Re: down to one test failure

On Tue, Aug 13, 2002 at 08:46:31PM +0100, Richard Clamp wrote:
> This leaves me with one failing test on my local boxes - t/comp.decl,
> which currently exits after 5 tests, and throws a big wobbly under
> valgrind[0].  I hope to trap this one later this evening.

I lied, there were two failing tests.  op.unshift was also not happy
on Linux/powerpc - valgrind later showed this to be reading outside of
where was malloced, which this I tracked to this in array.c.

     if (key > ar->ary_max) {
-       int newmax = key + ar->ary_max / 5;
+       int newmax = key + ar->ary_max;

        ar->ary_array = (STR**)saferealloc((char*)ar->ary_array,
            (newmax+1) * sizeof(STR*));

It feels like the right solution as I can't really see why it was / 5
before, can anyone correct/enlighten me?

I've badly papered over a crack in stab.c, maybe someone wiser can
debug this:

        case '~':
-           safefree(curoutstab->stab_io->fmt_name);
+           /* FIXME: investigate more carefully.  When the following
+            * safefree is allowed to happen the subsequent stabent call
+            * results in a segfault.  Commenting it out is a cheap band-aid
+            * and probably a memory leak rolled into one
+            *  -- richardc 2002-08-14
+            */
+           /* safefree(curoutstab->stab_io->fmt_name); */ 
            curoutstab->stab_io->fmt_name = str_get(str);
            curoutstab->stab_io->fmt_stab = stabent(str_get(str),TRUE);
            break;

I can now dubiously claim test passes on Linux and Mac OS X 10.1.5 on
powerpc, Linux and OpenBSD-3.1 on x86.

There are 2 current valgrind errors[0], and a whole bunch of memory
leaks, but I'm calling it progress.  Please could folks with other
architectures grab a version from my repository[1] and email me if it
works.  Thanks.


[0] http://unixbeard.net/~richardc/lab/grinder/errors.html
[1] http://mirth.unixbeard.net/cgi-bin/viewcvs.cgi/perl1/perl1.tar.gz?tarball=1

-- 
Richard Clamp <[email protected]>