Fix duplicate diagnostics for min_rust_version_invalid_attr

[Gri-ffin]

part of #12379

Separate attributes check from parsing.

changelog: [min_rust_version_invalid_attr]: fix duplicate diagnostics

[DanielEScherzer]

So this could just be me not understanding clippy, but if I put the code from the test case in the playground, the output there has all of the same errors that the updated test has - so it doesn't look like things were being duplicated previously? Or is something already suppressing the duplicates on the playground?

View changes since this review

[Gri-ffin]

Yes, rustc will deduplicate the lints before outputting them, you can test locally by running cargo clippy -- -Zdeduplicate-diagnostics=no. You should be on nightly though.

[ada4a]

you can test locally by running cargo clippy -- -Zdeduplicate-diagnostics=no

I think cargo dev lint path/to/test.rs should work as well

[Gri-ffin]

-Zdeduplicate-diagnostics is yes by default so you need to pass no.

[ada4a]

But this means you'll need to manually add this line to any check_x where x can have attributes, no? So I think this is missing check_stmt, check_local, etc. Overall I'm unsure if this approach is very maintainable..

View changes since the review

[Gri-ffin]

I honestly was conflicted between this and moving the macro expansion of extract_msrv_attr! (meaning check_attributes and check_attributes_post) inside, which would duplicate code, plus I genuinely thought no one would ever put msrv attrs above stmts. But you are right, the second option is probably better instead of sprinkling checks here and there.

[Gri-ffin]

Hmm now that I'm reviewing the implementation, I think duplicated_attributes::check should also be moved.