Handle opaques in the new solver (take 2?) #111473

compiler-errors · 2023-05-11T17:46:19Z

Implement a new strategy for handling opaques in the new solver.

First, queries now carry both their defining anchor and the opaques that were defined in the inference context at the time of canonicalization. These are both used to pre-populate the inference context used by the canonical query.

Second, use the normalizes-to goal to handle opaque types in the new solver. This means that opaques are handled like projection aliases, but with their own rules:

Can only define opaques if they're "defining uses" (i.e. have unique params in all their substs).
Can only define opaques that are from the anchor.
Opaque type definitions are modulo regions. So that means Opaque<'?0r> = HiddenTy1 and Opaque<?'1r> = HiddenTy2 equate HiddenTy1 and HiddenTy2 instead of defining them as different opaque type keys.

aliemjay · 2023-05-12T03:07:24Z

It's not clear to me what is wrong with the current approach of having the opaque type relations in the external constraints. Do we have a write-up of this issue somewhere?

bors · 2023-05-15T18:11:09Z

☔ The latest upstream changes (presumably #111601) made this pull request unmergeable. Please resolve the merge conflicts.

lcnr · 2023-05-22T18:15:00Z

r? @lcnr

compiler-errors · 2023-05-22T18:51:29Z

compiler/rustc_trait_selection/src/solve/assembly/mod.rs

-                    if result.has_only_region_constraints() {
-                        return Ok(result);
-                    }
+                    //if result.has_only_region_constraints() {


Leave a comment why

lcnr

very nice, "a few nits", then r=me

compiler/rustc_borrowck/src/type_check/mod.rs

lcnr · 2023-05-22T18:24:23Z

compiler/rustc_borrowck/src/type_check/mod.rs

+                            ocx.infcx.tcx.fold_regions((*opaque_ty, *hidden_ty), |_, _| {
+                                ocx.infcx.next_nll_region_var(
+                                    NllRegionVariableOrigin::Existential { from_forall: false },
+                                )
+                            });


can you use renumber_regions here

Region renumberer is private to rustc_borrowck::renumber :^) I don't think it offers much benefit over just folding here.

compiler/rustc_borrowck/src/type_check/mod.rs

compiler/rustc_hir_typeck/src/writeback.rs

compiler/rustc_infer/src/infer/nll_relate/mod.rs

lcnr · 2023-05-22T19:20:24Z

compiler/rustc_trait_selection/src/solve/opaques.rs

+                }
+            },
+            Reveal::All => {
+                let actual = tcx.type_of(opaque_ty.def_id).subst(tcx, opaque_ty.substs);


try add a debug assertion that predefined opaques are empty

:s

I would need to make the inner infcx reachable from this module, or at least expose a new EvalCtxt helper here. I don't think it's worth just yet, so I'll leave a fixme.

compiler/rustc_trait_selection/src/solve/search_graph/mod.rs

lcnr · 2023-05-22T19:25:32Z

tests/ui/type-alias-impl-trait/different_defining_uses_never_type-2.rs

@@ -0,0 +1,15 @@
+#![feature(type_alias_impl_trait)]


also add test with type params for #110601 and close that issue, i think this should fix it?

We actually don't need to include this commit anymore. It's actually fine that our HIR typeck results have a Map<DefId, HiddenTy>, since we do the mapping from hidden ty to definition params during writeback:

rust/compiler/rustc_hir_typeck/src/writeback.rs

Lines 582 to 588 in 8b4b208

let hidden_type = hidden_type.remap_generic_params_to_declaration_params(

opaque_type_key,

self.fcx.infcx.tcx,

true,

);

self.typeck_results.concrete_opaque_types.insert(opaque_type_key.def_id, hidden_type);

So I'll just land #111853 independently of this PR.

rustbot · 2023-05-23T01:08:55Z

Some changes occurred to the core trait solver

cc @rust-lang/initiative-trait-system-refactor

compiler-errors · 2023-05-23T01:10:28Z

Alright this is ready for review once again, @lcnr. Left a bunch of comments and a few more modifications. Should be very easy to review now.

lcnr · 2023-05-24T10:31:11Z

compiler/rustc_borrowck/src/type_check/mod.rs

@@ -1041,6 +1048,57 @@ impl<'a, 'tcx> TypeChecker<'a, 'tcx> {
        checker
    }

+    pub(super) fn register_predefined_opaques_in_new_solver(&mut self) {


how is that fine (for TAIT)? 🤔 aren't the preregistered opaque types in the wrong environment now, they have the generic params of the opaque while in the context of the defining scope?

as a thought which also deals with rust-lang/trait-system-refactor-initiative#17

long term we could maybe end up with the following to replace DefiningAnchor:

enum DefineOpaques { // the current `DefiningAnchor::Bind` Yes(&'tcx List<DefId>), // We do not allow any new definitions for our opaques, // but when using an opaque with `DefId`, we instantiate the // inferred hidden type with these substs. // // This doesn't quite work because of lifetimes, but we can // figure something out there :grin: No(SomeMapCollection<DefId, EarlyBinder<Ty<'tcx>>), }

anyways, your PR works for RPIT so I wouldn't mind us landing this as long as we put a fixme here.

You're absolutely right lol. We shouldn't map RPITITs back to their declaration params during writeback if we're in the new trait solver. 🤦

I'll add a fixme anyways.

lcnr · 2023-05-24T10:35:12Z

compiler/rustc_trait_selection/src/solve/assembly/mod.rs

@@ -356,8 +354,11 @@ impl<'tcx> EvalCtxt<'_, 'tcx> {
                        }),
                    );
                    ecx.add_goal(normalizes_to_goal);
-                    let _ = ecx.try_evaluate_added_goals()?;
+                    let _ = ecx.try_evaluate_added_goals().inspect_err(|_| {


Suggested change

let _ = ecx.try_evaluate_added_goals().inspect_err(|_| {

let _certainty = ecx.try_evaluate_added_goals().inspect_err(|_| {

lcnr · 2023-05-24T11:03:03Z

I guess r=me with FIXME for #111473 (comment)

bors · 2023-05-25T01:24:23Z

☔ The latest upstream changes (presumably #111919) made this pull request unmergeable. Please resolve the merge conflicts.

…::normalize_opaque_type Co-authored-by: lcnr <rust@lcnr.de>

compiler-errors · 2023-05-25T03:43:30Z

@bors r=lcnr

Follow-ups:

Fix tests/ui/dyn-star/param-env-infer.rs cycle error (for the record, this error can be made to happen with the old solver too: https://play.rust-lang.org/?version=nightly&mode=debug&edition=2021&gist=e9a462d482b988d17268cf8e8947d1ec)
Evaluate borrowck type ops locally in new solver
Figure out how to, then fix predefined opaques coming from HIR typeck

bors · 2023-05-25T03:43:32Z

📌 Commit dd98198 has been approved by lcnr

It is now in the queue for this repository.

bors · 2023-05-25T08:41:57Z

⌛ Testing commit dd98198 with merge eb9da7b...

bors · 2023-05-25T11:21:08Z

☀️ Test successful - checks-actions
Approved by: lcnr
Pushing eb9da7b to master...

rust-timer · 2023-05-25T14:58:55Z

Finished benchmarking commit (eb9da7b): comparison URL.

Overall result: ❌ regressions - no action needed

@rustbot label: -perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	0.6%	[0.6%, 0.6%]	3
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

This benchmark run did not return any relevant results for this metric.

Cycles

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	3.1%	[3.1%, 3.1%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 646.729s -> 646.591s (-0.02%)

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. WG-trait-system-refactor The Rustc Trait System Refactor Initiative (-Znext-solver) labels May 11, 2023

This comment has been minimized.

Sign in to view

rust-cloud-vms bot force-pushed the opaques branch from 2baabb7 to de57bac Compare May 11, 2023 17:55

rust-cloud-vms bot force-pushed the opaques branch from de57bac to aae783e Compare May 18, 2023 23:08

This comment has been minimized.

Sign in to view

rustbot assigned lcnr May 22, 2023

compiler-errors commented May 22, 2023

View reviewed changes

lcnr reviewed May 22, 2023

View reviewed changes

rust-cloud-vms bot force-pushed the opaques branch 2 times, most recently from 7b9f1ee to d576970 Compare May 23, 2023 01:06

compiler-errors marked this pull request as ready for review May 23, 2023 01:08

This comment has been minimized.

Sign in to view

lcnr mentioned this pull request May 23, 2023

changed how we arbitrarily choose candidates for trait goals rust-lang/trait-system-refactor-initiative#11

Closed

lcnr reviewed May 24, 2023

View reviewed changes

compiler-errors and others added 7 commits May 25, 2023 03:19

Pull out logic from rust-lang#111131, plus some new logic in EvalCtxt…

4d80b80

…::normalize_opaque_type Co-authored-by: lcnr <rust@lcnr.de>

Move DefiningAnchor

a2d7ffc

Prepopulate opaques in canonical input

f3c9c21

Check that opaque is a defining use, prefer pre-defined opaques

e3f8bea

Pre-populate MIR with opaques, prefer subst-relate candidate

a2d806d

Strongly prefer alias and param-env bounds

97c11ff

Add InferCtxt::register_hidden_type_in_new_solver

980da66

Match on both reveal and solver mode at the same time

dd98198

rust-cloud-vms bot force-pushed the opaques branch from d576970 to dd98198 Compare May 25, 2023 03:42

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels May 25, 2023

bors added the merged-by-bors This PR was explicitly merged by bors. label May 25, 2023

bors merged commit eb9da7b into rust-lang:master May 25, 2023

rustbot added this to the 1.71.0 milestone May 25, 2023

This was referenced May 25, 2023

Add AliasKind::Weak for type aliases. #108860

Merged

-Z trait-solver=next: Deduplicate region constraints in query responses #111172

Closed

lcnr mentioned this pull request May 31, 2023

Tracking Issue for -Znext-solver #107374

Open

14 tasks

compiler-errors deleted the opaques branch August 11, 2023 20:12

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Handle opaques in the new solver (take 2?) #111473

Handle opaques in the new solver (take 2?) #111473

compiler-errors commented May 11, 2023 •

edited

Loading

This comment has been minimized.

aliemjay commented May 12, 2023

bors commented May 15, 2023

This comment has been minimized.

lcnr commented May 22, 2023

compiler-errors May 22, 2023

lcnr left a comment •

edited

Loading

lcnr May 22, 2023

compiler-errors May 23, 2023

lcnr May 22, 2023

compiler-errors May 23, 2023

lcnr May 22, 2023

compiler-errors May 23, 2023

rustbot commented May 23, 2023

compiler-errors commented May 23, 2023

This comment has been minimized.

lcnr May 24, 2023 •

edited

Loading

lcnr May 24, 2023

compiler-errors May 24, 2023

lcnr May 24, 2023

lcnr commented May 24, 2023

bors commented May 25, 2023

compiler-errors commented May 25, 2023 •

edited

Loading

bors commented May 25, 2023

bors commented May 25, 2023

bors commented May 25, 2023

rust-timer commented May 25, 2023

	let hidden_type = hidden_type.remap_generic_params_to_declaration_params(
	opaque_type_key,
	self.fcx.infcx.tcx,
	true,
	);

	self.typeck_results.concrete_opaque_types.insert(opaque_type_key.def_id, hidden_type);

	let _ = ecx.try_evaluate_added_goals().inspect_err(\|_\| {
	let _certainty = ecx.try_evaluate_added_goals().inspect_err(\|_\| {

Handle opaques in the new solver (take 2?) #111473

Handle opaques in the new solver (take 2?) #111473

Conversation

compiler-errors commented May 11, 2023 • edited Loading

This comment has been minimized.

aliemjay commented May 12, 2023

bors commented May 15, 2023

This comment has been minimized.

lcnr commented May 22, 2023

Choose a reason for hiding this comment

lcnr left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rustbot commented May 23, 2023

compiler-errors commented May 23, 2023

This comment has been minimized.

lcnr May 24, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

lcnr commented May 24, 2023

bors commented May 25, 2023

compiler-errors commented May 25, 2023 • edited Loading

bors commented May 25, 2023

bors commented May 25, 2023

bors commented May 25, 2023

rust-timer commented May 25, 2023

Overall result: ❌ regressions - no action needed

compiler-errors commented May 11, 2023 •

edited

Loading

lcnr left a comment •

edited

Loading

lcnr May 24, 2023 •

edited

Loading

compiler-errors commented May 25, 2023 •

edited

Loading