diff options
Diffstat (limited to 'docs/wiki_src/technical')
| -rw-r--r-- | docs/wiki_src/technical/arithmetic.md | 0 | ||||
| -rw-r--r-- | docs/wiki_src/technical/performance.md | 24 | ||||
| -rw-r--r-- | docs/wiki_src/technical/reduction.md | 12 |
3 files changed, 36 insertions, 0 deletions
diff --git a/docs/wiki_src/technical/arithmetic.md b/docs/wiki_src/technical/arithmetic.md deleted file mode 100644 index e69de29..0000000 --- a/docs/wiki_src/technical/arithmetic.md +++ /dev/null diff --git a/docs/wiki_src/technical/performance.md b/docs/wiki_src/technical/performance.md new file mode 100644 index 0000000..05fd683 --- /dev/null +++ b/docs/wiki_src/technical/performance.md @@ -0,0 +1,24 @@ +# Performance + +The reduction of lambda calculus is (practically) not very efficient. As +an extension, bruijn also suffers from bad performance. + +Bruijn's interpreter works by substituting the entire program into one +huge lambda calculus term that will then get reduced by the +[reducer](reduction.md). As a result, many equivalent terms get +evaluated multiple times (although some of this get's solved by bruijn's +call-by-need reduction strategy). We currently work on a solution that +reduces all equivalent terms as one, which turns out is not actually +that trivial. Follow the [blog](https://text.marvinborner.de) to keep up +to date with the development. + +Aside from that, bruijn is still much faster than most of the hobby +programming languages based on pure lambda calculus. This is because of +the [RKNL reducer](reduction.md) and our choice of default [number/byte +encodings](../coding/data-structures.md). + +``` bruijn +> :import std/Math . +> :time fac (+30) +0.15 seconds +``` diff --git a/docs/wiki_src/technical/reduction.md b/docs/wiki_src/technical/reduction.md index e69de29..97bd78f 100644 --- a/docs/wiki_src/technical/reduction.md +++ b/docs/wiki_src/technical/reduction.md @@ -0,0 +1,12 @@ +# Reduction + +Bruijn uses the RKNL abstract machine reducer[^1]. RKNL uses the +call-by-need reduction strategy, similar to Haskell and other functional +programming languages. For you this means that you have efficient +support for [laziness](../coding/laziness.md) with generally less +redundant reductions. + +[^1]: [Biernacka, Małgorzata, Witold Charatonik, and Tomasz Drab. "A + simple and efficient implementation of strong call by need by an + abstract machine." Proceedings of the ACM on Programming Languages + 6.ICFP (2022): 109-136.](https://doi.org/10.5281/zenodo.6786796) |