Re: The evil Perl reads freed things

On Tue, Jun 22, 2004 at 05:29:25PM +0100, Tim Bunce wrote:
> On Tue, Jun 22, 2004 at 02:09:01PM +0100, Nicholas Clark wrote:
> > On Fri, Jun 18, 2004 at 08:36:59PM +0100, Nicholas Clark wrote:
> > 
> > > Other tests fail in strange ways, and valgrind reports errors which are
> > > reads in the heads of now-free scalars.
> > > 
> > > So, I'm not sure what to do, as this assumption that you can read any
> > > head even after it's undead seems to be quite deeply ingrained. And it
> > > needs to be solved before we can usefully replace heads with PMCs.
> 
> [I wrote the next two paragraphs on Friday but didn't send it]
> 
> Perl5 doesn't refcount what it puts on the stack. Assuming that's
> the cause of many/most of these cases, perhaps ponie will need to.
> 
> But then again, won't that problem go away if/when ponie starts
> using GC instead of ref counting?

It's the cause of the checks w.r.t. freeing an array you're iterating over
in a for loop. But there seem to be other places.
I'm hoping that all the stack related problems go away ones the stack is
garbage collected.


> Though free_tmps may trigger DESTROYs so freeing the list at the
> end of free_tmps would be wise, but you're probably doing that anyway.)

Yes. I figured that free_tmps would only be adding to the list of things
that needed freeing soon, so best to do it after that. :-)
> 
> > Ponie in CVS now has SV* being a pointer to a PMC. (and all tests pass)
> 
> Yeah. Good work.
> 
> How's the performance situation?

I've no idea. I expect it's slower than before, because now SvANY() and
SvFLAGS() involve (at least) 2 function calls.

Nicholas Clark