[Bug-openmcl] :allow-other-keys

[Gary Byers]

On Wed, 7 Jan 2004, bryan o'connor wrote:

> > I was going to say "I thought that I'd fixed that".  I did (in
> > 0.13.x), but the change doesn't seem to have made it correctly back
> > into 0.14.  Whoops.
>
> do you remember where the fix got applied?  i went looking
> through the 0.13 cvs tree, but it looks like the history
> got nuked when clozure's disk went belly up.
>
> :allow-other-keys should fix ~80 of the ansi tests.

It's actually some assembler code in the kernel.  It was in
ccl:lisp-kernel;sp_lambda.s in 0.13; practically all of the
"subprimitives" are in sp_entry.s in 0.14.

I think that I checked the fix into the bleeding-edge tree early
this evening.  I'll try to get it into 0.14 after I get some
sleep ...

>
> the second and third most popular failures seem to be bugs in
> #'subtypep (which took me a couple hours just to figure out how
> it's supposed to work) and #'make-array when the :element-type
> is nil.
>

I have real difficulty taking the whole :ELEMENT-TYPE NIL thing
seriously.  Paul's tests are generally very good and he's generally
very rigorous about things; when I ran his tests a year ago, I
often thought that he was mistaken about something only to read
the spec more carefully and discover that he was indeed right.
In this case, I think that he's misinterpreting a poor choice
of words to say something that I don't believe was ever intended;
if there's a more compelling argument that says that arrays specialized
to the NIL type have to be provided by the implementation (in the
sense that BIT and CHARACTER and BASE-CHAR have to have specialized
array types, but whether DOUBLE-FLOAT does or not is up to the
implementation) I certainly can't find it, and I think that a proposal
to introduce them would have been controversial enough to be memorable.

Thanks for doing this.  I found last year that fixing one bug would
often cause 20 or 30 previously-failed test to pass, so hopefully
the set of failures will start to get manageable soon.  I'd certainly
believe that a lot of SUBTYPEP failures are legitimate, and I'd also
expect mamy of them to be hard to fix.



> 					...bryan
>
> _______________________________________________
> Bug-openmcl mailing list
> Bug-openmcl at clozure.com
> http://clozure.com/mailman/listinfo/bug-openmcl
>
>