I wrote:
> FOREACH [0, 1];
> cols.$curcol = 1;
> curcol = curcol + 1;
> Dumper.dump([cols, curcol]);
> END;
> $VAR1 = [
> [
> 1
> ],
> 1
> ];
> $VAR1 = [
> [
> 1,
> ${\$VAR1->[0][0]}
> ],
> 2
> ];
>
>I expect that last reference to be "1", not a reference. If I use
>Template::Stash instead of Template::Stash::XS, then I get the expected
>result.
Upon further review, this workaround "fixes" it:
[%
USE Dumper;
curcol = 0;
cols = [];
Dumper.dump([cols, curcol]);
FOREACH [0, 1];
temp = 1;
cols.$curcol = temp;
curcol = curcol + 1;
Dumper.dump([cols, curcol]);
END;
%]
Yes, just putting it in a temp value fixes the problem. Ouch! So armed
with this new information, I delved into Stash.xs and came up with this
patch:
--- xs/Stash.xs.orig Thu Feb 20 16:29:13 2003
+++ xs/Stash.xs Thu Feb 20 16:29:17 2003
@@ -681,11 +681,12 @@
return &PL_sv_undef;
}
+ newsv = newSVsv(value);
if (looks_like_number(key_sv)) {
- if (av_store(rootav, SvIV(key_sv), value))
- SvREFCNT_inc(value);
+ if (av_store(rootav, SvIV(key_sv), newsv))
+ SvREFCNT_inc(newsv);
else
- SvSETMAGIC(value);
+ SvSETMAGIC(newsv);
return value;
}
I am not 100% sure that it is the right thing to do, but it mirrors what
happens immediately above it in case SVt_PVHV, and it seems to fix the
problem. In Slash, we are going to stick with the temp-var workaround for
now, since we don't want to deal with a possibly incorrect patch, and many
other sites using our code might not have this patch anyway. :)
Thanks to Jamie McCarthy who did most of the legwork identifying the
problem and Chip Salzenberg for helping me sanity-check the patch.
--
Chris Nandor [EMAIL PROTECTED] http://pudge.net/
Open Source Development Network [EMAIL PROTECTED] http://osdn.com/
_______________________________________________
templates mailing list
[EMAIL PROTECTED]
http://lists.template-toolkit.org/mailman/listinfo/templates
