diff options
author | Jeffrey Stedfast <fejj@ximian.com> | 2001-02-07 05:50:37 +0800 |
---|---|---|
committer | Jeffrey Stedfast <fejj@src.gnome.org> | 2001-02-07 05:50:37 +0800 |
commit | 7c6a688ab9a931e69435412d47b432d9c703d2df (patch) | |
tree | 9e53df50012f3df4156da08caa9c9e1ac5359630 | |
parent | b301988ec3c7089112b6346c3bb6b22d2edcdeef (diff) | |
download | gsoc2013-evolution-7c6a688ab9a931e69435412d47b432d9c703d2df.tar gsoc2013-evolution-7c6a688ab9a931e69435412d47b432d9c703d2df.tar.gz gsoc2013-evolution-7c6a688ab9a931e69435412d47b432d9c703d2df.tar.bz2 gsoc2013-evolution-7c6a688ab9a931e69435412d47b432d9c703d2df.tar.lz gsoc2013-evolution-7c6a688ab9a931e69435412d47b432d9c703d2df.tar.xz gsoc2013-evolution-7c6a688ab9a931e69435412d47b432d9c703d2df.tar.zst gsoc2013-evolution-7c6a688ab9a931e69435412d47b432d9c703d2df.zip |
Rewrote using unicode_get_utf8(). (camel_ustrcasecmp): Use unicode_char_t
2001-02-06 Jeffrey Stedfast <fejj@ximian.com>
* camel-search-private.c (utf8_get): Rewrote using unicode_get_utf8().
(camel_ustrcasecmp): Use unicode_char_t
(camel_ustrncasecmp): Same.
svn path=/trunk/; revision=8026
-rw-r--r-- | camel/ChangeLog | 6 | ||||
-rw-r--r-- | camel/camel-mime-filter-charset.c | 4 | ||||
-rw-r--r-- | camel/camel-mime-utils.c | 7 | ||||
-rw-r--r-- | camel/camel-search-private.c | 135 |
4 files changed, 94 insertions, 58 deletions
diff --git a/camel/ChangeLog b/camel/ChangeLog index 2a05238ce1..f69486b8ff 100644 --- a/camel/ChangeLog +++ b/camel/ChangeLog @@ -1,3 +1,9 @@ +2001-02-06 Jeffrey Stedfast <fejj@ximian.com> + + * camel-search-private.c (utf8_get): Rewrote using unicode_get_utf8(). + (camel_ustrcasecmp): Use unicode_char_t + (camel_ustrncasecmp): Same. + 2001-02-06 Christopher James Lahey <clahey@ximian.com> * camel-mime-filter-charset.c (complete, filter), diff --git a/camel/camel-mime-filter-charset.c b/camel/camel-mime-filter-charset.c index a45a318367..6c1f668713 100644 --- a/camel/camel-mime-filter-charset.c +++ b/camel/camel-mime-filter-charset.c @@ -85,7 +85,7 @@ complete(CamelMimeFilter *mf, char *in, size_t len, size_t prespace, char **out, { CamelMimeFilterCharset *f = (CamelMimeFilterCharset *)mf; int converted; - char *inbuf; + const char *inbuf; char *outbuf; int inlen, outlen; @@ -151,7 +151,7 @@ filter(CamelMimeFilter *mf, char *in, size_t len, size_t prespace, char **out, s { CamelMimeFilterCharset *f = (CamelMimeFilterCharset *)mf; int converted; - char *inbuf; + const char *inbuf; char *outbuf; int inlen, outlen; diff --git a/camel/camel-mime-utils.c b/camel/camel-mime-utils.c index ab41258582..a300700548 100644 --- a/camel/camel-mime-utils.c +++ b/camel/camel-mime-utils.c @@ -869,13 +869,14 @@ rfc2047_decode_word(const char *in, int len) { const char *inptr = in+2; const char *inend = in+len-2; + const char *inbuf; char *encname; int tmplen; int ret; char *decword = NULL; char *decoded = NULL; char *outbase = NULL; - char *inbuf, *outbuf; + char *outbuf; int inlen, outlen; iconv_t ic; @@ -1109,7 +1110,7 @@ rfc2047_encode_word(GString *outstring, const char *in, int len, const char *typ iconv_t ic = (iconv_t *)-1; char *buffer, *out, *ascii; size_t inlen, outlen, enclen, bufflen; - char *inptr, *p; + const char *inptr, *p; int first = 1; d(printf("Converting [%d] '%.*s' to %s\n", len, len, in, type)); @@ -1118,7 +1119,7 @@ rfc2047_encode_word(GString *outstring, const char *in, int len, const char *typ bufflen = len*6+16; buffer = alloca(bufflen); inlen = len; - inptr = (char *) in; + inptr = in; ascii = alloca(bufflen); diff --git a/camel/camel-search-private.c b/camel/camel-search-private.c index 1162986af2..3b0daef962 100644 --- a/camel/camel-search-private.c +++ b/camel/camel-search-private.c @@ -50,7 +50,8 @@ A small issue is that case-insenstivity wont work entirely correct for utf8 strings. */ int -camel_search_build_match_regex(regex_t *pattern, camel_search_flags_t type, int argc, struct _ESExpResult **argv, CamelException *ex) +camel_search_build_match_regex (regex_t *pattern, camel_search_flags_t type, int argc, + struct _ESExpResult **argv, CamelException *ex) { GString *match = g_string_new(""); int c, i, count=0, err; @@ -64,11 +65,13 @@ camel_search_build_match_regex(regex_t *pattern, camel_search_flags_t type, int if (argv[i]->type == ESEXP_RES_STRING) { if (count > 0) g_string_append_c(match, '|'); - /* escape any special chars (not sure if this list is complete) */ + word = argv[i]->value.string; if (type & CAMEL_SEARCH_MATCH_REGEX) { + /* no need to escape because this should already be a valid regex */ g_string_append(match, word); } else { + /* escape any special chars (not sure if this list is complete) */ if (type & CAMEL_SEARCH_MATCH_START) g_string_append_c(match, '^'); while ((c = *word++)) { @@ -182,7 +185,10 @@ header_soundex(const char *header, const char *match) return truth; } -static guint16 utf8_get(const char **inp) +#if 0 +/* Why do it this way when the unicode lib already has a function to do this? */ +static unicode_char_t +utf8_get (const char **inp) { guint32 c, v = 0, s, shift; const unsigned char *p = *inp; @@ -216,56 +222,78 @@ static guint16 utf8_get(const char **inp) *inp = p; return v; } +#endif -static const char * -camel_ustrstrcase(const char *haystack, const char *needle) +static unicode_char_t +utf8_get (const char **inp) { - unicode_char_t *uni, *puni, *suni, u, v; - const char *p, *s, *l; - - if (haystack == NULL || needle == NULL) - return NULL; - - if (needle[0] == 0) - return haystack; - - if (haystack[0] == 0) - return NULL; + const unsigned char *p = *inp; + unicode_char_t c; + + g_return_val_if_fail (p != NULL, 0); + + p = unicode_get_utf8 (p, &c); + *inp = p; + + return c; +} - puni = uni = alloca(sizeof(*uni)*(strlen(needle)+1)); +static const char * +camel_ustrstrcase (const char *haystack, const char *needle) +{ + unicode_char_t *nuni, *puni; + unicode_char_t u; + const char *p; + + g_return_val_if_fail (haystack != NULL, NULL); + g_return_val_if_fail (needle != NULL, NULL); + g_return_val_if_fail (strlen (needle) != 0, haystack); + g_return_val_if_fail (strlen (haystack) != 0, NULL); + + puni = nuni = alloca (sizeof (unicode_char_t) * strlen (needle)); + p = needle; - while ((u = utf8_get(&p))) - *puni++ = unicode_tolower(u); - - if (p == NULL) + while ((u = utf8_get (&p))) + *puni++ = unicode_tolower (u); + + /* NULL means there was illegal utf-8 sequence */ + if (!p) return NULL; - - l = p = haystack; - while ( (u = utf8_get(&p)) ) { - v = unicode_tolower(u); - if (uni[0] == v) { - s = p; - suni = uni+1; - while (suni < puni) { - u = utf8_get(&s); - v = unicode_tolower(u); - if (v != *suni) - goto next; - suni++; + + p = haystack; + while ((u = utf8_get (&p))) { + unicode_char_t c; + + c = unicode_tolower (u); + /* We have valid stripped char */ + if (c == nuni[0]) { + const gchar *q = p; + gint npos = 1; + + while (nuni + npos < puni) { + u = utf8_get (&q); + if (!q || !u) + return NULL; + + c = unicode_tolower (u); + if (c != nuni[npos]) + break; + + npos++; } - return l; + + if (nuni + npos == puni) + return p; } - next: - l = p; } - + return NULL; } static int -camel_ustrcasecmp(const char *s1, const char *s2) +camel_ustrcasecmp (const char *s1, const char *s2) { - guint16 u1, u2=0; + unicode_char_t u1, u2 = 0; if (s1 == NULL) { if (s2 == NULL) @@ -275,7 +303,7 @@ camel_ustrcasecmp(const char *s1, const char *s2) } if (s2 == NULL) return 1; - + while ((u1 = utf8_get(&s1)) && (u2 = utf8_get(&s2))) { u1 = unicode_tolower(u1); u2 = unicode_tolower(u2); @@ -302,9 +330,9 @@ camel_ustrcasecmp(const char *s1, const char *s2) } static int -camel_ustrncasecmp(const char *s1, const char *s2, size_t len) +camel_ustrncasecmp (const char *s1, const char *s2, size_t len) { - guint16 u1, u2=0; + unicode_char_t u1, u2 = 0; if (s1 == NULL) { if (s2 == NULL) @@ -314,7 +342,7 @@ camel_ustrncasecmp(const char *s1, const char *s2, size_t len) } if (s2 == NULL) return 1; - + while (len > 0 && (u1 = utf8_get(&s1)) && (u2 = utf8_get(&s2))) { u1 = unicode_tolower(u1); u2 = unicode_tolower(u2); @@ -347,19 +375,20 @@ camel_ustrncasecmp(const char *s1, const char *s2, size_t len) /* searhces for match inside value, if match is mixed case, hten use case-sensitive, else insensitive */ -gboolean camel_search_header_match(const char *value, const char *match, camel_search_match_t how) +gboolean +camel_search_header_match (const char *value, const char *match, camel_search_match_t how) { const char *p; - - if (how == CAMEL_SEARCH_MATCH_SOUNDEX) - return header_soundex(value, match); - - while (*value && isspace(*value)) + + while (*value && isspace (*value)) value++; - - if (strlen(value) < strlen(match)) + + if (how == CAMEL_SEARCH_MATCH_SOUNDEX) + return header_soundex (value, match); + + if (strlen (value) < strlen (match)) return FALSE; - + /* from dan the man, if we have mixed case, perform a case-sensitive match, otherwise not */ p = match; |