Re: Poniedate #43

On Mon, Mar 01, 2004 at 03:12:59PM +0000, Arthur Bergman wrote:
> 
> On 1 Mar 2004, at 15:06, Dave Mitchell wrote:
> 
> >
> >I'm probably teaching my grandmopther to suck eggs here, but...

Well, I didn't know this :-)

> >You do know that a a pad consists of an array-of-arrays, rather than 
> >the
> >usual array-of-refs-to-arrays? And that the outer array isn't AvREAL, 
> >and
> >so doesn't do ref-couning?
> 
> I know, and it shouldn't really matter since I haven't changed the 
> code, it is still using all the macros to get the normal pointers 
> except that it happens through an indirection.
> 
> So AvARRAY  calls    Perl_macro_AvARRAY that does a 
> Parrot_PMC_get_pointer and then return &(xav->xav_array)   for example. 
> So it shouldn't be changing anything.

If you want to join in staring at this bug, it's present in the current code
checked into CVS.
I did a clean checkout
  (cvs co ponie; cd ponie; cvs co parrot; perl Configure.pl; make)
and it fails with a SEGV in make of lib.pm
An example test failing from make minitest is chdir.t, which goes SEGV
while compiling a line local $\ = "\n" in Cwd.pm

We've got a local patch to valgrind to stop the "impossible" happening - not
sure when that will make it to an upstream release.

Nicholas Clark