Auto merge of rust-lang#133566 - lcnr:fast-reject-perf, r=<try>

fast-reject: add cache

slightly modified version of rust-lang#133524

I originally wanted to simply bail after recursion for a certain amount of times, however, looking at the number of steps taken while compiling different crates we get the following results[^1]:

typenum
```rust
1098842 counts
(  1)   670511 (61.0%, 61.0%): dropping after 1
(  2)   358785 (32.7%, 93.7%): dropping after 0
(  3)    25191 ( 2.3%, 96.0%): dropping after 2
(  4)    10912 ( 1.0%, 97.0%): dropping after 4
(  5)     6461 ( 0.6%, 97.5%): dropping after 3
(  6)     5239 ( 0.5%, 98.0%): dropping after 5
(  7)     2528 ( 0.2%, 98.3%): dropping after 8
(  8)     2188 ( 0.2%, 98.5%): dropping after 1094
(  9)     2097 ( 0.2%, 98.6%): dropping after 6
( 10)     1179 ( 0.1%, 98.7%): dropping after 34
( 11)     1148 ( 0.1%, 98.9%): dropping after 7
( 12)      822 ( 0.1%, 98.9%): dropping after 10
```
bitmaps
```rust
533346 counts
(  1)   526166 (98.7%, 98.7%): dropping after 1
(  2)     4562 ( 0.9%, 99.5%): dropping after 0
(  3)     2072 ( 0.4%, 99.9%): dropping after 1024
(  4)      305 ( 0.1%,100.0%): dropping after 2
(  5)      106 ( 0.0%,100.0%): dropping after 4
(  6)       30 ( 0.0%,100.0%): dropping after 8
(  7)       18 ( 0.0%,100.0%): dropping after 3
(  8)       17 ( 0.0%,100.0%): dropping after 44
(  9)       15 ( 0.0%,100.0%): dropping after 168
( 10)        8 ( 0.0%,100.0%): dropping after 14
( 11)        7 ( 0.0%,100.0%): dropping after 13
( 12)        7 ( 0.0%,100.0%): dropping after 24
```
stage 2 compiler is mostly trivial, but has a few cases where we get >5000
```rust
12987156 counts
(  1)  9280476 (71.5%, 71.5%): dropping after 0
(  2)  2277841 (17.5%, 89.0%): dropping after 1
(  3)   724888 ( 5.6%, 94.6%): dropping after 2
(  4)   204005 ( 1.6%, 96.2%): dropping after 4
(  5)   146537 ( 1.1%, 97.3%): dropping after 3
(  6)    64287 ( 0.5%, 97.8%): dropping after 5
(  7)    43938 ( 0.3%, 98.1%): dropping after 6
(  8)    43758 ( 0.3%, 98.4%): dropping after 8
(  9)    27220 ( 0.2%, 98.7%): dropping after 7
( 10)    17374 ( 0.1%, 98.8%): dropping after 9
( 11)    16015 ( 0.1%, 98.9%): dropping after 10
( 12)    12855 ( 0.1%, 99.0%): dropping after 12
( 13)    10494 ( 0.1%, 99.1%): dropping after 11
( 14)     7553 ( 0.1%, 99.2%): dropping after 14
```

Given that we have crates which frequently rely on fairly deep recursion, actually using a cache seems better than using an arbitrary cutoff here. Having an impl which is large enough to trigger a cutoff instead of getting rejected noticeably impacts perf, so just using a cache in these cases seems better to me. Does not matter too much in the end, we only have to make sure we don't regress crates which don't recurse deeply.

[^1]: i've incremented a counter in the place I now call `if cache.get(&(lhs, rhs))` and then printed it on drop
r? `@compiler-errors`

compiler/rustc_trait_selection/src/traits/coherence.rs

@@ -98,7 +98,7 @@ pub fn overlapping_impls(
     // Before doing expensive operations like entering an inference context, do
     // a quick check via fast_reject to tell if the impl headers could possibly
     // unify.
-    let drcx = DeepRejectCtxt::relate_infer_infer(tcx);
+    let mut drcx = DeepRejectCtxt::relate_infer_infer(tcx);
     let impl1_ref = tcx.impl_trait_ref(impl1_def_id);
     let impl2_ref = tcx.impl_trait_ref(impl2_def_id);
     let may_overlap = match (impl1_ref, impl2_ref) {

compiler/rustc_trait_selection/src/traits/effects.rs

@@ -79,7 +79,7 @@ fn evaluate_host_effect_from_bounds<'tcx>(
     obligation: &HostEffectObligation<'tcx>,
 ) -> Result<ThinVec<PredicateObligation<'tcx>>, EvaluationFailure> {
     let infcx = selcx.infcx;
-    let drcx = DeepRejectCtxt::relate_rigid_rigid(selcx.tcx());
+    let mut drcx = DeepRejectCtxt::relate_rigid_rigid(selcx.tcx());
     let mut candidate = None;
 
     for predicate in obligation.param_env.caller_bounds() {

compiler/rustc_trait_selection/src/traits/project.rs

@@ -866,7 +866,7 @@ fn assemble_candidates_from_predicates<'cx, 'tcx>(
     potentially_unnormalized_candidates: bool,
 ) {
     let infcx = selcx.infcx;
-    let drcx = DeepRejectCtxt::relate_rigid_rigid(selcx.tcx());
+    let mut drcx = DeepRejectCtxt::relate_rigid_rigid(selcx.tcx());
     for predicate in env_predicates {
         let bound_predicate = predicate.kind();
         if let ty::ClauseKind::Projection(data) = predicate.kind().skip_binder() {

compiler/rustc_trait_selection/src/traits/select/candidate_assembly.rs

@@ -232,7 +232,7 @@ impl<'cx, 'tcx> SelectionContext<'cx, 'tcx> {
             .filter(|p| p.def_id() == stack.obligation.predicate.def_id())
             .filter(|p| p.polarity() == stack.obligation.predicate.polarity());
 
-        let drcx = DeepRejectCtxt::relate_rigid_rigid(self.tcx());
+        let mut drcx = DeepRejectCtxt::relate_rigid_rigid(self.tcx());
         let obligation_args = stack.obligation.predicate.skip_binder().trait_ref.args;
         // Keep only those bounds which may apply, and propagate overflow if it occurs.
         for bound in bounds {
@@ -548,7 +548,7 @@ impl<'cx, 'tcx> SelectionContext<'cx, 'tcx> {
         obligation: &PolyTraitObligation<'tcx>,
         candidates: &mut SelectionCandidateSet<'tcx>,
     ) {
-        let drcx = DeepRejectCtxt::relate_rigid_infer(self.tcx());
+        let mut drcx = DeepRejectCtxt::relate_rigid_infer(self.tcx());
         let obligation_args = obligation.predicate.skip_binder().trait_ref.args;
         self.tcx().for_each_relevant_impl(
             obligation.predicate.def_id(),

compiler/rustc_type_ir/src/fast_reject.rs

@@ -11,6 +11,7 @@ use rustc_data_structures::stable_hasher::{HashStable, StableHasher, ToStableHas
 #[cfg(feature = "nightly")]
 use rustc_macros::{HashStable_NoContext, TyDecodable, TyEncodable};
 
+use crate::data_structures::DelayedSet;
 use crate::inherent::*;
 use crate::visit::TypeVisitableExt as _;
 use crate::{self as ty, Interner};
@@ -181,41 +182,44 @@ impl<DefId> SimplifiedType<DefId> {
 /// We also use this function during coherence. For coherence the
 /// impls only have to overlap for some value, so we treat parameters
 /// on both sides like inference variables.
-#[derive(Debug, Clone, Copy)]
+#[derive(Debug)]
 pub struct DeepRejectCtxt<
     I: Interner,
     const INSTANTIATE_LHS_WITH_INFER: bool,
     const INSTANTIATE_RHS_WITH_INFER: bool,
 > {
     _interner: PhantomData<I>,
+    /// We use a cache here as exponentially large - but self-similar - types otherwise
+    /// cause hangs, e.g. when compiling itertools with the `-Znext-solver`.
+    cache: DelayedSet<(I::Ty, I::Ty)>,
 }
 
 impl<I: Interner> DeepRejectCtxt<I, false, false> {
     /// Treat parameters in both the lhs and the rhs as rigid.
     pub fn relate_rigid_rigid(_interner: I) -> DeepRejectCtxt<I, false, false> {
-        DeepRejectCtxt { _interner: PhantomData }
+        DeepRejectCtxt { _interner: PhantomData, cache: Default::default() }
     }
 }
 
 impl<I: Interner> DeepRejectCtxt<I, true, true> {
     /// Treat parameters in both the lhs and the rhs as infer vars.
     pub fn relate_infer_infer(_interner: I) -> DeepRejectCtxt<I, true, true> {
-        DeepRejectCtxt { _interner: PhantomData }
+        DeepRejectCtxt { _interner: PhantomData, cache: Default::default() }
     }
 }
 
 impl<I: Interner> DeepRejectCtxt<I, false, true> {
     /// Treat parameters in the lhs as rigid, and in rhs as infer vars.
     pub fn relate_rigid_infer(_interner: I) -> DeepRejectCtxt<I, false, true> {
-        DeepRejectCtxt { _interner: PhantomData }
+        DeepRejectCtxt { _interner: PhantomData, cache: Default::default() }
     }
 }
 
 impl<I: Interner, const INSTANTIATE_LHS_WITH_INFER: bool, const INSTANTIATE_RHS_WITH_INFER: bool>
     DeepRejectCtxt<I, INSTANTIATE_LHS_WITH_INFER, INSTANTIATE_RHS_WITH_INFER>
 {
     pub fn args_may_unify(
-        self,
+        &mut self,
         obligation_args: I::GenericArgs,
         impl_args: I::GenericArgs,
     ) -> bool {
@@ -234,7 +238,7 @@ impl<I: Interner, const INSTANTIATE_LHS_WITH_INFER: bool, const INSTANTIATE_RHS_
         })
     }
 
-    pub fn types_may_unify(self, lhs: I::Ty, rhs: I::Ty) -> bool {
+    pub fn types_may_unify(&mut self, lhs: I::Ty, rhs: I::Ty) -> bool {
         match rhs.kind() {
             // Start by checking whether the `rhs` type may unify with
             // pretty much everything. Just return `true` in that case.
@@ -273,8 +277,12 @@ impl<I: Interner, const INSTANTIATE_LHS_WITH_INFER: bool, const INSTANTIATE_RHS_
             | ty::Placeholder(_) => {}
         };
 
+        if self.cache.contains(&(lhs, rhs)) {
+            return true;
+        }
+
         // For purely rigid types, use structural equivalence.
-        match lhs.kind() {
+        let result = match lhs.kind() {
             ty::Ref(_, lhs_ty, lhs_mutbl) => match rhs.kind() {
                 ty::Ref(_, rhs_ty, rhs_mutbl) => {
                     lhs_mutbl == rhs_mutbl && self.types_may_unify(lhs_ty, rhs_ty)
@@ -414,10 +422,16 @@ impl<I: Interner, const INSTANTIATE_LHS_WITH_INFER: bool, const INSTANTIATE_RHS_
             }
 
             ty::Error(..) => true,
+        };
+
+        if result {
+            self.cache.insert((lhs, rhs));
         }
+
+        result
     }
 
-    pub fn consts_may_unify(self, lhs: I::Const, rhs: I::Const) -> bool {
+    pub fn consts_may_unify(&mut self, lhs: I::Const, rhs: I::Const) -> bool {
         match rhs.kind() {
             ty::ConstKind::Param(_) => {
                 if INSTANTIATE_RHS_WITH_INFER {
@@ -465,7 +479,7 @@ impl<I: Interner, const INSTANTIATE_LHS_WITH_INFER: bool, const INSTANTIATE_RHS_
         }
     }
 
-    fn var_and_ty_may_unify(self, var: ty::InferTy, ty: I::Ty) -> bool {
+    fn var_and_ty_may_unify(&mut self, var: ty::InferTy, ty: I::Ty) -> bool {
         if !ty.is_known_rigid() {
             return true;
         }

src/librustdoc/html/render/write_shared.rs

@@ -918,8 +918,8 @@ impl<'item> DocVisitor<'item> for TypeImplCollector<'_, '_, 'item> {
             // Be aware of `tests/rustdoc/type-alias/deeply-nested-112515.rs` which might regress.
             let Some(impl_did) = impl_item_id.as_def_id() else { continue };
             let for_ty = self.cx.tcx().type_of(impl_did).skip_binder();
-            let reject_cx = DeepRejectCtxt::relate_infer_infer(self.cx.tcx());
-            if !reject_cx.types_may_unify(aliased_ty, for_ty) {
+            let mut drcx = DeepRejectCtxt::relate_infer_infer(self.cx.tcx());
+            if !drcx.types_may_unify(aliased_ty, for_ty) {
                 continue;
             }
             // Avoid duplicates