#[cfg(...)] ignores namespace

[kennytm]

Repro:

fn main() {
    println!("           cfg(a) = {}", cfg!(a));
    println!("cfg(something::a) = {}", cfg!(something::a));
}

$ rustc --cfg 'a' 1.rs && ./1
           cfg(a) = true
cfg(something::a) = true

$ rustc --cfg 'something::a' 1.rs && ./1
           cfg(a) = true
cfg(something::a) = true

$ rustc --cfg 'what::a' 1.rs && ./1
           cfg(a) = true
cfg(something::a) = true

while I expected these should print true/false, false/true and false/false instead.

(Note: this also affects --print cfg which something::a is collapsed to a)

cc @petrochenkov @eddyb

---

$ rustc -vV
rustc 1.30.0-nightly (33b923fd4 2018-08-18)
binary: rustc
commit-hash: 33b923fd44c5c5925e635815fce68bdf1f98740f
commit-date: 2018-08-18
host: x86_64-apple-darwin
release: 1.30.0-nightly
LLVM version: 7.0

[eddyb]

What happens if you also have cfg(a), does that interact with cfg(something::a)?

[kennytm]

@eddyb Oh no, cfg(something::a) is the same as cfg(a). Edited the issue.

[kennytm]

This behavior exists since 1.27. On 1.26 or before, #[cfg(module::path)] is simply a syntax error.

[petrochenkov]

This looks like an unintended side-effect of #50030 that allowed arbitrary paths in meta-items (including nested ones), but fn name() for meta-items still returns the last segment ignoring everything else.
cc @flip1995 @topecongiro

Most of attributes, including cfg want nested meta-item paths to be single-segment.
So, looks like fn name should return Option<Name> and all its callers need to be taught to deal with None somehow - report an error, or pass it further, etc.

[eddyb]

I'd suggest having a flag on "meta-item parsing" to say whether it should allow paths at all - most of the builtin attributes don't need paths.

[petrochenkov]

The cfg issue specifically was fixed in #53893.