allow_internal_unstable silently ignores typos in feature names

[bugadani]

While working on #77088 I wasted a lot of time looking for a bug in my code, when the issue was that I made a typo in the feature name in my test case.

I think it would make sense to check that the feature to be allowed actually exists.

[bugadani]

Soooo... I've tried doing this, but apparently, some features are not features 👀:

error: `allow_internal_unstable` expects feature names
    --> library/core/src/macros/mod.rs:1389:31
     |
1389 |     #[allow_internal_unstable(core_intrinsics, libstd_sys_internals)]
     |                               ^^^^^^^^^^^^^^^

error: `allow_internal_unstable` expects feature names
    --> library/core/src/macros/mod.rs:1389:48
     |
1389 |     #[allow_internal_unstable(core_intrinsics, libstd_sys_internals)]
     |                                                ^^^^^^^^^^^^^^^^^^^^

[est31]

@bugadani there are lib features and lang features. Only the second type is hardcoded in the compiler. For the first type, you need to check whether it's been declared in one of the upstream crates.

[bugadani]

As far as I can tell, lib features aren't available at the point of macro expansion. At least I can't see a way to list those, but let me know if I missed something.

This means that fixing this issue is far from trivial.

[varkor]

Somewhat related to #69399, as allow_internal_unstable doesn't currently accept library features anyway.

[bugadani]

Somewhat related to #69399, as allow_internal_unstable doesn't currently accept library features anyway.

Technically, all the attribute does is plain string comparison. In fact, allow_internal_unstable IS used to gate library features - the error excerpt I pasted above is from compiling std IIRC. One feature I distinctly remember used is core_panic. If these shouldn't be accepted, then there are several such uses to be cleaned up.

[varkor]

Ah, it only doesn't accept library functions in certain places. So it's less relevant than I thought.