libc and test features are special-cased for feature validity checks

[varkor]

When implementing https://github.com/rust-lang/rust/pull/52644, where we emit errors when using unknown features, I had to special case two features: libc and test, because they weren't being picked up (and there were time constraints).

The special-casing is here: https://github.com/rust-lang/rust/blob/0aa8d0320266b5579428312095fe49af05ada972/src/librustc/middle/stability.rs#L840-L847

• libc is declared unlike any other feature: https://github.com/rust-lang/libc/blob/6bdbf5dc937459bd10e6bc4dc52b0adbd8cf4358/src/lib.rs#L92-L94 so it's not entirely surprising that the detection overlooks it, but I didn't find the root cause.
• I didn't manage to figure out what makes test special.

Ideally these should both not be special-cased. The relevant code for feature collection is in src/librustc/middle/lib_features.rs, so that's a good place to start looking.

[eddyb]

cfg_attr shouldn't make libc special. when we build it in-tree, we set feature = "stdbuild" (which is not set when using it from crates.io). All you need to do is make sure you're not touching feature-gates in libsyntax{,_ext}, only later on, in librustc.

[varkor]

All you need to do is make sure you're not touching feature-gates in libsyntax{,_ext}

As in: just ignore any warnings for these? Though that means we would avoid the special-casing, it could lead to a build up of unnecessary feature attributes in the compiler.

[eddyb]

@varkor I mean, don't check for feature-gating attributes on the AST, only on HIR.

[varkor]

It's been using either hir::itemlikevisit::ItemLikeVisitor or hir::intravisit::Visitor, so that's what it should be doing already, no?