libwine: Add helper function to compare character weights.
Signed-off-by: Piotr Caban <piotr@codeweavers.com> Signed-off-by: Alexandre Julliard <julliard@winehq.org>
This commit is contained in:
parent
1eca63440d
commit
ac5ac588f3
|
@ -153,11 +153,37 @@ int wine_get_sortkey(int flags, const WCHAR *src, int srclen, char *dst, int dst
|
|||
return key_ptr[3] - dst;
|
||||
}
|
||||
|
||||
static inline int compare_unicode_weights(int flags, const WCHAR *str1, int len1,
|
||||
const WCHAR *str2, int len2)
|
||||
enum weight
|
||||
{
|
||||
UNICODE_WEIGHT,
|
||||
DIACRITIC_WEIGHT,
|
||||
CASE_WEIGHT
|
||||
};
|
||||
|
||||
static unsigned int get_weight(WCHAR ch, enum weight type)
|
||||
{
|
||||
unsigned int ret;
|
||||
|
||||
ret = collation_table[collation_table[ch >> 8] + (ch & 0xff)];
|
||||
if (ret == (unsigned int)-1)
|
||||
return ch;
|
||||
|
||||
switch(type)
|
||||
{
|
||||
case UNICODE_WEIGHT:
|
||||
return ret >> 16;
|
||||
case DIACRITIC_WEIGHT:
|
||||
return (ret >> 8) & 0xff;
|
||||
case CASE_WEIGHT:
|
||||
default:
|
||||
return (ret >> 4) & 0x0f;
|
||||
}
|
||||
}
|
||||
|
||||
static inline int compare_weights(int flags, const WCHAR *str1, int len1,
|
||||
const WCHAR *str2, int len2, enum weight type)
|
||||
{
|
||||
unsigned int ce1, ce2;
|
||||
int ret;
|
||||
|
||||
/* 32-bit collation element table format:
|
||||
* unicode weight - high 16 bit, diacritic weight - high 8 bit of low 16 bit,
|
||||
|
@ -187,7 +213,7 @@ static inline int compare_unicode_weights(int flags, const WCHAR *str1, int len1
|
|||
/* hyphen and apostrophe are treated differently depending on
|
||||
* whether SORT_STRINGSORT specified or not
|
||||
*/
|
||||
if (!(flags & SORT_STRINGSORT))
|
||||
if (type == UNICODE_WEIGHT && !(flags & SORT_STRINGSORT))
|
||||
{
|
||||
if (*str1 == '-' || *str1 == '\'')
|
||||
{
|
||||
|
@ -206,133 +232,10 @@ static inline int compare_unicode_weights(int flags, const WCHAR *str1, int len1
|
|||
}
|
||||
}
|
||||
|
||||
ce1 = collation_table[collation_table[*str1 >> 8] + (*str1 & 0xff)];
|
||||
ce2 = collation_table[collation_table[*str2 >> 8] + (*str2 & 0xff)];
|
||||
ce1 = get_weight(*str1, type);
|
||||
ce2 = get_weight(*str2, type);
|
||||
|
||||
if (ce1 != (unsigned int)-1 && ce2 != (unsigned int)-1)
|
||||
ret = (ce1 >> 16) - (ce2 >> 16);
|
||||
else
|
||||
ret = *str1 - *str2;
|
||||
|
||||
if (ret) return ret;
|
||||
|
||||
str1++;
|
||||
str2++;
|
||||
len1--;
|
||||
len2--;
|
||||
}
|
||||
while (len1 && !*str1)
|
||||
{
|
||||
str1++;
|
||||
len1--;
|
||||
}
|
||||
while (len2 && !*str2)
|
||||
{
|
||||
str2++;
|
||||
len2--;
|
||||
}
|
||||
return len1 - len2;
|
||||
}
|
||||
|
||||
static inline int compare_diacritic_weights(int flags, const WCHAR *str1, int len1,
|
||||
const WCHAR *str2, int len2)
|
||||
{
|
||||
unsigned int ce1, ce2;
|
||||
int ret;
|
||||
|
||||
/* 32-bit collation element table format:
|
||||
* unicode weight - high 16 bit, diacritic weight - high 8 bit of low 16 bit,
|
||||
* case weight - high 4 bit of low 8 bit.
|
||||
*/
|
||||
while (len1 > 0 && len2 > 0)
|
||||
{
|
||||
if (flags & NORM_IGNORESYMBOLS)
|
||||
{
|
||||
int skip = 0;
|
||||
/* FIXME: not tested */
|
||||
if (get_char_typeW(*str1) & (C1_PUNCT | C1_SPACE))
|
||||
{
|
||||
str1++;
|
||||
len1--;
|
||||
skip = 1;
|
||||
}
|
||||
if (get_char_typeW(*str2) & (C1_PUNCT | C1_SPACE))
|
||||
{
|
||||
str2++;
|
||||
len2--;
|
||||
skip = 1;
|
||||
}
|
||||
if (skip) continue;
|
||||
}
|
||||
|
||||
ce1 = collation_table[collation_table[*str1 >> 8] + (*str1 & 0xff)];
|
||||
ce2 = collation_table[collation_table[*str2 >> 8] + (*str2 & 0xff)];
|
||||
|
||||
if (ce1 != (unsigned int)-1 && ce2 != (unsigned int)-1)
|
||||
ret = ((ce1 >> 8) & 0xff) - ((ce2 >> 8) & 0xff);
|
||||
else
|
||||
ret = *str1 - *str2;
|
||||
|
||||
if (ret) return ret;
|
||||
|
||||
str1++;
|
||||
str2++;
|
||||
len1--;
|
||||
len2--;
|
||||
}
|
||||
while (len1 && !*str1)
|
||||
{
|
||||
str1++;
|
||||
len1--;
|
||||
}
|
||||
while (len2 && !*str2)
|
||||
{
|
||||
str2++;
|
||||
len2--;
|
||||
}
|
||||
return len1 - len2;
|
||||
}
|
||||
|
||||
static inline int compare_case_weights(int flags, const WCHAR *str1, int len1,
|
||||
const WCHAR *str2, int len2)
|
||||
{
|
||||
unsigned int ce1, ce2;
|
||||
int ret;
|
||||
|
||||
/* 32-bit collation element table format:
|
||||
* unicode weight - high 16 bit, diacritic weight - high 8 bit of low 16 bit,
|
||||
* case weight - high 4 bit of low 8 bit.
|
||||
*/
|
||||
while (len1 > 0 && len2 > 0)
|
||||
{
|
||||
if (flags & NORM_IGNORESYMBOLS)
|
||||
{
|
||||
int skip = 0;
|
||||
/* FIXME: not tested */
|
||||
if (get_char_typeW(*str1) & (C1_PUNCT | C1_SPACE))
|
||||
{
|
||||
str1++;
|
||||
len1--;
|
||||
skip = 1;
|
||||
}
|
||||
if (get_char_typeW(*str2) & (C1_PUNCT | C1_SPACE))
|
||||
{
|
||||
str2++;
|
||||
len2--;
|
||||
skip = 1;
|
||||
}
|
||||
if (skip) continue;
|
||||
}
|
||||
|
||||
ce1 = collation_table[collation_table[*str1 >> 8] + (*str1 & 0xff)];
|
||||
ce2 = collation_table[collation_table[*str2 >> 8] + (*str2 & 0xff)];
|
||||
|
||||
if (ce1 != (unsigned int)-1 && ce2 != (unsigned int)-1)
|
||||
ret = ((ce1 >> 4) & 0x0f) - ((ce2 >> 4) & 0x0f);
|
||||
else
|
||||
ret = *str1 - *str2;
|
||||
|
||||
if (ret) return ret;
|
||||
if (ce1 - ce2) return ce1 - ce2;
|
||||
|
||||
str1++;
|
||||
str2++;
|
||||
|
@ -357,13 +260,13 @@ int wine_compare_string(int flags, const WCHAR *str1, int len1,
|
|||
{
|
||||
int ret;
|
||||
|
||||
ret = compare_unicode_weights(flags, str1, len1, str2, len2);
|
||||
ret = compare_weights(flags, str1, len1, str2, len2, UNICODE_WEIGHT);
|
||||
if (!ret)
|
||||
{
|
||||
if (!(flags & NORM_IGNORENONSPACE))
|
||||
ret = compare_diacritic_weights(flags, str1, len1, str2, len2);
|
||||
ret = compare_weights(flags, str1, len1, str2, len2, DIACRITIC_WEIGHT);
|
||||
if (!ret && !(flags & NORM_IGNORECASE))
|
||||
ret = compare_case_weights(flags, str1, len1, str2, len2);
|
||||
ret = compare_weights(flags, str1, len1, str2, len2, CASE_WEIGHT);
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
|
|
Loading…
Reference in New Issue