Fix text_position to not scan past end of source string in multibyte

case, per report from Korea PostgreSQL Users' Group.

diff --git a/src/backend/utils/adt/varlena.c b/src/backend/utils/adt/varlena.c
index 50f6bf1cb780f3f6abea8d881c59738763ec4f1a..caf0250e886a93b1527e319c3e4c90260c3b9a5c 100644 (file)
--- a/src/backend/utils/adt/varlena.c
+++ b/src/backend/utils/adt/varlena.c
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *       $Header: /cvsroot/pgsql/src/backend/utils/adt/varlena.c,v 1.106.2.2 2003/12/19 04:56:54 tgl Exp $
+ *       $Header: /cvsroot/pgsql/src/backend/utils/adt/varlena.c,v 1.106.2.3 2004/01/31 00:45:34 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -761,9 +761,6 @@ text_position(Datum str, Datum search_str, int matchnum)
        len1 = (VARSIZE(t1) - VARHDRSZ);
        len2 = (VARSIZE(t2) - VARHDRSZ);
 
-       /* no use in searching str past point where search_str will fit */
-       px = (len1 - len2);
-
        if (eml == 1)                           /* simple case - single byte encoding */
        {
                char       *p1,
@@ -772,6 +769,9 @@ text_position(Datum str, Datum search_str, int matchnum)
                p1 = VARDATA(t1);
                p2 = VARDATA(t2);
 
+               /* no use in searching str past point where search_str will fit */
+               px = (len1 - len2);
+
                for (p = 0; p <= px; p++)
                {
                        if ((*p2 == *p1) && (strncmp(p1, p2, len2) == 0))
@@ -799,6 +799,9 @@ text_position(Datum str, Datum search_str, int matchnum)
                (void) pg_mb2wchar_with_len((unsigned char *) VARDATA(t2), p2, len2);
                len2 = pg_wchar_strlen(p2);
 
+               /* no use in searching str past point where search_str will fit */
+               px = (len1 - len2);
+
                for (p = 0; p <= px; p++)
                {
                        if ((*p2 == *p1) && (pg_wchar_strncmp(p1, p2, len2) == 0))