A few scanner microoptimizations

[414owen]

• Most strings won't even exist in the source after CPP
• Turned a few recursive functions into loops
• Turned a few ifs into switches
• un-inlined all functions, because it didn't improve performance. GCC knows what to inline anyway.

None of this has had any tangible impact on performance, so either the compiler already knew what to do, or the scanner is no longer the bottleneck.

We might avoid a few potential stackoverflows though.

[414owen]

Woops, test fail with -DDEBUG

[tek]

btw, what's the point of using static for functions?

[414owen]

btw, what's the point of using static for functions?

static functions are only visible within the current compilation unit, so the only functions our file exports are the tree_sitter_haskell* functions. It just lets the compiler do more optimizations, inline things that it otherwise wouldn't have been, do dead code analysis, that sort of thing.

[414owen]

I'd deleted an important line of code -_-

[wenkokke]

@414owen Any chance of getting rid of realloc? (Or conditionally getting rid of it with #ifdef __EMSCRIPTEN__?)

[414owen]

@wenkokke realloc seems pretty important. Let's see if upstream wants to include it: https://github.com/tree-sitter/tree-sitter/pull/1571

(I assume _realloc is the correct way to do this?)

[wenkokke]

@414owen _realloc is correct, yes :)

[414owen]

@tek this is safe to merge