> It looks like you've found a serious bug in the current
> implementation. Actually you don't even need a coercion:
> 
>   # (((Nat.z : Nat.t :> int) - 1) : Nat.t);;
>   - : Nat.t = -1
> 
> This should certainly not be accepted, as you can see by writing
> 
> # (((Nat.z : Nat.t :> int) - 1) : int :> Nat.t);;  (* fails *)
> 
> I think I know the reason, which is probably related to the way
> abbreviation expansions are cached, but this needs more investigating.

This is now fixed. I just hope this does not break anything...
It should be ok, but please report any new non-termination problem
immediately :-)

Cheers,

Jacques Garrigue