improve comments on dropck_outlives

Author: nikomatsakis

src/librustc/traits/query/type_op/mod.rs

@@ -49,6 +49,12 @@ pub trait QueryTypeOp<'gcx: 'tcx, 'tcx>: fmt::Debug + Sized {
 
     fn param_env(key: &Self::QueryKey) -> ParamEnv<'tcx>;
 
+    /// Performs the actual query with the canonicalized key -- the
+    /// real work happens here. This method is not given an `infcx`
+    /// because it shouldn't need one -- and if it had access to one,
+    /// it might do things like invoke `sub_regions`, which would be
+    /// bad, because it would create subregion relationships that are
+    /// not captured in the return value.
     fn perform_query(
         tcx: TyCtxt<'_, 'gcx, 'tcx>,
         canonicalized: Canonicalized<'gcx, Self::QueryKey>,

src/librustc/traits/query/type_op/outlives.rs

@@ -52,6 +52,15 @@ where
         tcx: TyCtxt<'_, 'gcx, 'tcx>,
         canonicalized: Canonicalized<'gcx, Self::QueryKey>,
     ) -> Fallible<CanonicalizedQueryResult<'gcx, Self::QueryResult>> {
+        // Subtle: note that we are not invoking
+        // `infcx.at(...).dropck_outlives(...)` here, but rather the
+        // underlying `dropck_outlives` query. This same underlying
+        // query is also used by the
+        // `infcx.at(...).dropck_outlives(...)` fn. Avoiding the
+        // wrapper means we don't need an infcx in this code, which is
+        // good because the interface doesn't give us one (so that we
+        // know we are not registering any subregion relations or
+        // other things).
         tcx.dropck_outlives(canonicalized)
     }