False positive diagnostic in proc-macro

[laurmaedje]

On the Typst codebase, rust-analyzer raises a false positive diagnostic since a while. The concrete error message (typst: explicit parent type required (rust-analyzer macro-error)) is specific to the codebase since it stems from a proc macro, but the important part is that the command line compiler doesn't produce this diagnostic.

I'm not sure when exactly the diagnostic started, but it used to work and if I switch back to a ~2 months old rust-analyzer release, it works again.

rust-analyzer version: 0.3.2037-standalone (e9afba5 2024-07-14)

rustc version: rustc 1.79.0 (129f3b996 2024-06-10)

editor or extension: VSCode, extension version v0.3.2037

repository link: https://github.com/typst/typst

code snippet to reproduce: crates/typst/src/foundations/array.rs

[Jeremiah-Griffin]

Thanks for the working rust analyzer build.

I'm having a vaguely similar issue in #17615 so I'll start here.

[Veykril]

The problem seems to be that we fail to resolve the ty and scope macros in the typst codebase. r-a seems to get confused by the glob re-exports being shadowed by the same named modules

[Leedehai]

r-a seems to get confused by the glob re-exports being shadowed by the same named modules

Hi, am I right to elaborate your finding like this: crates/typst/src/foundations/ty.rs re-exports the ty and scope macros, but since crates/typst/src/foundations/mod.rs has mod ty; and mod scope;, these macro names, which are exported by pub use self::ty::*; in that mod.rs file, are shadowed?

Further, I also tried adding pub use typst_macros::{scope, ty}; directly in that mod.rs, after the module declarations, and the same issue remains. But if I put that re-export statement before the module declarations, the issue goes away.

[Veykril]

Yes, I don't know the exact problem here but the setup of the mod declarations + more visible re-exports is what seems to be tripping up rust-analyzer