Performance improvements

[natefaubion]

This PR improves the performance across the board by better optimizing the sharing on pattern matches. There were a lot of duplicate checks for things like == and <= in guards. In my tests, about a 2x speedup overall. Ideally, a smarter pattern matcher could perform these optimizations, but for now we can do it manually.

[paf31]

👍 Fantastic, thanks!

[jdegoes]

👍

[garyb]

I wasn't immediately sure why you did this at first, but I get it. Nice. I'll definitely do some work on this one day 😉

[ethul]

I just tried the 0.5.5 release and I am getting the following on psc 0.7.6.1:

Error found:
at /.../bower_components/purescript-maps/src/Data/Map.purs line 132, column 18 - line 132, column 18

  Unable to parse module:
  unexpected ,
  expecting ., {, number, char literal, string literal, "true", "false", [, _, "\\", qualifier, proper name, identifier, (, "case", "if", "do", "let", ::, operator or "of"

Was this a breaking change from 0.5.4?

[garyb]

Argh, it wasn't supposed to be. Looks like it's the multi-valued case, I thought that went in before 0.8.

@natefaubion perhaps we should use a record/array literal there for now, and then remove again when we make the release for psc 0.8?

[garyb]

I've just reverted it for now, I did a version using records and arrays but then was greeted by a bunch of partial pattern warnings. Perhaps there is something we can do pre-0.8 still though.

[natefaubion]

I thought that went in before 0.8.

I would have sworn that as well.

I think records should work (you're just doing useless allocations :/). Arrays will always give you the partial warnings. You could also use other functions, but that may be slower since you can't use mkFn, so you'd have the overhead of more currying.

[natefaubion]

Both lookup and insert can be rewritten with single case matching with slightly more verbose branching, but delete cannot. Swapping out the multi-case for objects actually makes it about twice as slow as the current delete. There are fewer perf wins for delete, so if we want to make these perf changes compatible for 0.7.6, then should probably just leave delete as is.

Do we want to update lookup and insert as a 0.5.x change, and then fix warnings + delete for a 0.6 change?

[garyb]

Sounds good to me, thanks for looking into this again.