`#[link_args]` was accidentally stabilized; were other attributes as well?

[pnkfelix]

A comment in the tracking issue for #[link_args] points out that the attribute was accidentally stabilized (quite a long time ago).

I think it will be easy to fix the particular case of #[link_args]; the infrastructure here is pretty nice. But it does raise the question: Were other attributes accidentally stabilized at the same time?

This issue has two goals: 1. correct the #[link_args] feature gate, and 2. double-check whether other attributes in the BUILTIN_ATTRIBUTES array need similar treatment.

[retep998]

In particular, try to check by using the attribute in places where it was not intended to be used. As an example #[link_args] is still unstable when used on an extern block (which is even tested), but if you do a crate level #![link_args] then it is stable.

[pnkfelix]

(So PR #43109 is intended to address item (1.) in the description, but I want to keep this issue open until we have double-checked (2.) and potentially added corresponding unit tests.)

[pnkfelix]

Another detail that I had not digested before about this bug: We are even emitting a lint claiming that the link_args attribute is unused when it appears in these other contexts that elude the current feature gate.

This led me to think that this bug was not so bad, since I inferred from the emission of the unused_attribute warning that the directive to link_args was being ignored in those other contexts.

However, as @retep998 states, the link_args directive is included in the linker invocation when link_args occurs elsewhere, despite the compiler claiming that the occurrence of the attribute is unused.

So the compiler is not merely failing to stop us from putting in a useless occurrence of an unstable attribue in such cases; it is in fact actively lying regarding the effect of the occurrence of the attribute.