Would avoiding the per-token mapper slice allocation be worthwhile?

[srinathshrestha]

hey, i in the mapping lexer , closure in Build creates a temporary combined []Mapper for each token with applicable mappers. i tried benchmarking the actual wrapper locally, with an allocation-free underlying lexer and two identity mappers. that setup gives 16 B/op and 1 alloc/op.

i was thinking we could avoid the temporary slice and just loop over the global mappers first, then the token-specific ones. so it'd be just 2 short loops instead of one. so what i'm asking is , will removing that allocation be worth it, any reasons to keep the combined list? wanted to ask before going further.

[alecthomas]

If the change is relatively minimal and the code is still clear, no behaviour changes, and there are alloc improvements, I'm open to it.

[srinathshrestha]

sounds good ! i'll prepare a small PR with tests and before/after allocation benchmarks.

[srinathshrestha]

hey, i opened the PR here: #483 as mentioned above. take a look at your convenience.