NULL and sv_setpv vs. newSVpv

I've been seeing some coredumps in cygwin with blead for some time, and
finally got around to pinning it on:

Change 27612 by nicholas@entropy on 2006/03/26 22:12:57

	Replace all sv_mortalcopy(&PL_sv_no); sv_set*(...) with
	sv2mortal(newSV*(...)) to avoid needless upgrades to PVNV and needless
	allocation of a copy of "".

Affected files ...

... //depot/perl/pp_sys.c#499 edit
...
 	/* pw_class and pw_comment are mutually exclusive--.
 	 * see the above note for pw_change, pw_quota, and pw_age. */
-	PUSHs(sv = sv_mortalcopy(&PL_sv_no));
 #   ifdef PWCLASS
-	sv_setpv(sv, pwent->pw_class);
+	PUSHs(sv_2mortal(newSVpv(pwent->pw_class, 0)));
 #   else
 #       ifdef PWCOMMENT
-	sv_setpv(sv, pwent->pw_comment);
+	PUSHs(sv_2mortal(newSVpv(pwent->pw_comment, 0)));
+#	else
+	/* I think that you can never get this compiled, but just in case.  */
+	PUSHs(sv_mortalcopy(&PL_sv_no));
 #       endif
 #   endif

There's a subtle difference there, in that newSVpv doesn't allow a null
pv when length is 0, while sv_setpv does (doing a SvOK_off in that case.)
And of course pwent->pw_comment is NULL for cygwin.  I could just fix that,
like:

--- perl-current/pp_sys.c	2006-08-21 03:18:50.000000000 -0700
+++ perl/pp_sys.c	2006-09-10 16:23:41.843750000 -0700
@@ -5202,7 +5202,7 @@
 	PUSHs(sv_2mortal(newSVpv(pwent->pw_class, 0)));
 #   else
 #       ifdef PWCOMMENT
-	PUSHs(sv_2mortal(newSVpv(pwent->pw_comment, 0)));
+	PUSHs(sv_2mortal(newSVpv(pwent->pw_comment ? pwent->pw_comment : "", 0)));
 #	else
 	/* I think that you can never get this compiled, but just in case.  */
 	PUSHs(sv_mortalcopy(&PL_sv_no));


but that patch made a similar change in a lot of other cases.
Would it be better to do something like:

--- perl-current/sv.c	2006-09-05 03:44:58.000000000 -0700
+++ perl/sv.c	2006-09-10 20:25:52.656250000 -0700
@@ -6933,7 +6933,7 @@ Perl_newSVpv(pTHX_ const char *s, STRLEN
     register SV *sv;

     new_SV(sv);
-    sv_setpvn(sv,s,len ? len : strlen(s));
+    sv_setpvn(sv,s,len || s == NULL ? len : strlen(s));
     return sv;
 }

• NULL and sv_setpv vs. newSVpv by Yitzchak Scott-Thoennes
• Re: NULL and sv_setpv vs. newSVpv by Rafael Garcia-Suarez