Move placeholder error handling to before region inference

[amandasystems]

This PR breaks out yet another part of #130227. It prepares for the removal of precise placeholder tracking by colocating detecting placeholder errors for higher-ranked relations (placeholders outliving existentials that can't name them, other placeholders) with the preprocessing step that adds outlives-static constraints. This also relieves that step from the burden of ensuring these errors are detected via propagation, and fully splits the logic into one part that rewrites the constraint graph to add edges to 'static and one that detects hard errors.

For stepping-on-toes reasons it is currently based on #140737, but can be rebased on regular master and merged separately nope that's a terrible idea actually, it relies on some of the tracking introduced there.

This step also makes region errors more fine-grained, which is useful to simplify the error reporting pipeline, which is currently based on a very brittle method of reconstructing an error event after the fact to figure out what went wrong. Doing that is left as future work, though.

This changes the logic of error reporting slightly to not emit errors for placeholder/higher kinded constraints if there are other borrowck errors. This somewhat lessens the flood of errors and was described as if anything an improvement by @nikomatsakis in informal review.

r? lcnr

Fixes: #146467

[amandasystems]

Wow, a merge conflict with upstream for a commit I JUST PUSHED, that's record time!

[lcnr]

@rustbot blocked

[amandasystems]

@rustbot ready

[amandasystems]

I hate this, but can't find a way around it. My gut feeling says it's another case of an order-dependent extraction procedure somewhere. For what it's worth, it wasn't in the earlier version of this code but it seems to have appeared after the recent changes to the base commit. Presumably the difference is from selecting smallest placeholder by rvid, etc. Help very much appreciated!

[amandasystems]

I think the "checked sub" part is what really broke me here. Why are universes being adjusted? What's a base universe? And why can the subtraction sometimes overflow?!?!

[amandasystems]

See also bound_region_errors, line 502

[amandasystems]

For the record, these are the only reads of base_universe on CanonicalTypeOpDeeplyNormalizeGoal, CanonicalTypeOpAscribeUserTypeGoal and InstantiateOpaqueType.

[amandasystems]

The job pr-check-2 failed! Check out the build log: (web) (plain enhanced) (plain)
Click to see the possible cause of the failure (guessed by this bot)

 95 | |         constraints.clone(),
 96 | |         &universal_region_relations,
 97 | |         infcx,
 98 | |     );
    | |_____- argument #4 of type `&mut RegionErrors<'_>` is missing
    |
note: function defined here
   --> compiler/rustc_borrowck/src/handle_placeholders.rs:306:15
    |
306 | pub(crate) fn compute_sccs_applying_placeholder_outlives_constraints<'tcx>(
    |               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...
310 |     errors_buffer: &mut RegionErrors<'tcx>,
    |     --------------------------------------
help: provide the argument
    |
 94 -     let lowered_constraints = compute_sccs_applying_placeholder_outlives_constraints(
 95 -         constraints.clone(),
 96 -         &universal_region_relations,
 97 -         infcx,
 98 -     );
 94 +     let lowered_constraints = compute_sccs_applying_placeholder_outlives_constraints(constraints.clone(), &universal_region_relations, infcx, /* &mut RegionErrors<'_> */);
    |

For more information about this error, try `rustc --explain E0061`.
[RUSTC-TIMING] rustc_borrowck test:false 4.478
error: could not compile `rustc_borrowck` (lib) due to 1 previous error

I don't understand where these errors come from. On my branch every single commit passes the ./x check command. What's emitting these? The error refers to a line that doesn't exist in a file that doesn't look like that during any commit in the PR.

[rustbot]

This PR was rebased onto a different master 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.

[bors]

☔ The latest upstream changes (presumably #146727) made this pull request unmergeable. Please resolve the merge conflicts.