int)sys1(SYS_NET_CLOSE, (int)(socket))))
+		yield();
+	return res;
+}
+
+static inline int net_receive(struct socket *socket, void *buf, u32 len, u32 timeout)
+{
+	if (!socket || !buf || !len)
+		return 0;
+
+	int time = time();
+	while (!net_data_available(socket)) {
+		if (socket->state == S_FAILED || (timeout && time() - time >= timeout))
+			return 0;
+		yield();
+	}
+
+	// TODO: Only return once all segments are received?
+	return (int)sys3(SYS_NET_RECEIVE, (int)(socket), (int)(buf), (int)(len));
+}
+
+#endif
diff --git a/libs/libtxt/Makefile b/libs/libtxt/Makefile
new file mode 100644
index 0000000..93a7993
--- /dev/null
+++ b/libs/libtxt/Makefile
@@ -0,0 +1,20 @@
+# MIT License, Copyright (c) 2020 Marvin Borner
+
+COBJS = keymap.o #xml.o html.o
+CC = ccache ../../cross/opt/bin/i686-elf-gcc
+LD = ccache ../../cross/opt/bin/i686-elf-ld
+AR = ccache ../../cross/opt/bin/i686-elf-ar
+
+CFLAGS = $(CFLAGS_DEFAULT) -I../ -I../libc/inc/ -Duserspace
+
+all: libtxt
+
+%.o: %.c
+	@$(CC) -c $(CFLAGS) $< -o $@
+
+libtxt: $(COBJS)
+	@mkdir -p ../../build/
+	@$(AR) rcs ../../build/libtxt.a $+
+
+clean:
+	@find . -name "*.o" -type f -delete
diff --git a/libs/libtxt/html.c b/libs/libtxt/html.c
new file mode 100644
index 0000000..9295e17
--- /dev/null
+++ b/libs/libtxt/html.c
@@ -0,0 +1,238 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+// HTML parsing is mainly based on the XML parser
+
+#include <assert.h>
+#include <libgui/gui.h>
+#include <libtxt/html.h>
+#include <libtxt/xml.h>
+#include <list.h>
+#include <mem.h>
+#include <print.h>
+#include <str.h>
+
+static int is_self_closing(const char *tag)
+{
+	const char *void_elements[] = { "area",	 "base", "br",	 "col",	  "embed",  "hr",    "img",
+					"input", "link", "meta", "param", "source", "track", "wbr" };
+
+	for (u32 i = 0; i < sizeof(void_elements) / sizeof(void_elements[0]); ++i) {
+		if (!strcmp(void_elements[i], tag))
+			return 1;
+	}
+	return 0;
+}
+
+static char *normalize_tag_name(char *tag)
+{
+	for (char *p = tag; *p; ++p)
+		*p = *p > 0x40 && *p < 0x5b ? *p | 0x60 : *p;
+	return tag;
+}
+
+static struct dom *new_object(const char *tag, struct dom *parent)
+{
+	struct dom *object = malloc(sizeof(*object));
+	object->tag = strdup(tag);
+	object->parent = parent;
+	object->content = NULL;
+	object->children = list_new();
+	return object;
+}
+
+static void print_dom(struct dom *dom, u32 level)
+{
+	struct node *iterator = dom->children->head;
+	while (iterator != NULL) {
+		struct dom *obj = iterator->data;
+		for (u32 i = 0; i < level; i++)
+			print("\t");
+		printf("'%s': '%s'\n", obj->tag, obj->content ? obj->content : "");
+		if (obj->children->head)
+			print_dom(obj, level + 1);
+		iterator = iterator->next;
+	}
+}
+
+static struct dom *generate_dom(char *data, u32 length)
+{
+	struct xml_token tokens[128];
+	struct xml parser;
+	xml_init(&parser);
+	void *buffer = data;
+	enum xml_error error = xml_parse(&parser, buffer, length, tokens, 128);
+
+	if (error != XML_SUCCESS && error != XML_ERROR_BUFFERDRY) {
+		printf("XML parse error: %d\n", err);
+		printf("DATA: '%s'\n", data);
+		return NULL;
+	}
+
+	struct dom *root = new_object("root", NULL);
+	struct dom *current = root;
+
+	static char name[256] = { 0 };
+	for (u32 i = 0; i < parser.ntokens; i++) {
+		const struct xml_token *token = tokens + i;
+		name[0] = '\0';
+		switch (token->type) {
+		case XML_START_TAG:
+			memcpy(&name, (u8 *)buffer + token->start_pos,
+			       token->end_pos - token->start_pos);
+			name[token->end_pos - token->start_pos] = '\0';
+			normalize_tag_name(name);
+			current = new_object(name, current);
+			printf("Adding %s to %s\n", current->tag, current->parent->tag);
+			list_add(current->parent->children, current);
+			if (is_self_closing(name))
+				current = current->parent;
+			break;
+		case XML_END_TAG:
+			memcpy(&name, (u8 *)buffer + token->start_pos,
+			       token->end_pos - token->start_pos);
+			name[token->end_pos - token->start_pos] = '\0';
+			normalize_tag_name(name);
+
+			if (is_self_closing(name))
+				break;
+
+			if (!current || !current->parent || strcmp(name, current->tag))
+				return NULL;
+
+			current = current->parent;
+			break;
+		case XML_CHARACTER:
+			if (!current)
+				continue;
+
+			if (token->end_pos == token->start_pos + 2) {
+				const char *ptr = (char *)buffer + token->start_pos;
+
+				if (ptr[0] == '\r' && ptr[1] == '\n')
+					continue;
+			}
+			memcpy(&name, (u8 *)buffer + token->start_pos,
+			       token->end_pos - token->start_pos);
+			name[token->end_pos - token->start_pos] = '\0';
+			char *clean_name = name;
+			for (char *p = name; *p; p++) {
+				if (*p == ' ' || *p == '\n' || *p == '\r' || *p == '\t') {
+					clean_name++;
+				} else {
+					break;
+				}
+			}
+			if (!strlen(clean_name))
+				break;
+			current->content = strdup(clean_name);
+			break;
+		default:
+			break;
+		}
+
+		i += token->size;
+	}
+
+	assert(root);
+	print("GENERATED!\n");
+	print_dom(root, 0);
+	return root;
+}
+
+static struct html_element *new_html_element(struct element *container, struct dom *dom)
+{
+	struct html_element *elem = malloc(sizeof(*elem));
+	elem->x_offset = 0;
+	elem->y_offset = 0;
+	elem->dom = dom;
+	elem->obj = container;
+	return elem;
+}
+
+// TODO: Better structure?
+// TODO: Less code duplication (e.g. for headings)
+#define CMP(tag, tag_string) (!strcmp((tag), (tag_string)))
+static struct html_element *render_object(struct html_element *container, struct dom *dom)
+{
+	char *tag = dom->tag;
+
+	assert(container);
+	if (CMP(tag, "html")) {
+		struct element *obj =
+			gui_add_container(container->obj, 0, 0, 100, 100, COLOR_WHITE);
+		return new_html_element(obj, dom);
+	} else if (CMP(tag, "body")) {
+		struct element *obj =
+			gui_add_container(container->obj, 0, 0, 100, 100, COLOR_WHITE);
+		return new_html_element(obj, dom);
+	} else if (CMP(tag, "h1")) {
+		struct element *obj =
+			gui_add_label(container->obj, container->x_offset, container->y_offset,
+				      FONT_32, dom->content, COLOR_WHITE, COLOR_BLACK);
+		container->x_offset = 0;
+		container->y_offset += obj->ctx->size.y;
+		return new_html_element(obj, dom);
+	} else if (CMP(tag, "h2")) {
+		struct element *obj =
+			gui_add_label(container->obj, container->x_offset, container->y_offset,
+				      FONT_24, dom->content, COLOR_WHITE, COLOR_BLACK);
+		container->x_offset = 0;
+		container->y_offset += obj->ctx->size.y;
+		return new_html_element(obj, dom);
+	} else if (CMP(tag, "h3")) {
+		struct element *obj =
+			gui_add_label(container->obj, container->x_offset, container->y_offset,
+				      FONT_16, dom->content, COLOR_WHITE, COLOR_BLACK);
+		container->x_offset = 0;
+		container->y_offset += obj->ctx->size.y;
+		return new_html_element(obj, dom);
+	} else if (CMP(tag, "p")) {
+		struct element *obj =
+			gui_add_label(container->obj, container->x_offset, container->y_offset,
+				      FONT_16, dom->content, COLOR_WHITE, COLOR_BLACK);
+		container->x_offset = 0;
+		container->y_offset += obj->ctx->size.y;
+		return new_html_element(obj, dom);
+	} else if (CMP(tag, "hr")) {
+		gfx_draw_rectangle(container->obj->ctx,
+				   vec2(container->x_offset, container->y_offset),
+				   vec2(container->obj->ctx->size.x - container->x_offset,
+					container->y_offset + 2),
+				   COLOR_BLACK);
+		container->x_offset = 0;
+		container->y_offset += 2;
+		return container;
+	} else if (CMP(tag, "head") || CMP(tag, "meta") || CMP(tag, "title")) {
+		return container;
+	} else {
+		printf("UNKNOWN %s\n", tag);
+		if (dom->content && strlen(dom->content) > 0) {
+			struct element *obj = gui_add_label(container->obj, container->x_offset,
+							    container->y_offset, FONT_16,
+							    dom->content, COLOR_WHITE, COLOR_BLACK);
+			container->x_offset = 0;
+			container->y_offset += obj->ctx->size.y;
+			return new_html_element(obj, dom);
+		}
+		return container;
+	}
+}
+
+int html_render_dom(struct html_element *container, struct dom *dom)
+{
+	struct node *iterator = dom->children->head;
+	while (iterator != NULL) {
+		struct dom *obj = iterator->data;
+		struct html_element *rendered = render_object(container, obj);
+		if (obj->children->head && rendered)
+			html_render_dom(rendered, obj);
+		iterator = iterator->next;
+	}
+	return 1;
+}
+
+int html_render(struct element *container, char *data, u32 length)
+{
+	struct dom *dom = generate_dom(data, length);
+	struct html_element *obj = new_html_element(container, dom);
+	return dom && obj && html_render_dom(obj, dom);
+}
diff --git a/libs/libtxt/html.h b/libs/libtxt/html.h
new file mode 100644
index 0000000..c1b29f2
--- /dev/null
+++ b/libs/libtxt/html.h
@@ -0,0 +1,26 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+// HTML parsing is mainly based on the XML parser
+
+#ifndef HTML_H
+#define HTML_H
+
+#include <def.h>
+#include <list.h>
+
+struct dom {
+	char *tag;
+	char *content;
+	struct dom *parent;
+	struct list *children;
+};
+
+struct html_element {
+	u32 x_offset;
+	u32 y_offset;
+	struct dom *dom;
+	struct element *obj;
+};
+
+int html_render(struct element *container, char *data, u32 length);
+
+#endif
diff --git a/libs/libtxt/keymap.c b/libs/libtxt/keymap.c
new file mode 100644
index 0000000..67054f6
--- /dev/null
+++ b/libs/libtxt/keymap.c
@@ -0,0 +1,89 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#include <def.h>
+#include <libtxt/keymap.h>
+#include <mem.h>
+#include <print.h>
+#include <sys.h>
+
+static void map(struct keymap *keymap, int line, char ch, int index)
+{
+	switch (line) {
+	case 0:
+		keymap->map[index] = ch;
+		break;
+	case 1:
+		keymap->shift_map[index] = ch;
+		break;
+	case 2:
+		keymap->alt_map[index] = ch;
+		break;
+	default:
+		break;
+	}
+}
+
+// Very ugly code but it should work for now
+struct keymap *keymap_parse(const char *path)
+{
+	char *keymap_src = sread(path);
+	if (!keymap_src)
+		return NULL;
+	struct keymap *keymap = malloc(sizeof(*keymap));
+
+	int index = 0;
+	int ch_index = 0;
+	char ch;
+	int escaped = 0;
+	int line = 0;
+	int skip = 0;
+	while ((ch = keymap_src[index]) != '\0' || escaped) {
+		if (ch == ' ' && !skip) {
+			skip = 1;
+			index++;
+			continue;
+		} else if (ch == '\n') {
+			ch_index = 0;
+			index++;
+			line++;
+			continue;
+		} else if (ch == '\\' && !escaped) {
+			escaped = 1;
+			index++;
+			continue;
+		}
+		skip = 0;
+
+		if (ch == ' ' && !escaped)
+			ch = 0;
+
+		ch_index++;
+		if (escaped) {
+			switch (ch) {
+			case 'b':
+				ch = '\b';
+				break;
+			case 't':
+				ch = '\t';
+				break;
+			case 'n':
+				ch = '\n';
+				break;
+			case '\\':
+				ch = '\\';
+				break;
+			case ' ':
+				ch = ' ';
+				break;
+			default:
+				print("Unknown escape!\n");
+			}
+			escaped = 0;
+		}
+
+		map(keymap, line, ch, ch_index);
+		index++;
+	}
+
+	return keymap;
+}
diff --git a/libs/libtxt/keymap.h b/libs/libtxt/keymap.h
new file mode 100644
index 0000000..9f1966e
--- /dev/null
+++ b/libs/libtxt/keymap.h
@@ -0,0 +1,16 @@
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef KEYMAP_H
+#define KEYMAP_H
+
+#define KEYMAP_LENGTH 90
+
+struct keymap {
+	char map[KEYMAP_LENGTH];
+	char shift_map[KEYMAP_LENGTH];
+	char alt_map[KEYMAP_LENGTH];
+};
+
+struct keymap *keymap_parse(const char *path);
+
+#endif
diff --git a/libs/libtxt/xml.c b/libs/libtxt/xml.c
new file mode 100644
index 0000000..9a5fd76
--- /dev/null
+++ b/libs/libtxt/xml.c
@@ -0,0 +1,514 @@
+// Inspired by sxml (capmar)
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#include <assert.h>
+#include <libtxt/xml.h>
+#include <mem.h>
+#include <str.h>
+
+static const char *str_findchr(const char *start, const char *end, int c)
+{
+	const char *it;
+
+	assert(start <= end);
+	assert(0 <= c && c <= 127);
+
+	it = (const char *)memchr((void *)start, c, end - start);
+	return (it != NULL) ? it : end;
+}
+
+static const char *str_findstr(const char *start, const char *end, const char *needle)
+{
+	u32 needlelen;
+	int first;
+	assert(start <= end);
+
+	needlelen = strlen(needle);
+	assert(0 < needlelen);
+	first = (u8)needle[0];
+
+	while (start + needlelen <= end) {
+		const char *it =
+			(const char *)memchr((void *)start, first, (end - start) - (needlelen - 1));
+		if (it == NULL)
+			break;
+
+		if (memcmp(it, needle, needlelen) == 0)
+			return it;
+
+		start = it + 1;
+	}
+
+	return end;
+}
+
+static int str_starts_with(const char *start, const char *end, const char *prefix)
+{
+	long nbytes;
+	assert(start <= end);
+
+	nbytes = strlen(prefix);
+	if (end - start < nbytes)
+		return 0;
+
+	return memcmp(prefix, start, nbytes) == 0;
+}
+
+static int white_space(int c)
+{
+	switch (c) {
+	case ' ':
+	case '\t':
+	case '\r':
+	case '\n':
+		return 1;
+	}
+
+	return 0;
+}
+
+static int name_start_char(int c)
+{
+	if (0x80 <= c)
+		return 1;
+
+	return c == ':' || ('A' <= c && c <= 'Z') || c == '_' || ('a' <= c && c <= 'z');
+}
+
+static int name_char(int c)
+{
+	return name_start_char(c) || c == '-' || c == '.' || ('0' <= c && c <= '9') || c == 0xB7 ||
+	       (0x0300 <= c && c <= 0x036F) || (0x203F <= c && c <= 0x2040);
+}
+
+#define is_space(c) (white_space(((u8)(c))))
+#define is_alpha(c) (name_start_char(((u8)(c))))
+#define is_alnum(c) (name_char(((u8)(c))))
+
+static const char *str_ltrim(const char *start, const char *end)
+{
+	const char *it;
+	assert(start <= end);
+
+	for (it = start; it != end && is_space(*it); it++)
+		;
+
+	return it;
+}
+
+static const char *str_rtrim(const char *start, const char *end)
+{
+	const char *it, *prev;
+	assert(start <= end);
+
+	for (it = end; start != it; it = prev) {
+		prev = it - 1;
+		if (!is_space(*prev))
+			return it;
+	}
+
+	return start;
+}
+
+static const char *str_find_notalnum(const char *start, const char *end)
+{
+	const char *it;
+	assert(start <= end);
+
+	for (it = start; it != end && is_alnum(*it); it++)
+		;
+
+	return it;
+}
+
+#define buffer_from_offset(args, i) ((args)->buffer + (i))
+#define buffer_tooffset(args, ptr) (unsigned)((ptr) - (args)->buffer)
+#define buffer_getend(args) ((args)->buffer + (args)->buffer_length)
+
+static int state_push_token(struct xml *state, struct xml_args *args, enum xml_type type,
+			    const char *start, const char *end)
+{
+	struct xml_token *token;
+	u32 i;
+	if (args->num_tokens <= state->ntokens)
+		return 0;
+
+	i = state->ntokens++;
+	token = &args->tokens[i];
+	token->type = type;
+	token->start_pos = buffer_tooffset(args, start);
+	token->end_pos = buffer_tooffset(args, end);
+	token->size = 0;
+
+	switch (type) {
+	case XML_START_TAG:
+		state->tag_level++;
+		break;
+
+	case XML_END_TAG:
+		assert(0 < state->tag_level);
+		state->tag_level--;
+		break;
+
+	default:
+		break;
+	}
+
+	return 1;
+}
+
+static enum xml_error state_set_pos(struct xml *state, const struct xml_args *args, const char *ptr)
+{
+	state->buffer_pos = buffer_tooffset(args, ptr);
+	return (state->ntokens <= args->num_tokens) ? XML_SUCCESS : XML_ERROR_TOKENSFULL;
+}
+
+#define state_commit(dest, src) memcpy((dest), (src), sizeof(struct xml))
+
+#define XML_ERROR_STRICT XML_ERROR_INVALID
+#define ENTITY_MAXLEN 8
+#define MIN(a, b) ((a) < (b) ? (a) : (b))
+
+static enum xml_error parse_characters(struct xml *state, struct xml_args *args, const char *end)
+{
+	const char *start = buffer_from_offset(args, state->buffer_pos);
+	const char *limit, *colon, *ampr = str_findchr(start, end, '&');
+	assert(end <= buffer_getend(args));
+
+	if (ampr != start)
+		state_push_token(state, args, XML_CHARACTER, start, ampr);
+
+	if (ampr == end)
+		return state_set_pos(state, args, ampr);
+
+	limit = MIN(ampr + ENTITY_MAXLEN, end);
+	colon = str_findchr(ampr, limit, ';');
+	if (colon == limit)
+		return (limit == end) ? XML_ERROR_BUFFERDRY : XML_ERROR_INVALID;
+
+	start = colon + 1;
+	state_push_token(state, args, XML_CHARACTER, ampr, start);
+	return state_set_pos(state, args, start);
+}
+
+static enum xml_error parse_attrvalue(struct xml *state, struct xml_args *args, const char *end)
+{
+	while (buffer_from_offset(args, state->buffer_pos) != end) {
+		enum xml_error error = parse_characters(state, args, end);
+		if (error != XML_SUCCESS)
+			return error;
+	}
+
+	return XML_SUCCESS;
+}
+
+static enum xml_error parse_attributes(struct xml *state, struct xml_args *args)
+{
+	const char *start = buffer_from_offset(args, state->buffer_pos);
+	const char *end = buffer_getend(args);
+	const char *name = str_ltrim(start, end);
+
+	u32 ntokens = state->ntokens;
+	assert(0 < ntokens);
+
+	while (name != end && is_alpha(*name)) {
+		const char *eq, *space, *quot, *value;
+		enum xml_error error;
+
+		eq = str_findchr(name, end, '=');
+		if (eq == end)
+			return XML_ERROR_BUFFERDRY;
+
+		space = str_rtrim(name, eq);
+		state_push_token(state, args, XML_CDATA, name, space);
+
+		quot = str_ltrim(eq + 1, end);
+		if (quot == end)
+			return XML_ERROR_BUFFERDRY;
+		else if (*quot != '\'' && *quot != '"')
+			return XML_ERROR_INVALID;
+
+		value = quot + 1;
+		quot = str_findchr(value, end, *quot);
+		if (quot == end)
+			return XML_ERROR_BUFFERDRY;
+
+		state_set_pos(state, args, value);
+		error = parse_attrvalue(state, args, quot);
+		if (error != XML_SUCCESS)
+			return error;
+
+		name = str_ltrim(quot + 1, end);
+	}
+
+	{
+		struct xml_token *token = args->tokens + (ntokens - 1);
+		token->size = (u16)(state->ntokens - ntokens);
+	}
+
+	return state_set_pos(state, args, name);
+}
+
+#define TAG_LEN(str) (sizeof(str) - 1)
+#define TAG_MINSIZE 1
+
+static enum xml_error parse_comment(struct xml *state, struct xml_args *args)
+{
+	static const char START_TAG[] = "<!--";
+	static const char END_TAG[] = "-->";
+
+	const char *dash;
+	const char *start = buffer_from_offset(args, state->buffer_pos);
+	const char *end = buffer_getend(args);
+	if (end - start < (int)TAG_LEN(START_TAG))
+		return XML_ERROR_BUFFERDRY;
+
+	if (!str_starts_with(start, end, START_TAG))
+		return XML_ERROR_INVALID;
+
+	start += TAG_LEN(START_TAG);
+	dash = str_findstr(start, end, END_TAG);
+	if (dash == end)
+		return XML_ERROR_BUFFERDRY;
+
+	state_push_token(state, args, XML_COMMENT, start, dash);
+	return state_set_pos(state, args, dash + TAG_LEN(END_TAG));
+}
+
+static enum xml_error parse_instruction(struct xml *state, struct xml_args *args)
+{
+	static const char START_TAG[] = "<?";
+	static const char END_TAG[] = "?>";
+
+	enum xml_error error;
+	const char *quest, *space;
+	const char *start = buffer_from_offset(args, state->buffer_pos);
+	const char *end = buffer_getend(args);
+	assert(TAG_MINSIZE <= end - start);
+
+	if (!str_starts_with(start, end, START_TAG))
+		return XML_ERROR_INVALID;
+
+	start += TAG_LEN(START_TAG);
+	space = str_find_notalnum(start, end);
+	if (space == end)
+		return XML_ERROR_BUFFERDRY;
+
+	state_push_token(state, args, XML_INSTRUCTION, start, space);
+
+	state_set_pos(state, args, space);
+	error = parse_attributes(state, args);
+	if (error != XML_SUCCESS)
+		return error;
+
+	quest = buffer_from_offset(args, state->buffer_pos);
+	if (end - quest < (int)TAG_LEN(END_TAG))
+		return XML_ERROR_BUFFERDRY;
+
+	if (!str_starts_with(quest, end, END_TAG))
+		return XML_ERROR_INVALID;
+
+	return state_set_pos(state, args, quest + TAG_LEN(END_TAG));
+}
+
+static enum xml_error parse_doctype(struct xml *state, struct xml_args *args)
+{
+	static const char START_TAG[] = "<!DOCTYPE";
+	static const char END_TAG[] = ">";
+
+	const char *bracket;
+	const char *start = buffer_from_offset(args, state->buffer_pos);
+	const char *end = buffer_getend(args);
+	if (end - start < (int)TAG_LEN(START_TAG))
+		return XML_ERROR_BUFFERDRY;
+
+	if (!str_starts_with(start, end, START_TAG))
+		return XML_ERROR_BUFFERDRY;
+
+	start += TAG_LEN(START_TAG);
+	bracket = str_findstr(start, end, END_TAG);
+	if (bracket == end)
+		return XML_ERROR_BUFFERDRY;
+
+	state_push_token(state, args, XML_DOCTYPE, start, bracket);
+	return state_set_pos(state, args, bracket + TAG_LEN(END_TAG));
+}
+
+static enum xml_error parse_start(struct xml *state, struct xml_args *args)
+{
+	enum xml_error error;
+	const char *gt, *name, *space;
+	const char *start = buffer_from_offset(args, state->buffer_pos);
+	const char *end = buffer_getend(args);
+	assert(TAG_MINSIZE <= end - start);
+
+	if (!(start[0] == '<' && is_alpha(start[1])))
+		return XML_ERROR_INVALID;
+
+	name = start + 1;
+	space = str_find_notalnum(name, end);
+	if (space == end)
+		return XML_ERROR_BUFFERDRY;
+
+	state_push_token(state, args, XML_START_TAG, name, space);
+
+	state_set_pos(state, args, space);
+	error = parse_attributes(state, args);
+	if (error != XML_SUCCESS)
+		return error;
+
+	gt = buffer_from_offset(args, state->buffer_pos);
+
+	if (gt != end && *gt == '/') {
+		state_push_token(state, args, XML_END_TAG, name, space);
+		gt++;
+	}
+
+	if (gt == end)
+		return XML_ERROR_BUFFERDRY;
+
+	if (*gt != '>')
+		return XML_ERROR_INVALID;
+
+	return state_set_pos(state, args, gt + 1);
+}
+
+static enum xml_error parse_end(struct xml *state, struct xml_args *args)
+{
+	const char *gt, *space;
+	const char *start = buffer_from_offset(args, state->buffer_pos);
+	const char *end = buffer_getend(args);
+	assert(TAG_MINSIZE <= end - start);
+
+	if (!(str_starts_with(start, end, "</") && is_alpha(start[2])))
+		return XML_ERROR_INVALID;
+
+	start += 2;
+	gt = str_findchr(start, end, '>');
+	if (gt == end)
+		return XML_ERROR_BUFFERDRY;
+
+	space = str_find_notalnum(start, gt);
+	if (str_ltrim(space, gt) != gt)
+		return XML_ERROR_STRICT;
+
+	state_push_token(state, args, XML_END_TAG, start, space);
+	return state_set_pos(state, args, gt + 1);
+}
+
+static enum xml_error parse_cdata(struct xml *state, struct xml_args *args)
+{
+	static const char START_TAG[] = "<![CDATA[";
+	static const char END_TAG[] = "]]>";
+
+	const char *bracket;
+	const char *start = buffer_from_offset(args, state->buffer_pos);
+	const char *end = buffer_getend(args);
+	if (end - start < (int)TAG_LEN(START_TAG))
+		return XML_ERROR_BUFFERDRY;
+
+	if (!str_starts_with(start, end, START_TAG))
+		return XML_ERROR_INVALID;
+
+	start += TAG_LEN(START_TAG);
+	bracket = str_findstr(start, end, END_TAG);
+	if (bracket == end)
+		return XML_ERROR_BUFFERDRY;
+
+	state_push_token(state, args, XML_CDATA, start, bracket);
+	return state_set_pos(state, args, bracket + TAG_LEN(END_TAG));
+}
+
+void xml_init(struct xml *state)
+{
+	state->buffer_pos = 0;
+	state->ntokens = 0;
+	state->tag_level = 0;
+}
+
+#define ROOT_FOUND(state) (0 < (state)->tag_level)
+#define ROOT_PARSED(state) ((state)->tag_level == 0)
+
+enum xml_error xml_parse(struct xml *state, const char *buffer, u32 buffer_length,
+			 struct xml_token tokens[], u32 num_tokens)
+{
+	struct xml temp = *state;
+	const char *end = buffer + buffer_length;
+
+	struct xml_args args;
+	args.buffer = buffer;
+	args.buffer_length = buffer_length;
+	args.tokens = tokens;
+	args.num_tokens = num_tokens;
+
+	while (!ROOT_FOUND(&temp)) {
+		enum xml_error error;
+		const char *start = buffer_from_offset(&args, temp.buffer_pos);
+		const char *lt = str_ltrim(start, end);
+		state_set_pos(&temp, &args, lt);
+		state_commit(state, &temp);
+
+		if (end - lt < TAG_MINSIZE)
+			return XML_ERROR_BUFFERDRY;
+
+		if (*lt != '<')
+			return XML_ERROR_INVALID;
+
+		switch (lt[1]) {
+		case '?':
+			error = parse_instruction(&temp, &args);
+			break;
+		case '!':
+			error = (lt[2] == '-') ? parse_comment(&temp, &args) :
+						       parse_doctype(&temp, &args);
+			break;
+		default:
+			error = parse_start(&temp, &args);
+			break;
+		}
+
+		if (error != XML_SUCCESS)
+			return error;
+
+		state_commit(state, &temp);
+	}
+
+	while (!ROOT_PARSED(&temp)) {
+		enum xml_error error;
+		const char *start = buffer_from_offset(&args, temp.buffer_pos);
+		const char *lt = str_findchr(start, end, '<');
+		while (buffer_from_offset(&args, temp.buffer_pos) != lt) {
+			error = parse_characters(&temp, &args, lt);
+			if (error != XML_SUCCESS)
+				return error;
+
+			state_commit(state, &temp);
+		}
+
+		if (end - lt < TAG_MINSIZE)
+			return XML_ERROR_BUFFERDRY;
+
+		switch (lt[1]) {
+		case '?':
+			error = parse_instruction(&temp, &args);
+			break;
+		case '/':
+			error = parse_end(&temp, &args);
+			break;
+		case '!':
+			error = (lt[2] == '-') ? parse_comment(&temp, &args) :
+						       parse_cdata(&temp, &args);
+			break;
+		default:
+			error = parse_start(&temp, &args);
+			break;
+		}
+
+		if (error != XML_SUCCESS)
+			return error;
+
+		state_commit(state, &temp);
+	}
+
+	return XML_SUCCESS;
+}
diff --git a/libs/libtxt/xml.h b/libs/libtxt/xml.h
new file mode 100644
index 0000000..43a8005
--- /dev/null
+++ b/libs/libtxt/xml.h
@@ -0,0 +1,51 @@
+// Inspired by sxml (capmar)
+// MIT License, Copyright (c) 2020 Marvin Borner
+
+#ifndef XML_H
+#define XML_H
+
+#include <def.h>
+
+enum xml_error {
+	XML_ERROR_INVALID = -1,
+	XML_SUCCESS = 0,
+	XML_ERROR_BUFFERDRY = 1,
+	XML_ERROR_TOKENSFULL = 2
+};
+
+struct xml_token {
+	u16 type;
+	u16 size;
+	u32 start_pos;
+	u32 end_pos;
+};
+
+struct xml_args {
+	const char *buffer;
+	u32 buffer_length;
+	struct xml_token *tokens;
+	u32 num_tokens;
+};
+
+enum xml_type {
+	XML_START_TAG,
+	XML_END_TAG,
+	XML_CHARACTER,
+	XML_CDATA,
+	XML_INSTRUCTION,
+	XML_DOCTYPE,
+	XML_COMMENT
+};
+
+struct xml {
+	u32 buffer_pos;
+	u32 ntokens;
+	u32 tag_level;
+};
+
+enum xml_error xml_parse(struct xml *parser, const char *buffer, u32 buffer_length,
+			 struct xml_token *tokens, u32 num_tokens);
+
+void xml_init(struct xml *parser);
+
+#endif