summaryrefslogtreecommitdiff
path: root/test-suite/failure/universes3.v
diff options
context:
space:
mode:
Diffstat (limited to 'test-suite/failure/universes3.v')
-rw-r--r--test-suite/failure/universes3.v25
1 files changed, 25 insertions, 0 deletions
diff --git a/test-suite/failure/universes3.v b/test-suite/failure/universes3.v
new file mode 100644
index 00000000..8fb414d5
--- /dev/null
+++ b/test-suite/failure/universes3.v
@@ -0,0 +1,25 @@
+(* This example (found by coqchk) checks that an inductive cannot be
+ polymorphic if its constructors induce upper universe constraints.
+ Here: I cannot be polymorphic because its type is less than the
+ type of the argument of impl. *)
+
+Definition Type1 := Type.
+Definition Type3 : Type1 := Type. (* Type3 < Type1 *)
+Definition Type4 := Type.
+Definition impl (A B:Type3) : Type4 := A->B. (* Type3 <= Type4 *)
+Inductive I (B:Type (*6*)) := C : B -> impl Prop (I B).
+ (* Type(6) <= Type(7) because I contains, via C, elements in B
+ Type(7) <= Type3 because (I B) is argument of impl
+ Type(4) <= Type(7) because type of C less than I (see remark below)
+
+ where Type(7) is the auxiliary level used to infer the type of I
+*)
+
+(* We cannot enforce Type1 < Type(6) while we already have
+ Type(6) <= Type(7) < Type3 < Type1 *)
+Definition J := I Type1.
+
+(* Open question: should the type of an inductive be the max of the
+ types of the _arguments_ of its constructors (here B and Prop,
+ after unfolding of impl), or of the max of types of the
+ constructors itself (here B -> impl Prop (I B)), as done above. *)