Front page | perl.perl5.porters |
Postings from April 2003
Jarkko Hietaniemi wrote:
>
> I may be compleletly wrong on this but maybe the PL_nullstash
> was handling the "homeless" variables created by the "package;"
> feature which is what the #17942 was removing.
Yes :
(And applying #17942 in reverse solves the bug.)
$ perl5.8.0 -le 'package;@{"homeless"};print for keys %{"<none>::"}'
homeless
But it also holds the illegal variables :
$ perl5.8.0 -le 'use strict"vars";eval q{$foo};print for keys %{"<none>::"}'
foo
Now (in bleadperl) the illegal variables are simply thrown away.
Which leads to this segfault case.
> But your patch looks good enough for now.
(Paranoid statement follows) : I don't know, perhaps there are other
uncovered bugs caused by this change.
And as says Adi :
> > I don't think that gv_fetchpv returning NULL when its 'add' argument
> > is non-zero is exactly right - and I failed to see what was so
> > wrong with PL_nullstash.
So another fix is to have gv_fetchpv return a dummy stash instead of
null when it has an "add" argument (implemented by the undertested patch
below). This is equivalent as restoring PL_nullstash for that purpose.
(Except that this time I'd name it <buffy>::, because, you know, that's
sorta a hellmouth-stash. Bad things go there and don't go out.)
Thoughts ?
Index: gv.c
===================================================================
--- gv.c (revision 1094)
+++ gv.c (working copy)
@@ -764,8 +764,10 @@ Perl_gv_fetchpv(pTHX_ const char *nambeg
: sv_type == SVt_PVAV ? "@"
: sv_type == SVt_PVHV ? "%"
: ""), name));
+ stash = GvHV(gv_fetchpv("<buffy>::", GV_ADDMULTI, SVt_PVHV));
}
- return Nullgv;
+ else
+ return Nullgv;
}
if (!SvREFCNT(stash)) /* symbol table under destruction */
End.
--
Usable is not *NIX
Thread Previous
|
Thread Next