saweval = i;
seq = cxstack[i].blk_oldcop->cop_seq;
startcv = cxstack[i].blk_eval.cv;
- off = pad_findlex(name, newoff, seq, startcv, i-1,
- saweval, 0);
- if (off) /* continue looking if not found here */
- return off;
+ if (startcv && CvOUTSIDE(startcv)) {
+ off = pad_findlex(name, newoff, seq, CvOUTSIDE(startcv),
+ i-1, saweval, 0);
+ if (off) /* continue looking if not found here */
+ return off;
+ }
}
break;
case OP_DOFILE:
* CV, they don't hold a refcount on the outside CV. This avoids
* the refcount loop between the outer CV (which keeps a refcount to
* the closure prototype in the pad entry for pp_anoncode()) and the
- * closure prototype, and the ensuing memory leak. --GSAR */
- if (!CvANON(cv) || CvCLONED(cv))
+ * closure prototype, and the ensuing memory leak. This does not
+ * apply to closures generated within eval"", since eval"" CVs are
+ * ephemeral. --GSAR */
+ if (!CvANON(cv) || CvCLONED(cv)
+ || (CvOUTSIDE(cv) && CvEVAL(CvOUTSIDE(cv)) && !CvGV(CvOUTSIDE(cv))))
+ {
SvREFCNT_dec(CvOUTSIDE(cv));
+ }
CvOUTSIDE(cv) = Nullcv;
if (CvCONST(cv)) {
SvREFCNT_dec((SV*)CvXSUBANY(cv).any_ptr);
}
}
- /* If a potential closure prototype, don't keep a refcount on outer CV.
+ /* If a potential closure prototype, don't keep a refcount on
+ * outer CV, unless the latter happens to be a passing eval"".
* This is okay as the lifetime of the prototype is tied to the
* lifetime of the outer CV. Avoids memory leak due to reference
* loop. --GSAR */
- if (!name)
+ if (!name && CvOUTSIDE(cv)
+ && !(CvEVAL(CvOUTSIDE(cv)) && !CvGV(CvOUTSIDE(cv))))
+ {
SvREFCNT_dec(CvOUTSIDE(cv));
+ }
if (name || aname) {
char *s;
EXPECT
ok
########
+# test that closures generated by eval"" hold on to the CV of the eval""
+# for their entire lifetime
+$code = eval q[
+ sub { eval '$x = "ok 1\n"'; }
+];
+&{$code}();
+print $x;
+EXPECT
+ok 1
+########
# This test is here instead of pragma/locale.t because
# the bug depends on in the internal state of the locale
# settings and pragma/locale messes up that state pretty badly.