From f804c46da690ff8f394eac42bf9ba85f5d05fd88 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Julio=20S=C3=A1nchez=20Fern=C3=A1ndez?= Date: Mon, 31 May 1999 11:23:41 +0000 Subject: [PATCH] Cosmetic/spelling fixes. --- libraries/libldap/schema.c | 33 +++++++++++++++++++++------------ 1 file changed, 21 insertions(+), 12 deletions(-) diff --git a/libraries/libldap/schema.c b/libraries/libldap/schema.c index f0a22f81d2..cfe7c5ec20 100644 --- a/libraries/libldap/schema.c +++ b/libraries/libldap/schema.c @@ -22,7 +22,7 @@ /* * When pretty printing the entities we will be appending to a buffer. * Since checking for overflow, realloc'ing and checking if no error - * is extremely boring, we will use a pretection layer that will let + * is extremely boring, we will use a protection layer that will let * us blissfully ignore the error until the end. This layer is * implemented with the help of the next type. */ @@ -156,7 +156,7 @@ print_qdescrlist(safe_string *ss, char **sa) ret = print_qdescr(ss,*sp); } /* If the list was empty, we return zero that is potentially - * incorrect, but since we will still appending things, the + * incorrect, but since we will be still appending things, the * overflow will be detected later. Maybe FIX. */ return(ret); @@ -167,7 +167,9 @@ print_qdescrs(safe_string *ss, char **sa) { /* The only way to represent an empty list is as a qdescrlist * so, if the list is empty we treat it as a long list. - * Really, this is what the syntax mandates. + * Really, this is what the syntax mandates. We should not + * be here if the list was empty, but if it happens, a label + * has already been output and we cannot undo it. */ if ( !sa[0] || ( sa[0] && sa[1] ) ) { print_whsp(ss); @@ -455,8 +457,8 @@ charray_free( char **array ) #define TK_QDESCR TK_QDSTRING struct token { - int type; - char *sval; + int type; + char *sval; }; static int @@ -554,7 +556,10 @@ parse_numericoid(char **sp, int *code) /* Each iteration of this loops gets one decimal string */ while (**sp) { if ( !isdigit(**sp) ) { - /* Initial char is not a digit or char after dot is not a digit */ + /* + * Initial char is not a digit or char after dot is + * not a digit + */ *code = LDAP_SCHERR_NODIGIT; return NULL; } @@ -566,7 +571,7 @@ parse_numericoid(char **sp, int *code) /* Otherwise, gobble the dot and loop again */ (*sp)++; } - /* At this point, *sp points at the char past the numericoid. Perfect. */ + /* Now *sp points at the char past the numericoid. Perfect. */ len = *sp - start; res = LDAP_MALLOC(len+1); if (!res) { @@ -853,7 +858,7 @@ ldap_str2attributetype( char * s, int * code, char ** errp ) parse_whsp(&ss); /* - * Beyond this point we will be liberal an accept the items + * Beyond this point we will be liberal and accept the items * in any order. */ while (1) { @@ -1027,13 +1032,17 @@ ldap_str2attributetype( char * s, int * code, char ** errp ) return NULL; } if ( !strcasecmp(sval,"userApplications") ) - at->at_usage = LDAP_SCHEMA_USER_APPLICATIONS; + at->at_usage = + LDAP_SCHEMA_USER_APPLICATIONS; else if ( !strcasecmp(sval,"directoryOperation") ) - at->at_usage = LDAP_SCHEMA_DIRECTORY_OPERATION; + at->at_usage = + LDAP_SCHEMA_DIRECTORY_OPERATION; else if ( !strcasecmp(sval,"distributedOperation") ) - at->at_usage = LDAP_SCHEMA_DISTRIBUTED_OPERATION; + at->at_usage = + LDAP_SCHEMA_DISTRIBUTED_OPERATION; else if ( !strcasecmp(sval,"dSAOperation") ) - at->at_usage = LDAP_SCHEMA_DSA_OPERATION; + at->at_usage = + LDAP_SCHEMA_DSA_OPERATION; else { *code = LDAP_SCHERR_UNEXPTOKEN; *errp = ss; -- 2.39.5