compiler: Inform the solver of concurrency

Parallel compilation of a program can cause unexpected event sequencing.
Inform the solver when this is true so it can skip invalid asserts, then
assert replaced solutions are equal if Some

compiler/rustc_middle/src/ty/context.rs

@@ -181,6 +181,10 @@ fn with_global_cache<R>(
         }
     }
 
+    fn evaluation_is_concurrent(&self) -> bool {
+        self.sess.threads() > 1
+    }
+
     fn expand_abstract_consts<T: TypeFoldable<TyCtxt<'tcx>>>(self, t: T) -> T {
         self.expand_abstract_consts(t)
     }

compiler/rustc_type_ir/src/interner.rs

@@ -137,6 +137,8 @@ fn with_global_cache<R>(
         f: impl FnOnce(&mut search_graph::GlobalCache<Self>) -> R,
     ) -> R;
 
+    fn evaluation_is_concurrent(&self) -> bool;
+
     fn expand_abstract_consts<T: TypeFoldable<Self>>(self, t: T) -> T;
 
     type GenericsOf: GenericsOf<Self>;
@@ -404,4 +406,7 @@ fn with_global_cache<R>(
     ) -> R {
         I::with_global_cache(self, mode, f)
     }
+    fn evaluation_is_concurrent(&self) -> bool {
+        self.evaluation_is_concurrent()
+    }
 }

compiler/rustc_type_ir/src/search_graph/global_cache.rs

@@ -44,22 +44,28 @@ pub(super) fn insert(
         cx: X,
         input: X::Input,
 
-        result: X::Result,
+        origin_result: X::Result,
         dep_node: X::DepNodeIndex,
 
         additional_depth: usize,
         encountered_overflow: bool,
         nested_goals: NestedGoals<X>,
     ) {
-        let result = cx.mk_tracked(result, dep_node);
+        let result = cx.mk_tracked(origin_result, dep_node);
         let entry = self.map.entry(input).or_default();
         if encountered_overflow {
             let with_overflow = WithOverflow { nested_goals, result };
             let prev = entry.with_overflow.insert(additional_depth, with_overflow);
-            assert!(prev.is_none());
+            if let Some(prev) = &prev {
+                assert!(cx.evaluation_is_concurrent());
+                assert_eq!(cx.get_tracked(&prev.result), origin_result);
+            }
         } else {
             let prev = entry.success.replace(Success { additional_depth, nested_goals, result });
-            assert!(prev.is_none());
+            if let Some(prev) = &prev {
+                assert!(cx.evaluation_is_concurrent());
+                assert_eq!(cx.get_tracked(&prev.result), origin_result);
+            }
         }
     }
 

compiler/rustc_type_ir/src/search_graph/mod.rs

@@ -53,6 +53,8 @@ fn with_global_cache<R>(
         mode: SolverMode,
         f: impl FnOnce(&mut GlobalCache<Self>) -> R,
     ) -> R;
+
+    fn evaluation_is_concurrent(&self) -> bool;
 }
 
 pub trait Delegate {

tests/ui/traits/next-solver/global-cache-and-parallel-frontend.rs

@@ -0,0 +1,27 @@
+//@ compile-flags: -Zthreads=16
+
+// original issue: https://github.com/rust-lang/rust/issues/129112
+// Previously, the "next" solver asserted that each successful solution is only obtained once.
+// This test exhibits a repro that, with next-solver + -Zthreads, triggered that old assert.
+// In the presence of multithreaded solving, it's possible to concurrently evaluate things twice,
+// which leads to replacing already-solved solutions in the global solution cache!
+// We assume this is fine if we check to make sure they are solved the same way each time.
+
+// This test only nondeterministically fails but that's okay, as it will be rerun by CI many times,
+// so it should almost always fail before anything is merged. As other thread tests already exist,
+// we already face this difficulty, probably. If we need to fix this by reducing the error margin,
+// we should improve compiletest.
+
+#[derive(Clone, Eq)] //~ ERROR [E0277]
+pub struct Struct<T>(T);
+
+impl<T: Clone, U> PartialEq<U> for Struct<T>
+where
+    U: Into<Struct<T>> + Clone
+{
+    fn eq(&self, _other: &U) -> bool {
+        todo!()
+    }
+}
+
+fn main() {}

tests/ui/traits/next-solver/global-cache-and-parallel-frontend.stderr

@@ -0,0 +1,24 @@
+error[E0277]: the trait bound `T: Clone` is not satisfied
+  --> $DIR/global-cache-and-parallel-frontend.rs:15:17
+   |
+LL | #[derive(Clone, Eq)]
+   |                 ^^ the trait `Clone` is not implemented for `T`, which is required by `Struct<T>: PartialEq`
+   |
+note: required for `Struct<T>` to implement `PartialEq`
+  --> $DIR/global-cache-and-parallel-frontend.rs:18:19
+   |
+LL | impl<T: Clone, U> PartialEq<U> for Struct<T>
+   |         -----     ^^^^^^^^^^^^     ^^^^^^^^^
+   |         |
+   |         unsatisfied trait bound introduced here
+note: required by a bound in `Eq`
+  --> $SRC_DIR/core/src/cmp.rs:LL:COL
+   = note: this error originates in the derive macro `Eq` (in Nightly builds, run with -Z macro-backtrace for more info)
+help: consider restricting type parameter `T`
+   |
+LL | pub struct Struct<T: std::clone::Clone>(T);
+   |                    +++++++++++++++++++
+
+error: aborting due to 1 previous error
+
+For more information about this error, try `rustc --explain E0277`.