fixed generics of crate node ICE

[enthropy7]

when note_and_explain_type_err is called during dyn-compatibility checking, the body_owner_def_id can be CRATE_DEF_ID because ObligationCause::dummy() uses it as a placeholder when there is no real body owner. calling generics_of on CRATE_DEF_ID caused an ICE because the crate root is a module and modules don't have generics. the fix guards all generics_of call sites by checking whether body_owner_def_id is CRATE_DEF_ID before attempting to look up generics, and skips the type parameter span labeling in that case.

i also added regression test with original ice reproduce

fixes #152335

[rustbot]

r? @fee1-dead

rustbot has assigned @fee1-dead.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: compiler
• compiler expanded to 69 candidates
• Random selection from 14 candidates

[fee1-dead]

I think overall this is not pretty fool-proof (what if someone else decides to call tcx.generics_of(body_owner_def_id) down the line, in the function or somewhere else?) and also requires quite some involved changes for an ICE that doesn't happen quite often.

I wonder whether we should simply return and not attempt to add additional information if the obligation cause is dummy.

I wonder whether we can try to make the original obligation cause not dummy.

View changes since this review

[fee1-dead]

Suggested change

	if !tcx.is_impl_trait_in_trait(proj.def_id) && body_generics.is_some() =>
	{
	let param = tcx.generics_of(body_owner_def_id).type_param(p, tcx);
	let param = body_generics.unwrap().type_param(p, tcx);
	if !tcx.is_impl_trait_in_trait(proj.def_id) && let Some(body_generics) = body_generics =>
	{
	let param = body_generics.type_param(p, tcx);

[fee1-dead]

This should be (ty::Param(p), _) | (_, ty::Param(p)) if let Some(generics) = body_generics =>

[rustbot]

Reminder, once the PR becomes ready for a review, use @rustbot ready.

[enthropy7]

ok so my first uproach won't work, i went with guarding generics_of call sites, as early return turned out too aggressive, it suppressed useful notes/helps for async blocks and closures that don't need generics at all.
guard in suggest_constraint for the same reason.

regarding making the obligation cause non-dummy: ObligationCause::dummy() comes from receiver_is_dispatchable in dyn_compatibility.rs, and there's no real body owner in that context, so I think guarding in error reporting is the right place for now.

[rustbot]

Some changes occurred in src/tools/compiletest

cc @jieyouxu

compiletest directives have been modified. Please add or update docs for the
new or modified directive in src/doc/rustc-dev-guide/.

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[enthropy7]

ok so, i completely changed approach. body_owner_def_id is now Option<DefId> throughout note_and_explain_type_err and all its helpers (suggest_constraint, expected_projection, point_at_associated_type). CRATE_DEF_ID check happens once at the call site in mod.rs, and the Option propagates naturally. compiler won't let anyone call generics_of without handling None first. in fact, during rebase a new generics_of(body_owner_def_id) call site had been added upstream, and it was immediately caught as a compile error rather than a silent ICE.

[enthropy7]

@rustbot ready

[fee1-dead]

@bors r+

[rust-bors]

📌 Commit f1a4ae0 has been approved by fee1-dead

It is now in the queue for this repository.