Resolve attributes during name resolution

[jonas-schievink]

Another step required to support procedural attribute macros. This will emit diagnostics when attributes can't be resolved to either a built-in attribute, or to something in macro namespace.

Procedural attribute macros refuse to expand, because they aren't yet fully implemented. Instead, they always get the unresolved-proc-macro diagnostic to let the user know.

Note that we don't yet ignore items with unresolved attributes on them. That's the next step required to support attribute macros.

Notes for users: This will result in false positive diagnostics when attributes are used that are consumed by procedural macros. The diagnostic can be disabled by adding "unresolved-attribute" to the rust-analyzer.diagnostics.disabled setting.

[flodiebold]

We should make a note about this in the next changelog to prevent at least a few of the inevitable bug reports 😬

[jonas-schievink]

Yeah, good idea. I've added a note to the PR description too.

[matklad]

Procedural attribute macros refuse to expand, because they aren't yet fully implemented. Instead, they always get the unresolved-proc-macro diagnostic to let the user know.

Hm, this feels less than ideal.... Perhaps we should make this sub-diagnostic experemental?

[matklad]

other than that, lgtm! (maybe only add .iter method to Attrs?)

[jonas-schievink]

Hm, this feels less than ideal.... Perhaps we should make this sub-diagnostic experemental?

Yeah I can do that. unresolved-proc-macro is already meant to be unintrusive and easy to ignore though, so that users are aware that the macro is being ignored, without having red lines all over the place.

[jonas-schievink]

And this seems to result in false positives for derive helpers like #[serde(rename = "edition")], seems like we need to register those... somewhere?