Front page | perl.perl5.porters |
Postings from April 2000
On Monday, April 10, 2000 12:48 AM, "Gurusamy Sarathy"
<gsar@ActiveState.com> wrote:
> On Sun, 09 Apr 2000 22:21:27 MDT, eric@xmission.com wrote:
> >[excellent analysis snipped]
> >
> >These are the possibilities I see:
> >
> > 1. Have the tokenizer always create a new SV for constant subs.
> > 2. Have peek copy the constant instead of using it directly when
> > it moves it to the pad.
> > 3. Have pad_free never remove the flags when both SVs_PADTMP and
> > SVs_READONLY are set.
> > 4. Have pad_free only remove the flags when the reference count
> > for the SV is 1.
> >
> >I lean towards 4 or 2. I think that 4 would be safe but it might
> >not always free up a pad slot when it should. 2 should also be
> >safe but would cause additional memory usage.
>
> I'd say peek() should make a copy when moving it to the pad if the
> SV is already a PADTMP.
>
Sounds good. Here's the patch.
--- op.c~ Fri Mar 24 22:24:40 2000
+++ op.c Mon Apr 10 21:17:16 2000
@@ -6417,10 +6417,17 @@
* for reference counts, sv_upgrade() etc. */
if (cSVOP->op_sv) {
PADOFFSET ix = pad_alloc(OP_CONST, SVs_PADTMP);
- SvREFCNT_dec(PL_curpad[ix]);
- SvPADTMP_on(cSVOPo->op_sv);
- PL_curpad[ix] = cSVOPo->op_sv;
- cSVOPo->op_sv = Nullsv;
+ if (SvPADTMP(cSVOPo->op_sv)) {
+ /* If op_sv is already a PADTMP then it is being used by */
+ /* another pad so make a copy. */
+ sv_setsv(PL_curpad[ix],cSVOPo->op_sv);
+ SvREADONLY_on(PL_curpad[ix]);
+ } else {
+ SvREFCNT_dec(PL_curpad[ix]);
+ SvPADTMP_on(cSVOPo->op_sv);
+ PL_curpad[ix] = cSVOPo->op_sv;
+ cSVOPo->op_sv = Nullsv;
+ }
o->op_targ = ix;
}
#endif
Thread Previous
|
Thread Next