Uncertain maybe-incorrect suggestion for impl_trait_overcaptures

[ehuss]

This is mostly a question for @compiler-errors (and really just an edition-guide issue). This is regarding the maybe-incorrect suggestion for impl_trait_overcaptures in the following specific scenario:

https://crater-reports.s3.amazonaws.com/pr-132712/try%23da25749bf5e6ba5ed862ff361c19afff2a986b2d/reg/srdf-0.1.52/log.txt

where the general code looks like:

impl<RDF> RDFParser<RDF>
where
    RDF: FocusRDF,
{
   //................
    pub fn instances_of(
        &self,
        object: &RDF::Term,
    ) -> Result<impl Iterator<Item = RDF::Subject>, RDFParseError> {
        let values = self
            .rdf
            .subjects_with_predicate_object(&Self::rdf_type(), object)
            .map_err(|e| RDFParseError::SRDFError { err: e.to_string() })?;
        Ok(values.into_iter())
    }
}

This generates the impl_trait_overcaptures with maybe-incorrect applicability.

warning: `impl Iterator<Item = <RDF as srdf_basic::SRDFBasic>::Subject>` will capture more lifetimes than possibly intended in edition 2024
  --> src/srdf_parser/rdf_parser.rs:65:17
   |
65 |     ) -> Result<impl Iterator<Item = RDF::Subject>, RDFParseError> {
   |                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   |
   = warning: this changes meaning in Rust 2024
   = note: for more information, see <https://doc.rust-lang.org/nightly/edition-guide/rust-2024/rpit-lifetime-capture.html>
note: specifically, these lifetimes are in scope but not mentioned in the type's bounds
  --> src/srdf_parser/rdf_parser.rs:63:9
   |
63 |         &self,
   |         ^
64 |         object: &RDF::Term,
   |                 ^
   = note: all lifetimes in scope will be captured by `impl Trait`s in edition 2024
help: use the precise capturing `use<...>` syntax to make the captures explicit
   |
65 |     ) -> Result<impl Iterator<Item = RDF::Subject> + use<RDF>, RDFParseError> {

The question I have is that I do not see this scenario described in https://doc.rust-lang.org/nightly/edition-guide/rust-2024/rpit-lifetime-capture.html, and thus left being uncertain about why it thinks it is maybe incorrect. I assume it is something to do with the associated types, but it wasn't immediately obvious how to minimize this scenario.

If this isn't already documented on https://doc.rust-lang.org/nightly/edition-guide/rust-2024/rpit-lifetime-capture.html about why it might be wrong to blindly follow the compiler, would it be possible to describe the scenario and what considerations the author should take before accepting the suggestion?

Meta

rustc --version --verbose:

rustc 1.84.0-nightly (143ce0920 2024-11-10)
binary: rustc
commit-hash: 143ce0920a2307b19831160a01f06f107610f1b2
commit-date: 2024-11-10
host: aarch64-unknown-linux-gnu
release: 1.84.0-nightly
LLVM version: 19.1.3

[compiler-errors]

I can't remember why I marked it as maybe-incorrect, though in this case we probably will have a problem since it'll suggest + use<'_> which is ambiguous in this case.

I don't have the time to majorly overhaul how we issue this suggestion to turn that into a 'a or something.

I think we can probably make it machine applicable except for when there's a '_ lifetime in the suggestion.

[compiler-errors]

Another place where it cannot be machine applicable is when there are APITs, since turning a impl Trait argument into a T: Trait changes the API of the function (i.e. makes it turbofishable).

[ehuss]

The suggestions it gives in every case is + use<RDF> which seems to at least compile (except one function, which I don't know how to fix). That's partially why I'm a little confused, because it doesn't seem to be related to any named or unnamed lifetimes and it's not an APIT, either.

Regarding the case with the ambiguous lifetime, I don't think I have seen that in the crater run, yet.

And I'm now seeing that #132816 changed it so that all impl_trait_overcaptures suggestions are maybe-incorrect. Am I interpreting that correctly? The PR doesn't mention that change.

That could be a significant hurdle for some projects since some will have dozens of RPITs to update.

If we are going with all maybe-incorrect, we'll at least need to update the documentation in https://doc.rust-lang.org/nightly/edition-guide/rust-2024/rpit-lifetime-capture.html#migrating-while-avoiding-overcapturing that you'll need to make all changes manually.

[compiler-errors]

Am I interpreting that correctly? The PR doesn't mention that change.

Yes, and that was an oversight. Sorry, I'm not perfect :)

[compiler-errors]

Also, I misspoke. There should never be a case where we capture an elided lifetime that would be incorrect to name in the + use<'_> suggestion. It should only be in the case that we have APITs.

[compiler-errors]

#132938 addresses this