From: Stig Venaas Date: Tue, 17 Jul 2001 19:35:23 +0000 (+0000) Subject: Made caseIgnoreSubstringsMatch and caseExactSubstringsMatch use proper X-Git-Tag: LDBM_PRE_GIANT_RWLOCK~1227 X-Git-Url: https://git.sur5r.net/?a=commitdiff_plain;h=6c362d77acf927e997728c510de5872512d190ef;p=openldap Made caseIgnoreSubstringsMatch and caseExactSubstringsMatch use proper Unicode cannonical normalization --- diff --git a/servers/slapd/schema_init.c b/servers/slapd/schema_init.c index c69914ac9b..30ead1ac1a 100644 --- a/servers/slapd/schema_init.c +++ b/servers/slapd/schema_init.c @@ -498,99 +498,6 @@ booleanMatch( return LDAP_SUCCESS; } -#if UTF8MATCH -/* case insensitive UTF8 strncmp with offset for second string */ -static int -UTF8oncasecmp( - struct berval *right, - struct berval *left, - ber_len_t len, - ber_len_t offset ) -{ - ber_len_t r, l; - ber_len_t rlen, llen; - ber_len_t rslen, lslen; - ldap_unicode_t ru, lu; - ldap_unicode_t ruu, luu; - - rslen = len < right->bv_len ? len : right->bv_len; - lslen = len + offset < left->bv_len ? len : left->bv_len; - - for( r = 0, l = offset; - r < rslen && l < lslen; - r+=rlen, l+=llen ) - { - /* - * XXYYZ: we convert to ucs4 even though -llunicode - * expects ucs2 in an unsigned long - */ - ru = ldap_utf8_to_ucs4( &right->bv_val[r] ); - if( ru == LDAP_UCS4_INVALID ) { - return 1; - } - - lu = ldap_utf8_to_ucs4( &left->bv_val[l] ); - if( lu == LDAP_UCS4_INVALID ) { - return -1; - } - - ruu = uctoupper( ru ); - luu = uctoupper( lu ); - - if( ruu > luu ) { - return 1; - } else if( luu > ruu ) { - return -1; - } - - rlen = LDAP_UTF8_CHARLEN( &right->bv_val[r] ); - llen = LDAP_UTF8_CHARLEN( &left->bv_val[l] ); - } - - if( r < rslen ) { - /* less left */ - return -1; - } - - if( l < lslen ) { - /* less right */ - return 1; - } - - return 0; -} - -static char *UTF8casechr( const char *str, const char *c ) -{ - char *p, *lower, *upper; - ldap_ucs4_t tch, ch = ldap_utf8_to_ucs4(c); - - tch = uctolower ( ch ); - for( p = (char *) str; *p != '\0'; LDAP_UTF8_INCR(p) ) { - if( ldap_utf8_to_ucs4( p ) == tch ) { - break; - } - } - lower = *p != '\0' ? p : NULL; - - tch = uctoupper ( ch ); - for( p = (char *) str; *p != '\0'; LDAP_UTF8_INCR(p) ) { - if( ldap_utf8_to_ucs4( p ) == tch ) { - break; - } - } - upper = *p != '\0' ? p : NULL; - - if( lower && upper ) { - return lower < upper ? lower : upper; - } else if ( lower ) { - return lower; - } else { - return upper; - } -} -#endif - static int UTF8StringValidate( Syntax *syntax, @@ -1075,10 +982,31 @@ caseExactSubstringsMatch( void *assertedValue ) { int match = 0; - SubstringsAssertion *sub = assertedValue; - struct berval left = *value; + SubstringsAssertion *sub; + struct berval left; int i; ber_len_t inlen=0; +#if UTF8MATCH + char *nav; +#endif + +#if UTF8MATCH + nav = UTF8normalize( value->bv_val, UTF8_NOCASEFOLD ); + if( nav == NULL ) { + match = 1; + goto done; + } + left.bv_val = nav; + left.bv_len = strlen( nav ); + sub = UTF8SubstringsassertionNormalize( assertedValue, UTF8_NOCASEFOLD ); +#else + left = *value; + sub = assertedValue; +#endif + if( sub == NULL ) { + match = -1; + goto done; + } /* Add up asserted input length */ if( sub->sa_initial ) { @@ -1186,6 +1114,15 @@ retry: } done: +#if UTF8MATCH + free( nav ); + if( sub != NULL ) { + ch_free( sub->sa_final ); + ber_bvecfree( sub->sa_any ); + ch_free( sub->sa_initial ); + ch_free( sub ); + } +#endif *matchp = match; return LDAP_SUCCESS; } @@ -1711,10 +1648,31 @@ caseIgnoreSubstringsMatch( void *assertedValue ) { int match = 0; - SubstringsAssertion *sub = assertedValue; - struct berval left = *value; + SubstringsAssertion *sub; + struct berval left; int i; ber_len_t inlen=0; +#if UTF8MATCH + char *nav; +#endif + +#if UTF8MATCH + nav = UTF8normalize( value->bv_val, UTF8_CASEFOLD ); + if( nav == NULL ) { + match = 1; + goto done; + } + left.bv_val = nav; + left.bv_len = strlen( nav ); + sub = UTF8SubstringsassertionNormalize( assertedValue, UTF8_CASEFOLD ); +#else + left = *value; + sub = assertedValue; +#endif + if( sub == NULL ) { + match = -1; + goto done; + } /* Add up asserted input length */ if( sub->sa_initial ) { @@ -1736,8 +1694,8 @@ caseIgnoreSubstringsMatch( } #if UTF8MATCH - match = UTF8oncasecmp( sub->sa_initial, &left, - sub->sa_initial->bv_len, 0 ); + match = strncmp( sub->sa_initial->bv_val, left.bv_val, + sub->sa_initial->bv_len ); #else match = strncasecmp( sub->sa_initial->bv_val, left.bv_val, sub->sa_initial->bv_len ); @@ -1759,9 +1717,9 @@ caseIgnoreSubstringsMatch( } #if UTF8MATCH - match = UTF8oncasecmp( sub->sa_final, &left, - sub->sa_final->bv_len, - left.bv_len - sub->sa_final->bv_len ); + match = strncmp( sub->sa_final->bv_val, + &left.bv_val[left.bv_len - sub->sa_final->bv_len], + sub->sa_final->bv_len ); #else match = strncasecmp( sub->sa_final->bv_val, &left.bv_val[left.bv_len - sub->sa_final->bv_len], @@ -1793,7 +1751,7 @@ retry: } #if UTF8MATCH - p = UTF8casechr( left.bv_val, sub->sa_any[i]->bv_val ); + p = strchr( left.bv_val, *sub->sa_any[i]->bv_val ); #else p = strcasechr( left.bv_val, *sub->sa_any[i]->bv_val ); #endif @@ -1821,19 +1779,14 @@ retry: } #if UTF8MATCH - match = UTF8oncasecmp( &left, sub->sa_any[i], - sub->sa_any[i]->bv_len, 0 ); - - if( match != 0 ) { - int len = LDAP_UTF8_CHARLEN( left.bv_val ); - left.bv_val += len; - left.bv_len -= len; - goto retry; - } + match = strncmp( left.bv_val, + sub->sa_any[i]->bv_val, + sub->sa_any[i]->bv_len ); #else match = strncasecmp( left.bv_val, sub->sa_any[i]->bv_val, sub->sa_any[i]->bv_len ); +#endif if( match != 0 ) { left.bv_val++; @@ -1841,7 +1794,6 @@ retry: goto retry; } -#endif left.bv_val += sub->sa_any[i]->bv_len; left.bv_len -= sub->sa_any[i]->bv_len; @@ -1850,6 +1802,15 @@ retry: } done: +#if UTF8MATCH + free( nav ); + if( sub != NULL ) { + ch_free( sub->sa_final ); + ber_bvecfree( sub->sa_any ); + ch_free( sub->sa_initial ); + ch_free( sub ); + } +#endif *matchp = match; return LDAP_SUCCESS; }