Skip to content

M
MariaDB
Commit 0259b3cb authored by Alexander Barkov's avatar Alexander Barkov
Browse files
Options

MDEV-11255 LDML: allow defining 2-level UCA collations

parent 90c5b2f5
Show whitespace changes
Inline Side-by-side
Showing with 2000 additions and 90 deletions
mysql-test/std_data/ldml/Index.xml
View file @ 0259b3cb
......@@ -1167,6 +1167,45 @@
</rules>
</collation>
<collation name="utf8_czech_test_w2" id="370" version="5.2.0">
<settings strength="2"/>
<rules>
<reset>C</reset><p>\u010D</p><t>\u010C</t>
<reset>H</reset><p>ch</p><t>Ch</t><t>CH</t>
<reset>R</reset><p>\u0159</p><t>\u0158</t>
<reset>S</reset><p>\u0161</p><t>\u0160</t>
<reset>Z</reset><p>\u017E</p><t>\u017D</t>
</rules>
</collation>
<collation name="utf8_czech_test_nopad_w2" id="371" version="5.2.0" flag="nopad">
<settings strength="2"/>
<rules>
<reset>C</reset><p>\u010D</p><t>\u010C</t>
<reset>H</reset><p>ch</p><t>Ch</t><t>CH</t>
<reset>R</reset><p>\u0159</p><t>\u0158</t>
<reset>S</reset><p>\u0161</p><t>\u0160</t>
<reset>Z</reset><p>\u017E</p><t>\u017D</t>
</rules>
</collation>
<!--
This collation definition is bad.
It uses Unicode-4.0.0 (the default version), and requests strength="2".
Unicode-4.0.0 does not have information about the secondary weight level.
The version="5.2.0" collation attribute was forgotten in this definition.
-->
<collation name="utf8_czech_test_bad_w2" id="372">
<settings strength="2"/>
<rules>
<reset>C</reset><p>\u010D</p><t>\u010C</t>
<reset>H</reset><p>ch</p><t>Ch</t><t>CH</t>
<reset>R</reset><p>\u0159</p><t>\u0158</t>
<reset>S</reset><p>\u0161</p><t>\u0160</t>
<reset>Z</reset><p>\u017E</p><t>\u017D</t>
</rules>
</collation>
</charset>
</charsets>
mysql-test/t/ctype_ldml.test
View file @ 0259b3cb
......@@ -499,3 +499,106 @@ SELECT HEX(a), REPLACE(a,' ','<SP>') FROM t1 WHERE a='a';
SELECT HEX(a), REPLACE(a,' ','<SP>') FROM t1 ORDER BY a;
SELECT HEX(a), REPLACE(a,' ','<SP>') FROM t1 ORDER BY a DESC;
DROP TABLE t1;
SET NAMES utf8 COLLATE utf8_czech_test_w2;
CREATE TABLE t1 AS SELECT SPACE(10) AS c1 LIMIT 0;
--source include/ctype_unicode_latin.inc
INSERT INTO t1 VALUES ('a ');
SELECT c1, HEX(WEIGHT_STRING(c1 LEVEL 1)), HEX(WEIGHT_STRING(c1 LEVEL 2)) FROM t1 ORDER BY c1, BINARY c1;
SELECT c1, HEX(WEIGHT_STRING(c1 AS CHAR(3) LEVEL 1)), HEX(WEIGHT_STRING(c1 AS CHAR(3) LEVEL 2)) FROM t1 WHERE c1 BETWEEN 'a' AND 'aZ' ORDER BY c1, BINARY c1;
DROP TABLE t1;
SELECT 'a' = 'a ';
SELECT 'a' < 'á';
SELECT 'áa' < 'ab';
SELECT 'á' < 'ä';
SELECT 'äa' < 'áb';
SELECT 'c' < 'č';
SELECT 'cb' < 'ča';
SELECT 'd' < 'ď';
SELECT 'ďa' < 'db';
SELECT 'e' < 'é';
SELECT 'éa' < 'eb';
SELECT 'é' < 'ě';
SELECT 'ěa' < 'éb';
SELECT 'i' < 'í';
SELECT 'ía' < 'ib';
SELECT 'n' < 'ň';
SELECT 'ňa' < 'nb';
SELECT 'o' < 'ó';
SELECT 'óa' < 'ob';
SELECT 'ó' < 'ö';
SELECT 'öa' < 'ób';
SELECT 'r' < 'ř';
SELECT 'rb' < 'řa';
SELECT 's' < 'š';
SELECT 'sb' < 'ša';
SELECT 't' < 'ť';
SELECT 'ťa' < 'tb';
SELECT 'u' < 'ú';
SELECT 'úa' < 'ub';
SELECT 'ú' < 'ů';
SELECT 'ůa' < 'úb';
SELECT 'ů' < 'ü';
SELECT 'üa' < 'ůb';
SELECT 'y' < 'ý';
SELECT 'ýa' < 'yb';
SELECT 'z' < 'ž';
SELECT 'zb' < 'ža';
SELECT 'hž' < 'ch';
SELECT 'chž'< 'i';
SET NAMES utf8 COLLATE utf8_czech_test_nopad_w2;
CREATE TABLE t1 AS SELECT SPACE(10) AS c1 LIMIT 0;
--source include/ctype_unicode_latin.inc
INSERT INTO t1 VALUES ('a ');
SELECT c1, HEX(WEIGHT_STRING(c1 LEVEL 1)), HEX(WEIGHT_STRING(c1 LEVEL 2)) FROM t1 ORDER BY c1, BINARY c1;
SELECT c1, HEX(WEIGHT_STRING(c1 AS CHAR(3) LEVEL 1)), HEX(WEIGHT_STRING(c1 AS CHAR(3) LEVEL 2)) FROM t1 WHERE c1 BETWEEN 'a' AND 'aZ' ORDER BY c1, BINARY c1;
DROP TABLE t1;
SELECT 'a' = 'a ';
SELECT 'a' < 'á';
SELECT 'áa' < 'ab';
SELECT 'á' < 'ä';
SELECT 'äa' < 'áb';
SELECT 'c' < 'č';
SELECT 'cb' < 'ča';
SELECT 'd' < 'ď';
SELECT 'ďa' < 'db';
SELECT 'e' < 'é';
SELECT 'éa' < 'eb';
SELECT 'é' < 'ě';
SELECT 'ěa' < 'éb';
SELECT 'i' < 'í';
SELECT 'ía' < 'ib';
SELECT 'n' < 'ň';
SELECT 'ňa' < 'nb';
SELECT 'o' < 'ó';
SELECT 'óa' < 'ob';
SELECT 'ó' < 'ö';
SELECT 'öa' < 'ób';
SELECT 'r' < 'ř';
SELECT 'rb' < 'řa';
SELECT 's' < 'š';
SELECT 'sb' < 'ša';
SELECT 't' < 'ť';
SELECT 'ťa' < 'tb';
SELECT 'u' < 'ú';
SELECT 'úa' < 'ub';
SELECT 'ú' < 'ů';
SELECT 'ůa' < 'úb';
SELECT 'ů' < 'ü';
SELECT 'üa' < 'ůb';
SELECT 'y' < 'ý';
SELECT 'ýa' < 'yb';
SELECT 'z' < 'ž';
SELECT 'zb' < 'ža';
SELECT 'hž' < 'ch';
SELECT 'chž'< 'i';
--error ER_UNKNOWN_COLLATION
SELECT 'a' COLLATE utf8_czech_test_bad_w2;
mysys/charset.c
View file @ 0259b3cb
......@@ -258,12 +258,38 @@ static my_bool simple_cs_is_full(CHARSET_INFO *cs)
#if defined(HAVE_UCA_COLLATIONS) && (defined(HAVE_CHARSET_ucs2) || defined(HAVE_CHARSET_utf8))
/**
Initialize a loaded collation.
@param [OUT] to - The new charset_info_st structure to initialize.
@param [IN] from - A template collation, to fill the missing data from.
@param [IN] loaded - The collation data loaded from the LDML file.
some data may be missing in "loaded".
*/
static void
copy_uca_collation(struct charset_info_st *to, CHARSET_INFO *from)
copy_uca_collation(struct charset_info_st *to, CHARSET_INFO *from,
CHARSET_INFO *loaded)
{
to->cset= from->cset;
to->coll= from->coll;
to->strxfrm_multiply= from->strxfrm_multiply;
/*
Single-level UCA collation have strnxfrm_multiple=8.
In case of a multi-level UCA collation we use strnxfrm_multiply=4.
That means MY_COLLATION_HANDLER::strnfrmlen() will request the caller
to allocate a buffer smaller size for each level, for performance purpose,
and to fit longer VARCHARs to @@max_sort_length.
This makes filesort produce non-precise order for some rare Unicode
characters that produce more than 4 weights (long expansions).
UCA requires 2 bytes per weight multiplied by the number of levels.
In case of a 2-level collation, each character requires 4*2=8 bytes.
Therefore, the longest VARCHAR that fits into the default @@max_sort_length
is 1024/8=VARCHAR(128). With strnxfrm_multiply==8, only VARCHAR(64)
would fit.
Note, the built-in collation utf8_thai_520_w2 also uses strnxfrm_multiply=4,
for the same purpose.
TODO: we could add a new LDML syntax to choose strxfrm_multiply value.
*/
to->strxfrm_multiply= loaded->levels_for_order > 1 ?
4 : from->strxfrm_multiply;
to->min_sort_char= from->min_sort_char;
to->max_sort_char= from->max_sort_char;
to->mbminlen= from->mbminlen;
......@@ -312,7 +338,8 @@ static int add_collation(struct charset_info_st *cs)
#if defined(HAVE_CHARSET_ucs2) && defined(HAVE_UCA_COLLATIONS)
copy_uca_collation(newcs, newcs->state & MY_CS_NOPAD ?
&my_charset_ucs2_unicode_nopad_ci :
&my_charset_ucs2_unicode_ci);
&my_charset_ucs2_unicode_ci,
cs);
newcs->state|= MY_CS_AVAILABLE | MY_CS_LOADED | MY_CS_NONASCII;
#endif
}
......@@ -321,7 +348,8 @@ static int add_collation(struct charset_info_st *cs)
#if defined (HAVE_CHARSET_utf8) && defined(HAVE_UCA_COLLATIONS)
copy_uca_collation(newcs, newcs->state & MY_CS_NOPAD ?
&my_charset_utf8_unicode_nopad_ci :
&my_charset_utf8_unicode_ci);
&my_charset_utf8_unicode_ci,
cs);
newcs->ctype= my_charset_utf8_unicode_ci.ctype;
if (init_state_maps(newcs))
return MY_XML_ERROR;
......@@ -332,7 +360,8 @@ static int add_collation(struct charset_info_st *cs)
#if defined (HAVE_CHARSET_utf8mb4) && defined(HAVE_UCA_COLLATIONS)
copy_uca_collation(newcs, newcs->state & MY_CS_NOPAD ?
&my_charset_utf8mb4_unicode_nopad_ci :
&my_charset_utf8mb4_unicode_ci);
&my_charset_utf8mb4_unicode_ci,
cs);
newcs->ctype= my_charset_utf8mb4_unicode_ci.ctype;
newcs->state|= MY_CS_AVAILABLE | MY_CS_LOADED;
#endif
......@@ -342,7 +371,8 @@ static int add_collation(struct charset_info_st *cs)
#if defined (HAVE_CHARSET_utf16) && defined(HAVE_UCA_COLLATIONS)
copy_uca_collation(newcs, newcs->state & MY_CS_NOPAD ?
&my_charset_utf16_unicode_nopad_ci :
&my_charset_utf16_unicode_ci);
&my_charset_utf16_unicode_ci,
cs);
newcs->state|= MY_CS_AVAILABLE | MY_CS_LOADED | MY_CS_NONASCII;
#endif
}
......@@ -351,7 +381,8 @@ static int add_collation(struct charset_info_st *cs)
#if defined (HAVE_CHARSET_utf32) && defined(HAVE_UCA_COLLATIONS)
copy_uca_collation(newcs, newcs->state & MY_CS_NOPAD ?
&my_charset_utf32_unicode_nopad_ci :
&my_charset_utf32_unicode_ci);
&my_charset_utf32_unicode_ci,
cs);
newcs->state|= MY_CS_AVAILABLE | MY_CS_LOADED | MY_CS_NONASCII;
#endif
}
......
strings/ctype-uca.c
View file @ 0259b3cb
......@@ -6542,6 +6542,17 @@ MY_UCA_INFO my_uca_v400=
},
0 /* levelno */
},
{
0,
NULL,
NULL,
{
0,
NULL,
NULL
},
1 /* levelno */
},
},
/* Logical positions */
......@@ -30134,6 +30145,18 @@ MY_UCA_INFO my_uca_v520=
},
0 /* levelno */
},
{
0x10FFFF, /* maxchar */
(uchar *) uca520_length_w2,
(uint16 **) uca520_weight_w2,
{ /* Contractions: */
0, /* nitems */
NULL, /* item */
NULL /* flags */
},
1 /* levelno */
},
},
0x0009, /* first_non_ignorable p != ignore */
......@@ -31851,6 +31874,25 @@ static int my_strnncoll_uca_multilevel(CHARSET_INFO *cs,
}
static int
my_strnncollsp_generic_uca_nopad_multilevel(CHARSET_INFO *cs,
const uchar *s, size_t slen,
const uchar *t, size_t tlen)
{
uint num_level= cs->levels_for_order;
uint i;
for (i= 0; i != num_level; i++)
{
int ret= my_strnncoll_uca_onelevel(cs, &my_any_uca_scanner_handler,
&cs->uca->level[i],
s, slen, t, tlen, FALSE);
if (ret)
return ret;
}
return 0;
}
static inline int
my_space_weight(const MY_UCA_WEIGHT_LEVEL *level)
{
......@@ -32181,6 +32223,16 @@ my_strnxfrm_uca_onelevel(CHARSET_INFO *cs,
}
/*
Return the minimum possible weight on a level.
*/
static uint min_weight_on_level(MY_UCA_WEIGHT_LEVEL *level)
{
DBUG_ASSERT(level->levelno < 2); /* No 3-level NOPAD collations yet */
return level->levelno == 0 ? 0x0200 : 0x0020;
}
static uchar *
my_strnxfrm_uca_nopad_onelevel(CHARSET_INFO *cs,
my_uca_scanner_handler *scanner_handler,
......@@ -32194,12 +32246,9 @@ my_strnxfrm_uca_nopad_onelevel(CHARSET_INFO *cs,
dst, de, &nweights,
src, srclen);
DBUG_ASSERT(dst <= de);
/*
Pad with the minimum possible primary weight 0x0200.
*/
DBUG_ASSERT(level->levelno == 0); /* No multi-level NOPAD collations yet */
/* Pad with the minimum possible weight on this level */
if (dst < de && nweights && (flags & MY_STRXFRM_PAD_WITH_SPACE))
dst= my_strnxfrm_uca_padn(dst, de, nweights, 0x0200);
dst= my_strnxfrm_uca_padn(dst, de, nweights, min_weight_on_level(level));
DBUG_ASSERT(dst <= de);
my_strxfrm_desc_and_reverse(d0, dst, flags, 0);
return dst;
......@@ -32294,7 +32343,12 @@ my_strnxfrm_uca_multilevel(CHARSET_INFO *cs,
{
if (!(flags & MY_STRXFRM_LEVEL_ALL) ||
(flags & (MY_STRXFRM_LEVEL1 << current_level)))
dst= my_strnxfrm_uca_onelevel(cs, scanner_handler,
dst= cs->state & MY_CS_NOPAD ?
my_strnxfrm_uca_nopad_onelevel(cs, scanner_handler,
&cs->uca->level[current_level],
dst, de, nweights,
src, srclen, flags) :
my_strnxfrm_uca_onelevel(cs, scanner_handler,
&cs->uca->level[current_level],
dst, de, nweights,
src, srclen, flags);
......@@ -32970,6 +33024,7 @@ typedef enum
typedef struct my_coll_rules_st
{
uint version; /* Unicode version, e.g. 400 or 520 */
uint strength; /* Number of levels */
MY_UCA_INFO *uca; /* Unicode weight data */
size_t nrules; /* Number of rules in the rule array */
size_t mrules; /* Number of allocated rules */
......@@ -33251,6 +33306,10 @@ my_coll_parser_scan_setting(MY_COLL_RULE_PARSER *p)
{
rules->shift_after_method= my_shift_method_simple;
}
else if (!lex_cmp(lexem, C_STRING_WITH_LEN("[strength 1]")))
rules->strength= 1;
else if (!lex_cmp(lexem, C_STRING_WITH_LEN("[strength 2]")))
rules->strength= 2;
else
{
return 0;
......@@ -34189,6 +34248,10 @@ init_weight_level(MY_CHARSET_LOADER *loader, MY_COLL_RULES *rules,
}
MY_COLLATION_HANDLER my_collation_any_uca_handler_multilevel;
MY_COLLATION_HANDLER my_collation_generic_uca_nopad_handler_multilevel;
/*
This function copies an UCS2 collation from
the default Unicode Collation Algorithm (UCA)
......@@ -34213,6 +34276,7 @@ create_tailoring(struct charset_info_st *cs,
MY_COLL_RULES rules;
MY_UCA_INFO new_uca, *src_uca= NULL;
int rc= 0;
uint i;
*loader->error= '\0';
......@@ -34246,70 +34310,17 @@ create_tailoring(struct charset_info_st *cs,
if (!cs->caseinfo)
cs->caseinfo= &my_unicase_default;
}
cs->levels_for_order= rules.strength ? rules.strength : 1;
if ((rc= init_weight_level(loader, &rules,
&new_uca.level[0], &src_uca->level[0])))
goto ex;
if (!(cs->uca= (MY_UCA_INFO *) (loader->once_alloc)(sizeof(MY_UCA_INFO))))
{
rc= 1;
goto ex;
}
cs->uca[0]= new_uca;
ex:
(loader->free)(rules.rule);
if (rc != 0 && loader->error[0])
loader->reporter(ERROR_LEVEL, "%s", loader->error);
return rc;
}
static my_bool
create_tailoring_multilevel(struct charset_info_st *cs,
MY_CHARSET_LOADER *loader)
{
uint num_level= cs->levels_for_order;
MY_COLL_RULES rules;
MY_UCA_INFO new_uca, *src_uca= NULL;
int rc= 0;
uint i;
*loader->error= '\0';
if (!cs->tailoring)
return 0; /* Ok to add a collation without tailoring */
memset(&rules, 0, sizeof(rules));
rules.loader= loader;
rules.uca= cs->uca ? cs->uca : &my_uca_v400; /* For logical positions, etc */
memset(&new_uca, 0, sizeof(new_uca));
/* Parse ICU Collation Customization expression */
if ((rc= my_coll_rule_parse(&rules,
cs->tailoring,
cs->tailoring + strlen(cs->tailoring))))
goto ex;
if (rules.version == 520) /* Unicode-5.2.0 requested */
for (i= 0; i != cs->levels_for_order; i++)
{
src_uca= &my_uca_v520;
cs->caseinfo= &my_unicase_unicode520;
}
else if (rules.version == 400) /* Unicode-4.0.0 requested */
{
src_uca= &my_uca_v400;
cs->caseinfo= &my_unicase_default;
}
else /* No Unicode version specified */
if ((rc= (src_uca->level[i].maxchar == 0)))
{
src_uca= cs->uca ? cs->uca : &my_uca_v400;
if (!cs->caseinfo)
cs->caseinfo= &my_unicase_default;
my_snprintf(loader->error, sizeof(loader->error) - 1,
"%s: no level #%d data for this Unicode version.",
cs->name, i + 1);
goto ex;
}
for (i= 0; i != num_level; i++)
{
if ((rc= init_weight_level(loader, &rules,
&new_uca.level[i], &src_uca->level[i])))
goto ex;
......@@ -34321,6 +34332,10 @@ create_tailoring_multilevel(struct charset_info_st *cs,
goto ex;
}
cs->uca[0]= new_uca;
if (cs->levels_for_order > 1)
cs->coll= (cs->state & MY_CS_NOPAD) ?
&my_collation_generic_uca_nopad_handler_multilevel :
&my_collation_any_uca_handler_multilevel;
ex:
(loader->free)(rules.rule);
......@@ -34345,16 +34360,6 @@ my_coll_init_uca(struct charset_info_st *cs, MY_CHARSET_LOADER *loader)
return create_tailoring(cs, loader);
}
static my_bool
my_coll_init_uca_multilevel(struct charset_info_st *cs,
MY_CHARSET_LOADER *loader)
{
cs->pad_char= ' ';
cs->ctype= my_charset_utf8_unicode_ci.ctype;
if (!cs->caseinfo)
cs->caseinfo= &my_unicase_default;
return create_tailoring_multilevel(cs, loader);
}
static int my_strnncoll_any_uca(CHARSET_INFO *cs,
const uchar *s, size_t slen,
......@@ -34489,7 +34494,7 @@ MY_COLLATION_HANDLER my_collation_generic_uca_nopad_handler =
MY_COLLATION_HANDLER my_collation_any_uca_handler_multilevel=
{
my_coll_init_uca_multilevel,
my_coll_init_uca,
my_strnncoll_any_uca_multilevel,
my_strnncollsp_any_uca_multilevel,
my_strnxfrm_any_uca_multilevel,
......@@ -34503,6 +34508,22 @@ MY_COLLATION_HANDLER my_collation_any_uca_handler_multilevel=
};
MY_COLLATION_HANDLER my_collation_generic_uca_nopad_handler_multilevel =
{
my_coll_init_uca,
my_strnncoll_any_uca_multilevel,
my_strnncollsp_generic_uca_nopad_multilevel,
my_strnxfrm_any_uca_multilevel,
my_strnxfrmlen_any_uca_multilevel,
my_like_range_generic,
my_wildcmp_uca,
NULL,
my_instr_mb,
my_hash_sort_generic_uca_nopad,
my_propagate_complex
};
#ifdef HAVE_CHARSET_ucs2
/*
UCS2 optimized CHARSET_INFO compatible wrappers.
......@@ -35342,7 +35363,7 @@ struct charset_info_st my_charset_ucs2_thai_520_w2=
"ucs2", /* csname */
"ucs2_thai_520_w2", /* name */
"", /* comment */
"", /* tailoring */
"[strength 2]", /* tailoring */
NULL, /* ctype */
NULL, /* to_lower */
NULL, /* to_upper */
......@@ -36363,7 +36384,7 @@ struct charset_info_st my_charset_utf8_thai_520_w2=
MY_UTF8MB3, /* csname */
MY_UTF8MB3 "_thai_520_w2",/* name */
"", /* comment */
"", /* tailoring */
"[strength 2]", /* tailoring */
ctype_utf8, /* ctype */
NULL, /* to_lower */
NULL, /* to_upper */
......@@ -37275,7 +37296,7 @@ struct charset_info_st my_charset_utf8mb4_thai_520_w2=
MY_UTF8MB4, /* csname */
MY_UTF8MB4 "_thai_520_w2", /* name */
"", /* comment */
"", /* tailoring */
"[strength 2]", /* tailoring */
ctype_utf8, /* ctype */
NULL, /* to_lower */
NULL, /* to_upper */
......@@ -38237,7 +38258,7 @@ struct charset_info_st my_charset_utf32_thai_520_w2=
"utf32", /* csname */
"utf32_thai_520_w2",/* name */
"", /* comment */
"", /* tailoring */
"[strength 2]", /* tailoring */
NULL, /* ctype */
NULL, /* to_lower */
NULL, /* to_upper */
......@@ -39204,7 +39225,7 @@ struct charset_info_st my_charset_utf16_thai_520_w2=
"utf16", /* cs name */
"utf16_thai_520_w2",/* name */
"", /* comment */
"", /* tailoring */
"[strength 2]", /* tailoring */
NULL, /* ctype */
NULL, /* to_lower */
NULL, /* to_upper */
strings/ctype.c
View file @ 0259b3cb
......@@ -667,6 +667,8 @@ static int cs_value(MY_XML_PARSER *st,const char *attr, size_t len)
case _CS_ST_STRENGTH:
/* 1, 2, 3, 4, 5, or primary, secondary, tertiary, quaternary, identical */
rc= tailoring_append(st, "[strength %.*s]", len, attr);
if (len && attr[0] >= '1' && attr[0] <= '9')
i->cs.levels_for_order= attr[0] - '0';
break;
case _CS_ST_ALTERNATE:
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7