struct constructor of an alias ty under mgca produces error

[khyperia]

unsure whether to file this under min_generic_const_args, adt_const_params, inherent_associated_types, or something else ;-;

discovered when investigating how to progress with inherent_associated_types

//@ check-pass
#![feature(min_generic_const_args, adt_const_params)]
#![expect(incomplete_features)]

use std::marker::ConstParamTy;

#[derive(PartialEq, Eq, ConstParamTy)]
struct Struct {}

fn f<const S: Struct>() {}

trait Trait {
    type Assoc;
    fn g();
}

impl Trait for () {
    type Assoc = Struct;

    fn g() {
        let _ = Self::Assoc {}; // ok!
        f::<core::direct_const_arg!(Struct {})>(); // ok!
        f::<core::direct_const_arg!(Self::Assoc {})>(); // kaboom
    }
}

fn main() {}

error: struct expression with invalid base path
  --> src/main.rs:22:37
   |
22 |         f::<core::direct_const_arg!(Self::Assoc {})>();
   |                                     ^^^^^^^^^^^^^^

pardon the wonky setup, it can probably be minimized a bit, just wanting to avoid more_qualified_paths. I think (I am unsure) that this should compile.

relevant line of code:

rust/compiler/rustc_hir_analysis/src/hir_ty_lowering/mod.rs

Line 2679 in 8a2fbe3

let e = tcx.dcx().span_err(span, "struct expression with invalid base path");

relevant PR: #149114

Speculation on my part:

The easiest way to resolve this is to have ResolvedStructPath have its res: Result<Res, ErrorGuaranteed> field ripped out if at all possible. I haven't quite investigated why it exists, I thiiink it's only used in two places: this spot in mGCA direct arg lowering (hence me filing this issue), and fn_ctxt/checks.rs check_struct_path

rust/compiler/rustc_hir_typeck/src/fn_ctxt/checks.rs

Line 1049 in 8a2fbe3

let variant = match def {

... removing it would be real nice to make inherent_associated_types easier to implement.

Anyway, just filing an issue to track it for now, will fully investigate later.

[Nidhi-Gahlawat]

@rustbot claim

[khyperia]

ach, sorry, I was going to take a try looking into this at first! was mostly filing this as a self-issue to make sure I don't forget, sorry for not making that more clear. it's a potentially quite involved issue with no clear solution, not really a good intro issue, and a symptom of a deeper architectural problem rather than just an error to be resolved. I can maybe try to find something else you could work on if you'd like, though?

[Nidhi-Gahlawat]

No worries, thanks for clarifying! I’ll release this one.

[Nidhi-Gahlawat]

@rustbot release-assignment