Remove last traces of identifier hygiene from HIR

[petrochenkov]

e783a0a removed the last use of hygiene at post-resolve compilation stages, so we can avoid renaming during lowering to HIR and just keep original names.

r? @nrc

[petrochenkov]

I need to be sure clippy is okay with this change, while the compiler itself doesn't use information removed in this patch anymore, external lints still can use it.
Since this patch removes hygiene information from HIR names, lints running in the late lint pass and performing name-resolution-like functions requiring hygiene will have to be moved to the early lint pass, but then they will lose information from later stages (e.g. types) if they use it. Do such lints exist? Is it important to support hygiene in the compiler to be used exclusively by third party tools? I don't have a clear answer.

cc @Manishearth @mcarton

[Manishearth]

It looks like we already stopped using unhygenize. Might still break, though.

We rely on span expansion info now mostly.

[mcarton]

@Manishearth: We still use unhygienize 😛:

$ ag unhygienize
clippy_lints/src/shadow.rs
69:            bindings.push((ident.node.unhygienize(), ident.span))
123:            let name = ident.node.unhygienize();
330:    !path.global && path.segments.len() == 1 && path.segments[0].name.unhygienize() == name
340:        if self.name == name.unhygienize() {

clippy_lints/src/misc.rs
414:                   segment != segment.unhygienize() && // not in bang macro

From the comment, the last one looks like it could use in_macro. I won't have much time today to look at the other.

[petrochenkov]

unhygienize is not the problem, it can be just removed after this change. The question is about the need for hygienic comparisons in lints together with something like types.

[jseyfried]

Nice! LGTM.
Couldn't lints do hygienic comparisons by looking up the identifiers' resolutions in the def_map and comparing the resolutions?

[petrochenkov]

Couldn't lints do hygienic comparisons by looking up the identifiers' resolutions in the def_map and comparing the resolutions?

That's a good observation, all hygienic things (i.e. bindings, labels and some expression paths) have definitions in def_map, except for some function parameters, so yes, they can be used instead.

[mcarton]

LGTM, I have a working PR for Clippy based on this.

[jseyfried]

@bors r+

[bors]

📌 Commit f59afbc has been approved by jseyfried

[bors]

⌛ Testing commit f59afbc with merge 9604420...

[bors]

💔 Test failed - auto-mac-64-opt-rustbuild