Many feature gates are not checked in macros

[brson]

Feature gates are generally checked before expansion so feature gates are defeated by code inside macros.

[alexcrichton]

Notably, this compiles today when it shouldn't:

fn main() {
    let a = ('a', 1u);
    println!("{}", a.0);
}

[pnkfelix]

P-backcompat-lang, 1.0

[aturon]

Note that we may be able to use a technique similar to that for the stability lint: #17286

[japaric]

This is a duplicate of #12122

[alexcrichton]

While somewhat similar I don't think that this is a dupe of #12122. The definition of a macro should be subject to the feature gates in the defining crate, but the usage of a macro should be subject to the feature gates in the consumer crate.

#12122 indicates that we don't check macro definitions, and this indicates that we don't check macro invocations.

[cmr]

I think the best we can do right now is check after the expansion of macros, in the stable channel. Outside of the stable channel, this might cause some undesirable noise when the used features are supposed to be "internal implementation details" of a lib, but I think that sort of claim is baloney because the feature could disappear at any moment. It ought to require work on the macro user's part if it could stop working at any moment due to a compiler upgrade.