rustdoc: synthetic auto trait impls: accept unresolved region vars for now

src/librustdoc/clean/auto_trait.rs

@@ -181,8 +181,14 @@ fn clean_param_env<'tcx>(
         })
         .map(|pred| {
             tcx.fold_regions(pred, |r, _| match *r {
-                ty::ReVar(vid) => vid_to_region[&vid],
+                // FIXME: Don't `unwrap_or`, I think we should panic if we encounter an infer var that
+                // we can't map to a concrete region. However, `AutoTraitFinder` *does* leak those kinds
+                // of `ReVar`s for some reason at the time of writing. See `rustdoc-ui/` tests.
+                // This is in dire need of an investigation into `AutoTraitFinder`.
+                ty::ReVar(vid) => vid_to_region.get(&vid).copied().unwrap_or(r),
                 ty::ReEarlyParam(_) | ty::ReStatic | ty::ReBound(..) | ty::ReError(_) => r,
+                // FIXME(#120606): `AutoTraitFinder` can actually leak placeholder regions which feels
+                // incorrect. Needs investigation.
                 ty::ReLateParam(_) | ty::RePlaceholder(_) | ty::ReErased => {
                     bug!("unexpected region kind: {r:?}")
                 }

tests/rustdoc-ui/synthetic-auto-trait-impls/lifetime-generic-user-impl-normalize.rs

@@ -0,0 +1,17 @@
+// We used to ICE here while trying to synthesize auto trait impls.
+// issue: 112242
+//@ check-pass
+//@ compile-flags: -Znormalize-docs
+
+pub trait MyTrait<'a> {
+    type MyItem;
+}
+pub struct Inner<Q>(Q);
+pub struct Outer<Q>(Inner<Q>);
+
+impl<'a, Q> std::marker::Unpin for Inner<Q>
+where
+    Q: MyTrait<'a>,
+    <Q as MyTrait<'a>>::MyItem: Copy,
+{
+}

tests/rustdoc-ui/synthetic-auto-trait-impls/lifetime-generic-user-impl.rs

@@ -0,0 +1,11 @@
+// We used to ICE here while trying to synthesize auto trait impls.
+// issue: 123370
+//@ check-pass
+
+pub struct Inner<'a, Q>(&'a (), Q);
+
+pub struct Outer<'a, Q>(Inner<'a, Q>);
+
+impl<'a, Q: Trait<'a>> std::marker::Unpin for Inner<'static, Q> {}
+
+pub trait Trait<'a> {}