bjoernager/rust
bjoernager/rust
1
Fork 0
Code Activity

Rollup merge of #123578 - lqd:regression-123275, r=compiler-errors

Browse source 
Restore `pred_known_to_hold_modulo_regions`

As requested by `@lcnr` in https://github.com/rust-lang/rust/issues/123275#issuecomment-2031885563 this PR restores `pred_known_to_hold_modulo_regions` to fix that "unexpected unsized tail" beta regression.

This also adds the reduced repro from https://github.com/rust-lang/rust/issues/123275#issuecomment-2041222851 as a sub-optimal test is better than no test at all, and it'll also cover #108721. It still ICEs on master, even though https://github.com/phlip9/rustc-warp-ice doesn't on nightly anymore, since https://github.com/rust-lang/rust/pull/122493.

Fixes #123275.

r? `@compiler-errors` but feel free to close if you'd rather have a better test instead
cc `@wesleywiser` who had signed up to do the revert

Will need a backport if we go with this PR: `@rustbot` label +beta-nominated
This commit is contained in:
Matthias Krüger 2024-04-08 22:06:23 +02:00 • committed by GitHub
commit 984767e500
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 291 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

50
compiler/rustc_trait_selection/src/traits/mod.rs
View file

@ -119,7 +119,9 @@ pub fn predicates_for_generics<'tcx>(
/// Determines whether the type `ty` is known to meet `bound` and
/// returns true if so. Returns false if `ty` either does not meet
/// `bound` or is not known to meet bound.
/// `bound` or is not known to meet bound (note that this is
/// conservative towards *no impl*, which is the opposite of the
/// `evaluate` methods).
pub fn type_known_to_meet_bound_modulo_regions<'tcx>(
infcx: &InferCtxt<'tcx>,
param_env: ty::ParamEnv<'tcx>,
@ -127,8 +129,50 @@ pub fn type_known_to_meet_bound_modulo_regions<'tcx>(
def_id: DefId,
) -> bool {
let trait_ref = ty::TraitRef::new(infcx.tcx, def_id, [ty]);
let obligation = Obligation::new(infcx.tcx, ObligationCause::dummy(), param_env, trait_ref);
infcx.predicate_must_hold_modulo_regions(&obligation)
pred_known_to_hold_modulo_regions(infcx, param_env, trait_ref)
}
/// FIXME(@lcnr): this function doesn't seem right and shouldn't exist?
///
/// Ping me on zulip if you want to use this method and need help with finding
/// an appropriate replacement.
#[instrument(level = "debug", skip(infcx, param_env, pred), ret)]
fn pred_known_to_hold_modulo_regions<'tcx>(
infcx: &InferCtxt<'tcx>,
param_env: ty::ParamEnv<'tcx>,
pred: impl ToPredicate<'tcx>,
) -> bool {
let obligation = Obligation::new(infcx.tcx, ObligationCause::dummy(), param_env, pred);
let result = infcx.evaluate_obligation_no_overflow(&obligation);
debug!(?result);
if result.must_apply_modulo_regions() {
true
} else if result.may_apply() {
// Sometimes obligations are ambiguous because the recursive evaluator
// is not smart enough, so we fall back to fulfillment when we're not certain
// that an obligation holds or not. Even still, we must make sure that
// the we do no inference in the process of checking this obligation.
let goal = infcx.resolve_vars_if_possible((obligation.predicate, obligation.param_env));
infcx.probe(|_| {
let ocx = ObligationCtxt::new(infcx);
ocx.register_obligation(obligation);
let errors = ocx.select_all_or_error();
match errors.as_slice() {
// Only known to hold if we did no inference.
[] => infcx.shallow_resolve(goal) == goal,
errors => {
debug!(?errors);
false
}
}
})
} else {
false
}
}
#[instrument(level = "debug", skip(tcx, elaborated_env))]