diff options
| author | Dave Mitchell <davem@fdisolutions.com> | 2004-01-08 22:32:28 +0000 |
|---|---|---|
| committer | Dave Mitchell <davem@fdisolutions.com> | 2004-01-08 22:32:28 +0000 |
| commit | b23f1a86dd8a81ae1056d802240e99273db9cd3a (patch) | |
| tree | 27ba4e3e880421d36c75d1d83f13be0340261541 /sv.c | |
| parent | 95143fff86b7efc2ca19a65dbad4c11842f65021 (diff) | |
| download | perl-b23f1a86dd8a81ae1056d802240e99273db9cd3a.tar.gz | |
segv in pad.c with threads (was: DBD::Oracle and Perl 5.8.2 threads)
Message-ID: <20040107121357.GD82921@dansat.data-plan.com>
Returning a closure from a thread (via join) could mess up because
pointers to PL_sv_undef weren't rejigged to point at the joiner's
version of PL_sv_undef. Also, the closure's CvGV got cloned too
but never freed, since CvGV isn't refcounted.
p4raw-id: //depot/perl@22102
Diffstat (limited to 'sv.c')
| -rw-r--r-- | sv.c | 5 |
1 files changed, 4 insertions, 1 deletions
@@ -10374,7 +10374,10 @@ Perl_sv_dup(pTHX_ SV *sstr, CLONE_PARAMS* param) SvREFCNT_inc(CvXSUBANY(sstr).any_ptr) : sv_dup_inc(CvXSUBANY(sstr).any_ptr, param); } - CvGV(dstr) = gv_dup(CvGV(sstr), param); + /* don't dup if copying back - CvGV isn't refcounted, so the + * duped GV may never be freed. A bit of a hack! DAPM */ + CvGV(dstr) = (param->flags & CLONEf_JOIN_IN) ? + Nullgv : gv_dup(CvGV(sstr), param) ; if (param->flags & CLONEf_COPY_STACKS) { CvDEPTH(dstr) = CvDEPTH(sstr); } else { |