Re: 5.7.X : problem with attributed variables and memberchk ?

On Saturday 04 October 2008 22:08:59 Bart Demoen wrote:
> > ?- A in 0..5, B in 6..7, memberchk(A, [B, C]).
> > false.
> >
> >
> > Shouldn't the last one return :
> > A = C,
> > C in 0..5,
> > B in 6..7.
> >
> > ?
>
> First a look at:
>
>        ?- help(memberchk).
>        memberchk(?Elem, +List)
>            Equivalent to member/2, but leaves no choice point.
>
> This is enough to answer your question: yes, you are correct about
> your expectation.
>
> In fact, the implementation of pl_memberchk in pl-list.c cuts the
> choicepoint before executing woken goals. It is a bit tricky to get
> this correct (from within C), but currently it doesn't deliver what
> the manual promises I think. If memchk/2 is written in Prolog, you get
> what you expect.

Thanks Bart. There is a call foreignWakeup() that executes delayed calls
and can be used (only) at places where C code is about to do something
that is effectively a cut. Also used by (e.g.) \=. Fixed in 5.6.x and
merged into 5.7.x.

memberchk/2 is now getting a bit complicated and hopefully 5.7.x itself
will get fast enough someday to avoid the desire to write memberchk in C
. Setup and exit are a bit slower. Luckily the iteration over
non-unifying elements without attributes remains the same.

> How did you find this bug ?

I'm still surprised how obvious bugs remain unnoticed for years while
at the same time there are reports on very weird bugs.

	--- Jan

_______________________________________________
SWI-Prolog mailing list
SWI-Prolog <at> iai.uni-bonn.de
https://mailbox.iai.uni-bonn.de/mailman/listinfo.cgi/swi-prolog