fix: inline predicate_may_hold_fatal and remove expect call in it
#112163
Conversation
rustbot
added
S-waiting-on-review
| .find(|o| !selcx.predicate_may_hold_fatal(o)); | ||
| .find(|o| { | ||
| selcx.evaluate_root_obligation(o).map_or( | ||
| true, // Overflow has occurred, and treat the obligation as possibly holding. |
There was a problem hiding this comment.
An overflow is not an failing obligation. It is an ambiguity, and this is coherence, which must never treat an overflow as an error for soundness reasons.
There was a problem hiding this comment.
Understood. Thank you for explaining that.
There was a problem hiding this comment.
Specifically, an overflow represents a state of "we don't know whether or not a predicate holds because we didn't iterate deep enough" (where "deep enough" is sometimes infinitely deep -- as in this case -- or sometimes just because we set the recursion depth too low).
In coherence, we must never treat "we don't know" as "certainly does not hold", because in this code, "certainly does not hold" is used as proof that two impls may not overlap. The soundness of the program relies on coherence being correct.
Elsewhere in the compiler, outside of coherence, we can use overflows as "probably does not hold" because even if it did hold, we'd never be able to instantiate it during codegen. That's why other code, such as evaluate_obligation_no_overflow treats this case as an error. The worst it can do is affect inference or the choice of a method during method dispatch.
There was a problem hiding this comment.
Thank you so much for your clear and concise explanation. I now have a better understanding of this topic.
|
@bors r+ |
bors
added
S-waiting-on-bors
…rrors fix: inline `predicate_may_hold_fatal` and remove expect call in it - Fixes rust-lang#105231 - Discussion: rust-lang#111985 (comment) r? `@compiler-errors`
Rollup of 7 pull requests Successful merges: - rust-lang#112163 (fix: inline `predicate_may_hold_fatal` and remove expect call in it) - rust-lang#112399 (Instantiate closure synthetic substs in root universe) - rust-lang#112443 (Opportunistically resolve regions in new solver) - rust-lang#112535 (reorder attributes to make miri-test-libstd work again) - rust-lang#112579 (Fix building libstd documentation on FreeBSD.) - rust-lang#112639 (Fix `dead_code_cgu` computation) - rust-lang#112642 (Handle interpolated literal errors) r? `@ghost` `@rustbot` modify labels: rollup
r? @compiler-errors