Re: PATCH: zsh/regex and =~

[Phil Pennock <zsh-workers+phil.pennock@xxxxxxxxxxxx>]

On 2007-04-28 at 00:56 -0700, Phil Pennock wrote:
> The attached patch and files, which includes documentation, adds a new
> loadable module, zsh/regex.  I've not examined widechar issues and which
> regex libraries actually do handle these.  I've not looked at linkage
> issues on platforms where regex (the POSIX interface, not regexp) is not
> a part of libc.

I noticed a gcc complaint that a variable might be used uninitialised;
this was bogus, but understandable.  The first patch below fixes it.
The second patch is a spelling correction in the docs.  Alternatively,
it might be a sign that the option needs to be renamed ...

arr is only initialised if nelem.  Later there are two references to it.
The second is guarded by "if (nelem)"; the first is guarded by
"if (isset(BASHREMATCH))".  If BASHREMATCH is set, nelem is always at
least 1.  Assuming re.re_nsub is never negative, which it isn't.  And
the patch also affirms this for sheer paranoia's sake.

-Phil

--- Src/Modules/regex.c.old	Sat Apr 28 17:40:12 2007
+++ Src/Modules/regex.c	Sat Apr 28 17:43:27 2007
@@ -75,6 +75,11 @@ zcond_regex_match(char **a, int id)
 	/* re.re_nsub is number of parenthesized groups, we also need
 	 * 1 for the 0 offset, which is the entire matched portion
 	 */
+	if (re.re_nsub < 0) {
+	    zwarn("INTERNAL ERROR: regcomp() returned "
+		    "negative subpattern count %d", re.re_nsub);
+	    break;
+	}
 	matchessz = (re.re_nsub + 1) * sizeof(regmatch_t);
 	matches = zalloc(matchessz);
 	r = regexec(&re, lhstr, re.re_nsub+1, matches, reflags);
@@ -88,6 +93,7 @@ zcond_regex_match(char **a, int id)
 		start = 1;
 		nelem = re.re_nsub;
 	    }
+	    arr = NULL; /* bogus gcc warning of used uninitialised */
 	    /* entire matched portion + re_nsub substrings + NULL */
 	    if (nelem) {
 		arr = x = (char **) zalloc(sizeof(char *) * (nelem + 1));

--- Doc/Zsh/mod_regex.yo.old	Sat Apr 28 17:45:57 2007
+++ Doc/Zsh/mod_regex.yo	Sat Apr 28 17:45:35 2007
@@ -15,7 +15,7 @@
 
 [[ alphabetical -regex-match ^a([^a]+)a([^a]+)a ]] && print -l $MATCH X $match
 
-If tt(REGMATCH_PCRE) is not set, then the tt(=~) operator will automatically
+If tt(REMATCH_PCRE) is not set, then the tt(=~) operator will automatically
 load this module as needed and will invoke the tt(-regex-match) operator.
 
 If tt(BASH_REMATCH) is set, then tt($BASH_REMATCH) will be set instead of