Add combine based benchmark to catch exponential trait selection

[Marwes]

Similar to #588 but shows a newer regression and without pulling in combine itself.

Reduced to run in ~1.5s locally, could be lowered or increased by adding more parsers to the chain.

[Marwes]

cc rust-lang/rust#68666

[nnethercote]

You said "newer regression" -- is there another regression I'm not aware of?

[Marwes]

rust-lang/rust#67454 and rust-lang/rust#68666 is the ones I refer to, but I probably should not have used regression here. The issues contain practically the same code, only the Parser trait has been altered slightly as one is based on combine-3 and the newer on combine-4.

So it has regressed between combine versions, but it hasn't regressed between rust versions. Sorry for the confusion!

[Mark-Simulacrum]

I'll leave this up to @nnethercote to decide on merging or not, I'm personally not sure but have no specific concerns.

[Marwes]

It might be better to merge this after rust-lang/rust#68666 has been fixed, in which case the number of "parser"s could be increased without penalty to the benchmarks.

[nnethercote]

It might be better to merge this after rust-lang/rust#68666 has been fixed

My understanding is that #68666 (and #67454) were both fixed by #67471, which reverted the changes from #66405 that caused the regression. Furthermore, the regression fix was backported to beta. (#68666 was not marked as fixed, though; I did that just now.)

So I think you could increase the number of parsers safely now. But 1.5s sounds fine, and as long as it demonstrates the original regression (i.e. the old code is drastically slower) I think the current number of parsers should be fine.

[Marwes]

So I think you could increase the number of parsers safely now. But 1.5s sounds fine, and as long as it demonstrates the original regression (i.e. the old code is drastically slower) I think the current number of parsers should be fine.

I can't, since this is a separate performance issue.

[rylev]

Going to close since this is a bit stale. Feel free to reopen if you'd like to continue working on this (or if you think this should be rebased and re-reviewed).