In fact 'assert false' is explicitely replaced by 'raise
Assert_failure(...)' by the parser.
See the documentation about that in the "Language extensions" section.

With any other argument than false, assert is considered to have type bool
-> unit



Esther Baruk


On Wed, Aug 31, 2011 at 1:40 AM, Philippe Wang <mail@philippewang.info>wrote:

> On Wed, Aug 31, 2011 at 12:31 AM, John Carr <jfc@mit.edu> wrote:
> > # let f () = assert false; 0;;
> > Warning 21: this statement never returns (or has an unsound type.)
> > val f : unit -> int = <fun>
>
> > Assert behaves like a function with type bool -> 'a.  The return type is
> > an unconstrained type, a type variable not mentioned on the left hand
> > side of the arrow.
>
> Actually, assert is more like a function with type bool -> unit.
> Indeed, (assert false) is a special case where it is "so obvious that
> it'll raise an exception that it's made type 'a".
>
> Notably, if you write
> assert(1=2)
> it's not "obvious enough" for the compiler to assume it will raise an
> exception.
> (Which is normal, by the way, because it's generally an undecidable
> problem: in general, it's not possible to know that the evaluation of
> assert's argument will return a value, especially when there is a
> function application such as in "1=2".)
>
> So,
> in
> let f () = assert false
> f : unit -> 'a
> let f () = assert (1=2)
> f : unit -> unit
>
> And while
> let f () = assert false; 0;;
> raises a warning,
> let f () = assert (1=2); 0;;
> won't raise any.
>
> Somehow, the choice of discriminating (assert false) is not so perfect...
>
> Cheers,
>
> --
> Philippe Wang
>
>
> N.B. Sorry for double-replying.
> I wonder if "reply-to: caml-list@inria.fr" field could be added
> automatically to headers of all mails received by the list...
>
>
> --
> Caml-list mailing list.  Subscription management and archives:
> https://sympa-roc.inria.fr/wws/info/caml-list
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs
>
>