Remove hack skipping comparison of algebraic universes in subtyping.

When inferring [u <= v+k] I replaced the exception and instead add
[u <= v]. This is trivially sound and it doesn't seem possible to have
the one without the other (except specially for [Set <= v+k] which was
already handled).

I don't know an example where this used to fail and now succeeds (the
point was to remove an anomaly, but the example
~~~
Module Type SG. Definition DG := Type. End SG.
Module MG : SG. Definition DG := Type : Type. Fail End MG.
~~~
now fails with universe inconsistency.

Fix #7695 (soundness bug!).

1 files changed, 20 insertions, 0 deletions

diff --git a/test-suite/bugs/closed/7695.v b/test-suite/bugs/closed/7695.v
new file mode 100644
index 000000000..42bdb076b
--- /dev/null
+++ b/test-suite/bugs/closed/7695.v
@@ -0,0 +1,20 @@
+Require Import Hurkens.
+
+Universes i j k.
+Module Type T.
+  Parameter T1 : Type@{i+1}.
+  Parameter e : Type@{j} = T1 : Type@{k}.
+End T.
+
+Module M.
+  Definition T1 := Type@{j}.
+  Definition e : Type@{j} = T1 : Type@{k} := eq_refl.
+End M.
+
+Module F (A:T).
+  Definition bad := TypeNeqSmallType.paradox _ A.e.
+End F.
+
+Set Printing Universes.
+Fail Module X := F M.
+(* Universe inconsistency. Cannot enforce j <= i because i < Coq.Logic.Hurkens.105 = j. *)