summaryrefslogtreecommitdiffstats
path: root/src/test/ui/regions/regions-implied-bounds-projection-gap-hr-1.rs
diff options
context:
space:
mode:
Diffstat (limited to 'src/test/ui/regions/regions-implied-bounds-projection-gap-hr-1.rs')
-rw-r--r--src/test/ui/regions/regions-implied-bounds-projection-gap-hr-1.rs26
1 files changed, 26 insertions, 0 deletions
diff --git a/src/test/ui/regions/regions-implied-bounds-projection-gap-hr-1.rs b/src/test/ui/regions/regions-implied-bounds-projection-gap-hr-1.rs
new file mode 100644
index 000000000..c1dab6086
--- /dev/null
+++ b/src/test/ui/regions/regions-implied-bounds-projection-gap-hr-1.rs
@@ -0,0 +1,26 @@
+// The "projection gap" is particularly "fun" around higher-ranked
+// projections. This is because the current code is hard-coded to say
+// that a projection that contains escaping regions, like `<T as
+// Trait2<'y, 'z>>::Foo` where `'z` is bound, can only be found to
+// outlive a region if all components that appear free (`'y`, where)
+// outlive that region. However, we DON'T add those components to the
+// implied bounds set, but rather we treat projections with escaping
+// regions as opaque entities, just like projections without escaping
+// regions.
+
+trait Trait1<T> { }
+
+trait Trait2<'a, 'b> {
+ type Foo;
+}
+
+// As a side-effect of the conservative process above, the type of
+// this argument `t` is not automatically considered well-formed,
+// since for it to be WF, we would need to know that `'y: 'x`, but we
+// do not infer that.
+fn callee<'x, 'y, T>(t: &'x dyn for<'z> Trait1< <T as Trait2<'y, 'z>>::Foo >)
+ //~^ the trait bound `for<'z> T: Trait2<'y, 'z>` is not satisfied
+{
+}
+
+fn main() { }