actually create the ndisj HintDb. this makes it unfold constants so that we can use 'Hint Resolve'.

theories/base_logic/lib/namespaces.v

@@ -87,13 +87,12 @@ of the forms:
 - [N1 ⊥ N2] 
 - [↑N1 ⊆ E ∖ ↑N2 ∖ .. ∖ ↑Nn]
 - [E1 ∖ ↑N1 ⊆ E2 ∖ ↑N2 ∖ .. ∖ ↑Nn] *)
+Create HintDb ndisj.
 Hint Resolve ndisj_subseteq_difference : ndisj.
 Hint Extern 0 (_ ⊥ _) => apply ndot_ne_disjoint; congruence : ndisj.
-Hint Extern 1 (_ ⊥ _) => apply ndot_preserve_disjoint_l : ndisj.
-Hint Extern 1 (_ ⊥ _) => apply ndot_preserve_disjoint_r : ndisj.
-Hint Extern 1 (_ ⊆ _) => apply nclose_subseteq' : ndisj.
+Hint Resolve ndot_preserve_disjoint_l ndot_preserve_disjoint_r
+     nclose_subseteq' ndisj_difference_l : ndisj.
 Hint Resolve namespace_subseteq_difference_l | 100 : ndisj.
-Hint Resolve ndisj_difference_l : ndisj.
 
 Ltac solve_ndisj :=
   repeat match goal with

[Robbert Krebbers]

Nice catch. Any idea why this is so? Is this documented behavior? If not, it looks inconsistent, so we should create a bug report.

[Jacques-Henri Jourdan]

That's already done: https://coq.inria.fr/bugs/show_bug.cgi?id=5381

[Robbert Krebbers]

Excellent!