Change 29953 by [EMAIL PROTECTED] on 2007/01/24 17:01:34
Integrate:
[ 27792]
Remove SAVEt_FREESHAREDPV, as nothing is using it, and it isn't catered
for in Perl_ss_dup, hence nothing is testing it and nothing will alert
us if it breaks.
[ 27866]
Add Dave's explainations of why certain pointers are always non-NULL,
and assert() that it is true.
[ 27869]
Subject: [PATCH] (tentative) regcomp.c: try convincing Coverity that
data ptr in study_chunk is non-NULL when needed
From: [EMAIL PROTECTED] (Jarkko Hietaniemi)
Message-Id: <[EMAIL PROTECTED]>
Date: Sun, 16 Apr 2006 12:26:48 +0300 (EEST)
[ 27894]
Revert 27869 at Jarkko's request.
Affected files ...
... //depot/maint-5.8/perl/regcomp.c#85 integrate
... //depot/maint-5.8/perl/regexec.c#73 edit
... //depot/maint-5.8/perl/scope.c#60 integrate
... //depot/maint-5.8/perl/scope.h#22 integrate
Differences ...
==== //depot/maint-5.8/perl/regexec.c#73 (text) ====
Index: perl/regexec.c
--- perl/regexec.c#72~29949~ 2007-01-24 07:28:38.000000000 -0800
+++ perl/regexec.c 2007-01-24 09:01:34.000000000 -0800
@@ -3129,6 +3129,18 @@
/* No need to save/restore up to this paren */
I32 parenfloor = scan->flags;
+ /* Dave says:
+
+ CURLYX and WHILEM are always paired: they're the moral
+ equivalent of pp_enteriter anbd pp_iter.
+
+ The only time next could be null is if the node tree is
+ corrupt. This was mentioned on p5p a few days ago.
+
+ See
http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html
+ So we'll assert that this is true:
+ */
+ assert(next);
if (next && (OP(PREVOPER(next)) == NOTHING)) /* LONGJMP */
next += ARG(next);
cc.oldcc = PL_regcc;
@@ -3152,7 +3164,18 @@
saySAME(n);
}
/* NOTREACHED */
- case WHILEM: {
+ case WHILEM:
+ /* Dave says:
+
+ st->cc gets initialised by CURLYX ready for use by WHILEM.
+ So again, unless somethings been corrupted, st->cc cannot
+ be null at that point in WHILEM.
+
+ See
http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2006-04/msg00556.html
+ So we'll assert that this is true:
+ */
+ assert(PL_regcc);
+ {
/*
* This is really hard to understand, because after we match
* what we're trying to match, we must make sure the rest of
End of Patch.
