Mailing List Archive: Known OVERLOAD bugs?

On Thu, 14 Sep 1995 03:14:20 CDT, Gerd Knops wrote:
>Are there any known bugs with OVERLOAD? I am hunting
>a very ugly problem since a couple of days now.
>It looks like some memory gets overwritten, but I
>can not find anything wrong with my code.
>
[...]
>
>If that rings a bell, please please please drop me
>a line!
>
>gerti
>

Yes, it does ring a bell :-)

The following is not really about any bugs in OVERLOAD, but might very well
be one in base perl that shows up as bugs in OVERLOAD+mysterious memory
corruption. I have found that the following conditional compilation of
free_tmps() in scope.c:

#ifdef DEBUGGING
SvTEMP_off(sv);
#endif
SvREFCNT_dec(sv); /* note, can modify tmps_ix!!! */

might really be rather wrong (given that the SVf_TEMP flag is not exclusive
to -DDEBUGGING). This means that there's a possibility to get different
behavior in debugging-perl when SvREFCNT(sv) > 1.

So my suggestion to you would be to test your program again with a perl
compiled with -DDEBUGGING and see if it changes anything.

[.Aside: I ran into this when Chip Salzenberg reported that my first release
consolidated patches were blowing on OVERLOAD tests in mysterious ways. I
couldn't replicate the problem because I was compiling with -DDEBUGGING and
he wasn't. The problem shows up in OVERLOAD tests when you do an
sv_2mortal() on an SV that has a REFCNT > 1. I decided at that time to avoid
this problematic code in my patches by only doing sv_2mortal(sv) after first
checking for SvREFCNT(sv) == 1 (because I wasn't sure why this conditional
existed). I now think I should have fixed above code instead in my
patches...]

- Sarathy.
gsar@engin.umich.edu