(ITS#3538) Substring filter normalization

[weber@bucknell.edu]

Full_Name: Chris Weber
Version: 2.2.23
OS: Solaris 9
URL: ftp://ftp.openldap.org/incoming/
Submission from: (NULL) (134.82.42.1)


  I feel that the way in which OpenLDAP 2.2.23 normalizes substring filters is
probably incorrect.  At present, substring filters are normalized like:

     (cn=* ab) => (cn=*ab)
     (cn=ab *) => (cn=ab*)
     (cn=* ab *) => (cn=*ab*)
     (cn=* ab * cd *) => (cn=*ab*cd*)

  This treatment is due to each of a substring filter's "initial", "any", and
"final" strings being individually normalized out of context, as though they
were each an entire string rather than internal substrings of a larger string.

  I've applied the following patch locally to restore the expected behavior. 
The changes are to:

               schema_init.c:UTF8StringNormalize

and cause substring filter parts to be normalized in their proper contexts. 
Perhaps there's a better place or more general way to accomplish this within
OpenLDAP's source tree, but this is straightforward and seems to have no other
consequences.


--- openldap-2.2.23/servers/slapd/schema_init.c	2005-01-20 12:01:09.000000000
-0500
+++ openldap-2.2.23BU/servers/slapd/schema_init.c	2005-02-12 09:33:19.568860000
-0500
@@ -1361,7 +1361,10 @@
 	nvalue.bv_len = 0;
 	nvalue.bv_val = tmp.bv_val;
 
-	wasspace=1; /* trim leading spaces */
+	wasspace = ( ( ( use & SLAP_MR_SUBSTR_ANY ) == SLAP_MR_SUBSTR_ANY )
+		|| ( ( use & SLAP_MR_SUBSTR_FINAL ) == SLAP_MR_SUBSTR_FINAL ) )
+		? 0 : 1; /* trim leading spaces? */
+
 	for( i=0; i<tmp.bv_len; i++) {
 		if ( ASCII_SPACE( tmp.bv_val[i] )) {
 			if( wasspace++ == 0 ) {
@@ -1375,7 +1378,9 @@
 	}
 
 	if( nvalue.bv_len ) {
-		if( wasspace ) {
+		if( wasspace
+			&& ( ( use & SLAP_MR_SUBSTR_INITIAL ) != SLAP_MR_SUBSTR_INITIAL )
+			&& ( ( use & SLAP_MR_SUBSTR_ANY ) != SLAP_MR_SUBSTR_ANY ) ) {
 			/* last character was a space, trim it */
 			--nvalue.bv_len;
 		}