database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-06-11 20:28:21 +03:00
Code Activity

Fix up handling of C/POSIX collations.

Browse Source 
Install just one instance of the "C" and "POSIX" collations into
pg_collation, rather than one per encoding.  Make these instances exist
and do something useful even in machines without locale_t support: to wit,
it's now possible to force comparisons and case-folding functions to use C
locale in an otherwise non-C database, whether or not the platform has
support for using any additional collations.

Fix up severely broken upper/lower/initcap functions, too: the C/POSIX
fastpath now does what it is supposed to, and non-default collations are
handled correctly in single-byte database encodings.

Merge the two separate collation hashtables that were being maintained in
pg_locale.c, and be more wary of the possibility that we fail partway
through filling a cache entry.
This commit is contained in:
Tom Lane
2011-03-20 12:43:39 -04:00
parent c2f4ea469b
commit 176d5bae1d
8 changed files with 586 additions and 296 deletions
Download Patch File Download Diff File Expand all files Collapse all files

140
doc/src/sgml/charset.sgml
View File

@ -68,7 +68,7 @@ initdb --locale=sv_SE
<para> <para>
This example for Unix systems sets the locale to Swedish This example for Unix systems sets the locale to Swedish
(<literal>sv</>) as spoken (<literal>sv</>) as spoken
in Sweden (<literal>SE</>). Other possibilities might be in Sweden (<literal>SE</>). Other possibilities might include
<literal>en_US</> (U.S. English) and <literal>fr_CA</> (French <literal>en_US</> (U.S. English) and <literal>fr_CA</> (French
Canadian). If more than one character set can be used for a Canadian). If more than one character set can be used for a
locale then the specifications can take the form locale then the specifications can take the form
@ -133,7 +133,8 @@ initdb --locale=sv_SE
<para> <para>
If you want the system to behave as if it had no locale support, If you want the system to behave as if it had no locale support,
use the special locale <literal>C</> or <literal>POSIX</>. use the special locale name <literal>C</>, or equivalently
<literal>POSIX</>.
</para> </para>
<para> <para>
@ -257,7 +258,9 @@ initdb --locale=sv_SE
operator classes exist. These allow the creation of an index that operator classes exist. These allow the creation of an index that
performs a strict character-by-character comparison, ignoring performs a strict character-by-character comparison, ignoring
locale comparison rules. Refer to <xref linkend="indexes-opclass"> locale comparison rules. Refer to <xref linkend="indexes-opclass">
for more information. for more information. Another approach is to create indexes using
the <literal>C</> collation, as discussed in
<xref linkend="collation">.
</para> </para>
</sect2> </sect2>
@ -321,13 +324,6 @@ initdb --locale=sv_SE
of a database cannot be changed after its creation. of a database cannot be changed after its creation.
</para> </para>
<note>
<para>
Collation support is currently only known to work on
Linux (glibc) and Mac OS X platforms.
</para>
</note>
<sect2> <sect2>
<title>Concepts</title> <title>Concepts</title>
@ -335,7 +331,8 @@ initdb --locale=sv_SE
Conceptually, every expression of a collatable data type has a Conceptually, every expression of a collatable data type has a
collation. (The built-in collatable data types are collation. (The built-in collatable data types are
<type>text</type>, <type>varchar</type>, and <type>char</type>. <type>text</type>, <type>varchar</type>, and <type>char</type>.
User-defined base types can also be marked collatable.) If the User-defined base types can also be marked collatable, and of course
a domain over a collatable data type is collatable.) If the
expression is a column reference, the collation of the expression is the expression is a column reference, the collation of the expression is the
defined collation of the column. If the expression is a constant, the defined collation of the column. If the expression is a constant, the
collation is the default collation of the data type of the collation is the default collation of the data type of the
@ -346,8 +343,8 @@ initdb --locale=sv_SE
<para> <para>
The collation of an expression can be the <quote>default</quote> The collation of an expression can be the <quote>default</quote>
collation, which means the locale settings defined for the collation, which means the locale settings defined for the
database. In some cases, an expression can also have no known database. It is also possible for an expression's collation to be
collation. In such cases, ordering operations and other indeterminate. In such cases, ordering operations and other
operations that need to know the collation will fail. operations that need to know the collation will fail.
</para> </para>
@ -379,7 +376,7 @@ initdb --locale=sv_SE
The <firstterm>collation derivation</firstterm> of an expression can be The <firstterm>collation derivation</firstterm> of an expression can be
implicit or explicit. This distinction affects how collations are implicit or explicit. This distinction affects how collations are
combined when multiple different collations appear in an combined when multiple different collations appear in an
expression. An explicit collation derivation arises when a expression. An explicit collation derivation occurs when a
<literal>COLLATE</literal> clause is used; all other collation <literal>COLLATE</literal> clause is used; all other collation
derivations are implicit. When multiple collations need to be derivations are implicit. When multiple collations need to be
combined, for example in a function call, the following rules are combined, for example in a function call, the following rules are
@ -399,34 +396,90 @@ initdb --locale=sv_SE
<listitem> <listitem>
<para> <para>
Otherwise, all input expressions must have the same implicit Otherwise, all input expressions must have the same implicit
collation derivation or the default collation. If any collation derivation or the default collation. If any non-default
implicitly derived collation is present, that is the result of collation is present, that is the result of the collation combination.
the collation combination. Otherwise, the result is the Otherwise, the result is the default collation.
default collation. </para>
</listitem>
<listitem>
<para>
If there are conflicting non-default implicit collations among the
input expressions, then the combination is deemed to have indeterminate
collation. This is not an error condition unless the particular
function being invoked requires knowledge of the collation it should
apply. If it does, an error will be raised at run-time.
</para> </para>
</listitem> </listitem>
</orderedlist> </orderedlist>
For example, take this table definition: For example, consider this table definition:
<programlisting> <programlisting>
CREATE TABLE test1 ( CREATE TABLE test1 (
a text COLLATE "x", a text COLLATE "de_DE",
b text COLLATE "es_ES",
... ...
); );
</programlisting> </programlisting>
Then in Then in
<programlisting> <programlisting>
SELECT a || 'foo' FROM test1; SELECT a &lt; 'foo' FROM test1;
</programlisting> </programlisting>
the result collation of the <literal>||</literal> operator is the <literal>&lt;</literal> comparison is performed according to
<literal>"x"</literal> because it combines an implicitly derived <literal>de_DE</literal> rules, because the expression combines an
collation with the default collation. But in implicitly derived collation with the default collation. But in
<programlisting> <programlisting>
SELECT a || ('foo' COLLATE "y") FROM test1; SELECT a &lt; ('foo' COLLATE "fr_FR") FROM test1;
</programlisting>
the comparison is performed using <literal>fr_FR</literal> rules,
because the explicit collation derivation overrides the implicit one.
Furthermore, given
<programlisting>
SELECT a &lt; b FROM test1;
</programlisting>
the parser cannot determine which collation to apply, since the
<structfield>a</> and <structfield>b</> columns have conflicting
implicit collations. Since the <literal>&lt;</literal> operator
does need to know which collation to use, this will result in an
error. The error can be resolved by attaching an explicit collation
specifier to either input expression, thus:
<programlisting>
SELECT a &lt; b COLLATE "de_DE" FROM test1;
</programlisting>
or equivalently
<programlisting>
SELECT a COLLATE "de_DE" &lt; b FROM test1;
</programlisting>
On the other hand, the structurally similar case
<programlisting>
SELECT a || b FROM test1;
</programlisting>
does not result in an error, because the <literal>||</> operator
does not care about collations: its result is the same regardless
of the collation.
</para>
<para>
The collation assigned to a function or operator's combined input
expressions is also considered to apply to the function or operator's
result, if the function or operator delivers a result of a collatable
data type. So, in
<programlisting>
SELECT * FROM test1 ORDER BY a || 'foo';
</programlisting>
the ordering will be done according to <literal>de_DE</literal> rules.
But this query:
<programlisting>
SELECT * FROM test1 ORDER BY a || b;
</programlisting>
results in an error, because even though the <literal>||</> operator
doesn't need to know a collation, the <literal>ORDER BY</> clause does.
As before, the conflict can be resolved with an explicit collation
specifier:
<programlisting>
SELECT * FROM test1 ORDER BY a || b COLLATE "fr_FR";
</programlisting> </programlisting>
the result collation is <literal>"y"</literal> because the explicit
collation derivation overrides the implicit one.
</para> </para>
</sect2> </sect2>
@ -449,7 +502,22 @@ SELECT a || ('foo' COLLATE "y") FROM test1;
</para> </para>
<para> <para>
When a database cluster is initialized, <command>initdb</command> On all platforms, the collations named <literal>default</>,
<literal>C</>, and <literal>POSIX</> are available. Additional
collations may be available depending on operating system support.
The <literal>default</> collation selects the <symbol>LC_COLLATE</symbol>
and <symbol>LC_CTYPE</symbol> values specified at database creation time.
The <literal>C</> and <literal>POSIX</> collations both specify
<quote>traditional C</> behavior, in which only the ASCII letters
<quote><literal>A</></quote> through <quote><literal>Z</></quote>
are treated as letters, and sorting is done strictly by character
code byte values.
</para>
<para>
If the operating system provides support for using multiple locales
within a single program (<function>newlocale</> and related functions),
then when a database cluster is initialized, <command>initdb</command>
populates the system catalog <literal>pg_collation</literal> with populates the system catalog <literal>pg_collation</literal> with
collations based on all the locales it finds on the operating collations based on all the locales it finds on the operating
system at the time. For example, the operating system might system at the time. For example, the operating system might
@ -484,7 +552,21 @@ SELECT a || ('foo' COLLATE "y") FROM test1;
within a given database even though it would not be unique globally. within a given database even though it would not be unique globally.
Use of the stripped collation names is recommendable, since it will Use of the stripped collation names is recommendable, since it will
make one less thing you need to change if you decide to change to make one less thing you need to change if you decide to change to
another database encoding. another database encoding. Note however that the <literal>default</>,
<literal>C</>, and <literal>POSIX</> collations can be used
regardless of the database encoding.
</para>
<para>
<productname>PostgreSQL</productname> considers distinct collation
objects to be incompatible even when they have identical properties.
Thus for example,
<programlisting>
SELECT a COLLATE "C" &lt; b COLLATE "POSIX" FROM test1;
</programlisting>
will draw an error even though the <literal>C</> and <literal>POSIX</>
collations have identical behaviors. Mixing stripped and non-stripped
collation names is therefore not recommended.
</para> </para>
</sect2> </sect2>
</sect1> </sect1>

159
src/backend/utils/adt/formatting.c
View File

@ -1462,10 +1462,16 @@ str_numth(char *dest, char *num, int type)
* in multibyte character sets. Note that in either case we are effectively * in multibyte character sets. Note that in either case we are effectively
* assuming that the database character encoding matches the encoding implied * assuming that the database character encoding matches the encoding implied
* by LC_CTYPE. * by LC_CTYPE.
*
* If the system provides locale_t and associated functions (which are
* standardized by Open Group's XBD), we can support collations that are
* neither default nor C. The code is written to handle both combinations
* of have-wide-characters and have-locale_t, though it's rather unlikely
* a platform would have the latter without the former.
*/ */
/* /*
* wide-character-aware lower function * collation-aware, wide-character-aware lower function
* *
* We pass the number of bytes so we can pass varlena and char* * We pass the number of bytes so we can pass varlena and char*
* to this function. The result is a palloc'd, null-terminated string. * to this function. The result is a palloc'd, null-terminated string.
@ -1474,21 +1480,31 @@ char *
str_tolower(const char *buff, size_t nbytes, Oid collid) str_tolower(const char *buff, size_t nbytes, Oid collid)
{ {
char *result; char *result;
pg_locale_t mylocale = 0;
if (!buff) if (!buff)
return NULL; return NULL;
if (collid != DEFAULT_COLLATION_OID) /* C/POSIX collations use this path regardless of database encoding */
mylocale = pg_newlocale_from_collation(collid); if (lc_ctype_is_c(collid))
#ifdef USE_WIDE_UPPER_LOWER
if (pg_database_encoding_max_length() > 1 && !lc_ctype_is_c(collid))
{ {
char *p;
result = pnstrdup(buff, nbytes);
for (p = result; *p; p++)
*p = pg_ascii_tolower((unsigned char) *p);
}
#ifdef USE_WIDE_UPPER_LOWER
else if (pg_database_encoding_max_length() > 1)
{
pg_locale_t mylocale = 0;
wchar_t *workspace; wchar_t *workspace;
size_t curr_char; size_t curr_char;
size_t result_size; size_t result_size;
if (collid != DEFAULT_COLLATION_OID)
mylocale = pg_newlocale_from_collation(collid);
/* Overflow paranoia */ /* Overflow paranoia */
if ((nbytes + 1) > (INT_MAX / sizeof(wchar_t))) if ((nbytes + 1) > (INT_MAX / sizeof(wchar_t)))
ereport(ERROR, ereport(ERROR,
@ -1501,12 +1517,14 @@ str_tolower(const char *buff, size_t nbytes, Oid collid)
char2wchar(workspace, nbytes + 1, buff, nbytes, collid); char2wchar(workspace, nbytes + 1, buff, nbytes, collid);
for (curr_char = 0; workspace[curr_char] != 0; curr_char++) for (curr_char = 0; workspace[curr_char] != 0; curr_char++)
{
#ifdef HAVE_LOCALE_T #ifdef HAVE_LOCALE_T
if (mylocale) if (mylocale)
workspace[curr_char] = towlower_l(workspace[curr_char], mylocale); workspace[curr_char] = towlower_l(workspace[curr_char], mylocale);
else else
#endif #endif
workspace[curr_char] = towlower(workspace[curr_char]); workspace[curr_char] = towlower(workspace[curr_char]);
}
/* Make result large enough; case change might change number of bytes */ /* Make result large enough; case change might change number of bytes */
result_size = curr_char * pg_database_encoding_max_length() + 1; result_size = curr_char * pg_database_encoding_max_length() + 1;
@ -1515,22 +1533,40 @@ str_tolower(const char *buff, size_t nbytes, Oid collid)
wchar2char(result, workspace, result_size, collid); wchar2char(result, workspace, result_size, collid);
pfree(workspace); pfree(workspace);
} }
else
#endif /* USE_WIDE_UPPER_LOWER */ #endif /* USE_WIDE_UPPER_LOWER */
else
{ {
pg_locale_t mylocale = 0;
char *p; char *p;
if (collid != DEFAULT_COLLATION_OID)
mylocale = pg_newlocale_from_collation(collid);
result = pnstrdup(buff, nbytes); result = pnstrdup(buff, nbytes);
/*
* Note: we assume that tolower_l() will not be so broken as to need
* an isupper_l() guard test. When using the default collation, we
* apply the traditional Postgres behavior that forces ASCII-style
* treatment of I/i, but in non-default collations you get exactly
* what the collation says.
*/
for (p = result; *p; p++) for (p = result; *p; p++)
{
#ifdef HAVE_LOCALE_T
if (mylocale)
*p = tolower_l((unsigned char) *p, mylocale);
else
#endif
*p = pg_tolower((unsigned char) *p); *p = pg_tolower((unsigned char) *p);
} }
}
return result; return result;
} }
/* /*
* wide-character-aware upper function * collation-aware, wide-character-aware upper function
* *
* We pass the number of bytes so we can pass varlena and char* * We pass the number of bytes so we can pass varlena and char*
* to this function. The result is a palloc'd, null-terminated string. * to this function. The result is a palloc'd, null-terminated string.
@ -1539,21 +1575,31 @@ char *
str_toupper(const char *buff, size_t nbytes, Oid collid) str_toupper(const char *buff, size_t nbytes, Oid collid)
{ {
char *result; char *result;
pg_locale_t mylocale = 0;
if (!buff) if (!buff)
return NULL; return NULL;
if (collid != DEFAULT_COLLATION_OID) /* C/POSIX collations use this path regardless of database encoding */
mylocale = pg_newlocale_from_collation(collid); if (lc_ctype_is_c(collid))
#ifdef USE_WIDE_UPPER_LOWER
if (pg_database_encoding_max_length() > 1 && !lc_ctype_is_c(collid))
{ {
char *p;
result = pnstrdup(buff, nbytes);
for (p = result; *p; p++)
*p = pg_ascii_toupper((unsigned char) *p);
}
#ifdef USE_WIDE_UPPER_LOWER
else if (pg_database_encoding_max_length() > 1)
{
pg_locale_t mylocale = 0;
wchar_t *workspace; wchar_t *workspace;
size_t curr_char; size_t curr_char;
size_t result_size; size_t result_size;
if (collid != DEFAULT_COLLATION_OID)
mylocale = pg_newlocale_from_collation(collid);
/* Overflow paranoia */ /* Overflow paranoia */
if ((nbytes + 1) > (INT_MAX / sizeof(wchar_t))) if ((nbytes + 1) > (INT_MAX / sizeof(wchar_t)))
ereport(ERROR, ereport(ERROR,
@ -1566,12 +1612,14 @@ str_toupper(const char *buff, size_t nbytes, Oid collid)
char2wchar(workspace, nbytes + 1, buff, nbytes, collid); char2wchar(workspace, nbytes + 1, buff, nbytes, collid);
for (curr_char = 0; workspace[curr_char] != 0; curr_char++) for (curr_char = 0; workspace[curr_char] != 0; curr_char++)
{
#ifdef HAVE_LOCALE_T #ifdef HAVE_LOCALE_T
if (mylocale) if (mylocale)
workspace[curr_char] = towupper_l(workspace[curr_char], mylocale); workspace[curr_char] = towupper_l(workspace[curr_char], mylocale);
else else
#endif #endif
workspace[curr_char] = towupper(workspace[curr_char]); workspace[curr_char] = towupper(workspace[curr_char]);
}
/* Make result large enough; case change might change number of bytes */ /* Make result large enough; case change might change number of bytes */
result_size = curr_char * pg_database_encoding_max_length() + 1; result_size = curr_char * pg_database_encoding_max_length() + 1;
@ -1580,22 +1628,40 @@ str_toupper(const char *buff, size_t nbytes, Oid collid)
wchar2char(result, workspace, result_size, collid); wchar2char(result, workspace, result_size, collid);
pfree(workspace); pfree(workspace);
} }
else
#endif /* USE_WIDE_UPPER_LOWER */ #endif /* USE_WIDE_UPPER_LOWER */
else
{ {
pg_locale_t mylocale = 0;
char *p; char *p;
if (collid != DEFAULT_COLLATION_OID)
mylocale = pg_newlocale_from_collation(collid);
result = pnstrdup(buff, nbytes); result = pnstrdup(buff, nbytes);
/*
* Note: we assume that toupper_l() will not be so broken as to need
* an islower_l() guard test. When using the default collation, we
* apply the traditional Postgres behavior that forces ASCII-style
* treatment of I/i, but in non-default collations you get exactly
* what the collation says.
*/
for (p = result; *p; p++) for (p = result; *p; p++)
{
#ifdef HAVE_LOCALE_T
if (mylocale)
*p = toupper_l((unsigned char) *p, mylocale);
else
#endif
*p = pg_toupper((unsigned char) *p); *p = pg_toupper((unsigned char) *p);
} }
}
return result; return result;
} }
/* /*
* wide-character-aware initcap function * collation-aware, wide-character-aware initcap function
* *
* We pass the number of bytes so we can pass varlena and char* * We pass the number of bytes so we can pass varlena and char*
* to this function. The result is a palloc'd, null-terminated string. * to this function. The result is a palloc'd, null-terminated string.
@ -1605,21 +1671,42 @@ str_initcap(const char *buff, size_t nbytes, Oid collid)
{ {
char *result; char *result;
int wasalnum = false; int wasalnum = false;
pg_locale_t mylocale = 0;
if (!buff) if (!buff)
return NULL; return NULL;
if (collid != DEFAULT_COLLATION_OID) /* C/POSIX collations use this path regardless of database encoding */
mylocale = pg_newlocale_from_collation(collid); if (lc_ctype_is_c(collid))
#ifdef USE_WIDE_UPPER_LOWER
if (pg_database_encoding_max_length() > 1 && !lc_ctype_is_c(collid))
{ {
char *p;
result = pnstrdup(buff, nbytes);
for (p = result; *p; p++)
{
char c;
if (wasalnum)
*p = c = pg_ascii_tolower((unsigned char) *p);
else
*p = c = pg_ascii_toupper((unsigned char) *p);
/* we don't trust isalnum() here */
wasalnum = ((c >= 'A' && c <= 'Z') ||
(c >= 'a' && c <= 'z') ||
(c >= '0' && c <= '9'));
}
}
#ifdef USE_WIDE_UPPER_LOWER
else if (pg_database_encoding_max_length() > 1)
{
pg_locale_t mylocale = 0;
wchar_t *workspace; wchar_t *workspace;
size_t curr_char; size_t curr_char;
size_t result_size; size_t result_size;
if (collid != DEFAULT_COLLATION_OID)
mylocale = pg_newlocale_from_collation(collid);
/* Overflow paranoia */ /* Overflow paranoia */
if ((nbytes + 1) > (INT_MAX / sizeof(wchar_t))) if ((nbytes + 1) > (INT_MAX / sizeof(wchar_t)))
ereport(ERROR, ereport(ERROR,
@ -1660,15 +1747,38 @@ str_initcap(const char *buff, size_t nbytes, Oid collid)
wchar2char(result, workspace, result_size, collid); wchar2char(result, workspace, result_size, collid);
pfree(workspace); pfree(workspace);
} }
else
#endif /* USE_WIDE_UPPER_LOWER */ #endif /* USE_WIDE_UPPER_LOWER */
else
{ {
pg_locale_t mylocale = 0;
char *p; char *p;
if (collid != DEFAULT_COLLATION_OID)
mylocale = pg_newlocale_from_collation(collid);
result = pnstrdup(buff, nbytes); result = pnstrdup(buff, nbytes);
/*
* Note: we assume that toupper_l()/tolower_l() will not be so broken
* as to need guard tests. When using the default collation, we apply
* the traditional Postgres behavior that forces ASCII-style treatment
* of I/i, but in non-default collations you get exactly what the
* collation says.
*/
for (p = result; *p; p++) for (p = result; *p; p++)
{ {
#ifdef HAVE_LOCALE_T
if (mylocale)
{
if (wasalnum)
*p = tolower_l((unsigned char) *p, mylocale);
else
*p = toupper_l((unsigned char) *p, mylocale);
wasalnum = isalnum_l((unsigned char) *p, mylocale);
}
else
#endif
{
if (wasalnum) if (wasalnum)
*p = pg_tolower((unsigned char) *p); *p = pg_tolower((unsigned char) *p);
else else
@ -1676,6 +1786,7 @@ str_initcap(const char *buff, size_t nbytes, Oid collid)
wasalnum = isalnum((unsigned char) *p); wasalnum = isalnum((unsigned char) *p);
} }
} }
}
return result; return result;
} }

436
src/backend/utils/adt/pg_locale.c
View File

@ -99,15 +99,24 @@ static char lc_monetary_envbuf[LC_ENV_BUFSIZE];
static char lc_numeric_envbuf[LC_ENV_BUFSIZE]; static char lc_numeric_envbuf[LC_ENV_BUFSIZE];
static char lc_time_envbuf[LC_ENV_BUFSIZE]; static char lc_time_envbuf[LC_ENV_BUFSIZE];
/* Cache for collation-related knowledge */
typedef struct
{
Oid collid; /* hash key: pg_collation OID */
bool collate_is_c; /* is collation's LC_COLLATE C? */
bool ctype_is_c; /* is collation's LC_CTYPE C? */
bool flags_valid; /* true if above flags are valid */
pg_locale_t locale; /* locale_t struct, or 0 if not valid */
} collation_cache_entry;
static HTAB *collation_cache = NULL;
#if defined(WIN32) && defined(LC_MESSAGES) #if defined(WIN32) && defined(LC_MESSAGES)
static char *IsoLocaleName(const char *); /* MSVC specific */ static char *IsoLocaleName(const char *); /* MSVC specific */
#endif #endif
static HTAB *locale_cness_cache = NULL;
#ifdef HAVE_LOCALE_T
static HTAB *locale_t_cache = NULL;
#endif
/* /*
* pg_perm_setlocale * pg_perm_setlocale
@ -312,136 +321,6 @@ locale_messages_assign(const char *value, bool doit, GucSource source)
} }
/*
* We'd like to cache whether LC_COLLATE or LC_CTYPE is C (or POSIX),
* so we can optimize a few code paths in various places.
*
* Note that some code relies on this not reporting false negatives
* (that is, saying it's not C when it is). For example, char2wchar()
* could fail if the locale is C, so str_tolower() shouldn't call it
* in that case.
*/
struct locale_cness_cache_entry
{
Oid collid;
bool collate_is_c;
bool ctype_is_c;
};
static void
init_locale_cness_cache(void)
{
HASHCTL ctl;
memset(&ctl, 0, sizeof(ctl));
ctl.keysize = sizeof(Oid);
ctl.entrysize = sizeof(struct locale_cness_cache_entry);
ctl.hash = oid_hash;
locale_cness_cache = hash_create("locale C-ness cache", 1000, &ctl, HASH_ELEM | HASH_FUNCTION);
}
/*
* Handle caching of locale "C-ness" for nondefault collation objects.
* Relying on the system cache directly isn't fast enough.
*/
static bool
lookup_collation_cness(Oid collation, int category)
{
struct locale_cness_cache_entry *cache_entry;
bool found;
HeapTuple tp;
char *localeptr;
Assert(OidIsValid(collation));
Assert(category == LC_COLLATE || category == LC_CTYPE);
if (!locale_cness_cache)
init_locale_cness_cache();
cache_entry = hash_search(locale_cness_cache, &collation, HASH_ENTER, &found);
if (found)
{
if (category == LC_COLLATE)
return cache_entry->collate_is_c;
else
return cache_entry->ctype_is_c;
}
tp = SearchSysCache1(COLLOID, ObjectIdGetDatum(collation));
if (!HeapTupleIsValid(tp))
elog(ERROR, "cache lookup failed for collation %u", collation);
localeptr = NameStr(((Form_pg_collation) GETSTRUCT(tp))->collcollate);
cache_entry->collate_is_c = (strcmp(localeptr, "C") == 0) || (strcmp(localeptr, "POSIX") == 0);
localeptr = NameStr(((Form_pg_collation) GETSTRUCT(tp))->collctype);
cache_entry->ctype_is_c = (strcmp(localeptr, "C") == 0) || (strcmp(localeptr, "POSIX") == 0);
ReleaseSysCache(tp);
return category == LC_COLLATE ? cache_entry->collate_is_c : cache_entry->ctype_is_c;
}
bool
lc_collate_is_c(Oid collation)
{
/* Cache result so we only have to compute it once */
static int result = -1;
char *localeptr;
if (!OidIsValid(collation))
return false;
if (collation != DEFAULT_COLLATION_OID)
return lookup_collation_cness(collation, LC_COLLATE);
if (result >= 0)
return (bool) result;
localeptr = setlocale(LC_COLLATE, NULL);
if (!localeptr)
elog(ERROR, "invalid LC_COLLATE setting");
if (strcmp(localeptr, "C") == 0)
result = true;
else if (strcmp(localeptr, "POSIX") == 0)
result = true;
else
result = false;
return (bool) result;
}
bool
lc_ctype_is_c(Oid collation)
{
/* Cache result so we only have to compute it once */
static int result = -1;
char *localeptr;
if (!OidIsValid(collation))
return false;
if (collation != DEFAULT_COLLATION_OID)
return lookup_collation_cness(collation, LC_CTYPE);
if (result >= 0)
return (bool) result;
localeptr = setlocale(LC_CTYPE, NULL);
if (!localeptr)
elog(ERROR, "invalid LC_CTYPE setting");
if (strcmp(localeptr, "C") == 0)
result = true;
else if (strcmp(localeptr, "POSIX") == 0)
result = true;
else
result = false;
return (bool) result;
}
/* /*
* Frees the malloced content of a struct lconv. (But not the struct * Frees the malloced content of a struct lconv. (But not the struct
* itself.) * itself.)
@ -844,116 +723,295 @@ IsoLocaleName(const char *winlocname)
#endif /* WIN32 && LC_MESSAGES */ #endif /* WIN32 && LC_MESSAGES */
#ifdef HAVE_LOCALE_T /*
struct locale_t_cache_entry * Cache mechanism for collation information.
{ *
Oid collid; * We cache two flags: whether the collation's LC_COLLATE or LC_CTYPE is C
locale_t locale; * (or POSIX), so we can optimize a few code paths in various places.
}; * For the built-in C and POSIX collations, we can know that without even
* doing a cache lookup, but we want to support aliases for C/POSIX too.
* For the "default" collation, there are separate static cache variables,
* since consulting the pg_collation catalog doesn't tell us what we need.
*
* Also, if a pg_locale_t has been requested for a collation, we cache that
* for the life of a backend.
*
* Note that some code relies on the flags not reporting false negatives
* (that is, saying it's not C when it is). For example, char2wchar()
* could fail if the locale is C, so str_tolower() shouldn't call it
* in that case.
*
* Note that we currently lack any way to flush the cache. Since we don't
* support ALTER COLLATION, this is OK. The worst case is that someone
* drops a collation, and a useless cache entry hangs around in existing
* backends.
*/
static void static collation_cache_entry *
init_locale_t_cache(void) lookup_collation_cache(Oid collation, bool set_flags)
{ {
collation_cache_entry *cache_entry;
bool found;
Assert(OidIsValid(collation));
Assert(collation != DEFAULT_COLLATION_OID);
if (collation_cache == NULL)
{
/* First time through, initialize the hash table */
HASHCTL ctl; HASHCTL ctl;
memset(&ctl, 0, sizeof(ctl)); memset(&ctl, 0, sizeof(ctl));
ctl.keysize = sizeof(Oid); ctl.keysize = sizeof(Oid);
ctl.entrysize = sizeof(struct locale_t_cache_entry); ctl.entrysize = sizeof(collation_cache_entry);
ctl.hash = oid_hash; ctl.hash = oid_hash;
locale_t_cache = hash_create("locale_t cache", 1000, &ctl, HASH_ELEM | HASH_FUNCTION); collation_cache = hash_create("Collation cache", 100, &ctl,
HASH_ELEM | HASH_FUNCTION);
} }
#endif /* HAVE_LOCALE_T */
cache_entry = hash_search(collation_cache, &collation, HASH_ENTER, &found);
if (!found)
{
/*
* Make sure cache entry is marked invalid, in case we fail before
* setting things.
*/
cache_entry->flags_valid = false;
cache_entry->locale = 0;
}
if (set_flags && !cache_entry->flags_valid)
{
/* Attempt to set the flags */
HeapTuple tp;
Form_pg_collation collform;
const char *collcollate;
const char *collctype;
tp = SearchSysCache1(COLLOID, ObjectIdGetDatum(collation));
if (!HeapTupleIsValid(tp))
elog(ERROR, "cache lookup failed for collation %u", collation);
collform = (Form_pg_collation) GETSTRUCT(tp);
collcollate = NameStr(collform->collcollate);
collctype = NameStr(collform->collctype);
cache_entry->collate_is_c = ((strcmp(collcollate, "C") == 0) ||
(strcmp(collcollate, "POSIX") == 0));
cache_entry->ctype_is_c = ((strcmp(collctype, "C") == 0) ||
(strcmp(collctype, "POSIX") == 0));
cache_entry->flags_valid = true;
ReleaseSysCache(tp);
}
return cache_entry;
}
/*
* Detect whether collation's LC_COLLATE property is C
*/
bool
lc_collate_is_c(Oid collation)
{
/*
* If we're asked about "collation 0", return false, so that the code
* will go into the non-C path and report that the collation is bogus.
*/
if (!OidIsValid(collation))
return false;
/*
* If we're asked about the default collation, we have to inquire of
* the C library. Cache the result so we only have to compute it once.
*/
if (collation == DEFAULT_COLLATION_OID)
{
static int result = -1;
char *localeptr;
if (result >= 0)
return (bool) result;
localeptr = setlocale(LC_COLLATE, NULL);
if (!localeptr)
elog(ERROR, "invalid LC_COLLATE setting");
if (strcmp(localeptr, "C") == 0)
result = true;
else if (strcmp(localeptr, "POSIX") == 0)
result = true;
else
result = false;
return (bool) result;
}
/*
* If we're asked about the built-in C/POSIX collations, we know that.
*/
if (collation == C_COLLATION_OID ||
collation == POSIX_COLLATION_OID)
return true;
/*
* Otherwise, we have to consult pg_collation, but we cache that.
*/
return (lookup_collation_cache(collation, true))->collate_is_c;
}
/*
* Detect whether collation's LC_CTYPE property is C
*/
bool
lc_ctype_is_c(Oid collation)
{
/*
* If we're asked about "collation 0", return false, so that the code
* will go into the non-C path and report that the collation is bogus.
*/
if (!OidIsValid(collation))
return false;
/*
* If we're asked about the default collation, we have to inquire of
* the C library. Cache the result so we only have to compute it once.
*/
if (collation == DEFAULT_COLLATION_OID)
{
static int result = -1;
char *localeptr;
if (result >= 0)
return (bool) result;
localeptr = setlocale(LC_CTYPE, NULL);
if (!localeptr)
elog(ERROR, "invalid LC_CTYPE setting");
if (strcmp(localeptr, "C") == 0)
result = true;
else if (strcmp(localeptr, "POSIX") == 0)
result = true;
else
result = false;
return (bool) result;
}
/*
* If we're asked about the built-in C/POSIX collations, we know that.
*/
if (collation == C_COLLATION_OID ||
collation == POSIX_COLLATION_OID)
return true;
/*
* Otherwise, we have to consult pg_collation, but we cache that.
*/
return (lookup_collation_cache(collation, true))->ctype_is_c;
}
/* /*
* Create a locale_t from a collation OID. Results are cached for the * Create a locale_t from a collation OID. Results are cached for the
* lifetime of the backend. Thus, do not free the result with * lifetime of the backend. Thus, do not free the result with freelocale().
* freelocale().
* *
* As a special optimization, the default/database collation returns * As a special optimization, the default/database collation returns 0.
* 0. Callers should then revert to the non-locale_t-enabled code * Callers should then revert to the non-locale_t-enabled code path.
* path. In fact, they shouldn't call this function at all when they * In fact, they shouldn't call this function at all when they are dealing
* are dealing with the default locale. That can save quite a bit in * with the default locale. That can save quite a bit in hotspots.
* hotspots. * Also, callers should avoid calling this before going down a C/POSIX
* fastpath, because such a fastpath should work even on platforms without
* locale_t support in the C library.
* *
* For simplicity, we always generate COLLATE + CTYPE even though we * For simplicity, we always generate COLLATE + CTYPE even though we
* might only need one of them. Since this is called only once per * might only need one of them. Since this is called only once per session,
* session, it shouldn't cost much. * it shouldn't cost much.
*/ */
pg_locale_t pg_locale_t
pg_newlocale_from_collation(Oid collid) pg_newlocale_from_collation(Oid collid)
{ {
collation_cache_entry *cache_entry;
/* Return 0 for "default" collation, just in case caller forgets */
if (collid == DEFAULT_COLLATION_OID)
return (pg_locale_t) 0;
/*
* This is where we'll fail if a collation-aware function is invoked
* and no collation OID is passed. This typically means that the
* parser could not resolve a conflict of implicit collations, so
* report it that way.
*/
if (!OidIsValid(collid))
ereport(ERROR,
(errcode(ERRCODE_INDETERMINATE_COLLATION),
errmsg("locale operation to be invoked, but no collation was derived")));
cache_entry = lookup_collation_cache(collid, false);
if (cache_entry->locale == 0)
{
/* We haven't computed this yet in this session, so do it */
#ifdef HAVE_LOCALE_T #ifdef HAVE_LOCALE_T
HeapTuple tp; HeapTuple tp;
Form_pg_collation collform;
const char *collcollate; const char *collcollate;
const char *collctype; const char *collctype;
locale_t result; locale_t result;
struct locale_t_cache_entry *cache_entry;
bool found;
if (collid == DEFAULT_COLLATION_OID)
return (locale_t) 0;
if (!OidIsValid(collid))
elog(ERROR, "locale operation to be invoked, but no collation was derived");
if (!locale_t_cache)
init_locale_t_cache();
cache_entry = hash_search(locale_t_cache, &collid, HASH_ENTER, &found);
if (found)
return cache_entry->locale;
tp = SearchSysCache1(COLLOID, ObjectIdGetDatum(collid)); tp = SearchSysCache1(COLLOID, ObjectIdGetDatum(collid));
if (!HeapTupleIsValid(tp)) if (!HeapTupleIsValid(tp))
elog(ERROR, "cache lookup failed for collation %u", collid); elog(ERROR, "cache lookup failed for collation %u", collid);
collform = (Form_pg_collation) GETSTRUCT(tp);
collcollate = NameStr(((Form_pg_collation) GETSTRUCT(tp))->collcollate); collcollate = NameStr(collform->collcollate);
collctype = NameStr(((Form_pg_collation) GETSTRUCT(tp))->collctype); collctype = NameStr(collform->collctype);
if (strcmp(collcollate, collctype) == 0) if (strcmp(collcollate, collctype) == 0)
{ {
result = newlocale(LC_COLLATE_MASK | LC_CTYPE_MASK, collcollate, NULL); /* Normal case where they're the same */
result = newlocale(LC_COLLATE_MASK | LC_CTYPE_MASK, collcollate,
NULL);
if (!result) if (!result)
ereport(ERROR, ereport(ERROR,
(errcode_for_file_access(), (errcode_for_file_access(),
errmsg("could not create locale \"%s\": %m", collcollate))); errmsg("could not create locale \"%s\": %m",
collcollate)));
} }
else else
{ {
/* We need two newlocale() steps */
locale_t loc1; locale_t loc1;
loc1 = newlocale(LC_COLLATE_MASK, collcollate, NULL); loc1 = newlocale(LC_COLLATE_MASK, collcollate, NULL);
if (!loc1) if (!loc1)
ereport(ERROR, ereport(ERROR,
(errcode_for_file_access(), (errcode_for_file_access(),
errmsg("could not create locale \"%s\": %m", collcollate))); errmsg("could not create locale \"%s\": %m",
collcollate)));
result = newlocale(LC_CTYPE_MASK, collctype, loc1); result = newlocale(LC_CTYPE_MASK, collctype, loc1);
if (!result) if (!result)
ereport(ERROR, ereport(ERROR,
(errcode_for_file_access(), (errcode_for_file_access(),
errmsg("could not create locale \"%s\": %m", collctype))); errmsg("could not create locale \"%s\": %m",
collctype)));
} }
ReleaseSysCache(tp);
cache_entry->locale = result; cache_entry->locale = result;
return result; ReleaseSysCache(tp);
#else /* not HAVE_LOCALE_T */ #else /* not HAVE_LOCALE_T */
/* /*
* For platforms that don't support locale_t, check that we are * For platforms that don't support locale_t, we can't do anything
* dealing with the default locale. It's unlikely that we'll get * with non-default collations.
* here, but it's possible if users are creating collations even
* though they are not supported, or they are mixing builds in odd
* ways.
*/ */
if (!OidIsValid(collid))
elog(ERROR, "locale operation to be invoked, but no collation was derived");
else if (collid != DEFAULT_COLLATION_OID)
ereport(ERROR, ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("nondefault collations are not supported on this platform"))); errmsg("nondefault collations are not supported on this platform")));
return 0;
#endif /* not HAVE_LOCALE_T */ #endif /* not HAVE_LOCALE_T */
} }
return cache_entry->locale;
}

18
src/bin/initdb/initdb.c
View File

@ -1616,17 +1616,21 @@ setup_collation(void)
*/ */
skip = false; skip = false;
for (i = 0; i < len; i++) for (i = 0; i < len; i++)
{
if (IS_HIGHBIT_SET(localebuf[i])) if (IS_HIGHBIT_SET(localebuf[i]))
{
skip = true;
break;
}
}
if (skip)
{ {
if (debug) if (debug)
fprintf(stderr, _("%s: locale name has non-ASCII characters, skipped: %s\n"), fprintf(stderr, _("%s: locale name has non-ASCII characters, skipped: %s\n"),
progname, localebuf); progname, localebuf);
skipped++; skipped++;
skip = true;
break;
}
if (skip)
continue; continue;
}
enc = pg_get_encoding_from_locale(localebuf, debug); enc = pg_get_encoding_from_locale(localebuf, debug);
if (enc < 0) if (enc < 0)
@ -1635,7 +1639,7 @@ setup_collation(void)
continue; /* error message printed by pg_get_encoding_from_locale() */ continue; /* error message printed by pg_get_encoding_from_locale() */
} }
if (enc == PG_SQL_ASCII) if (enc == PG_SQL_ASCII)
continue; /* SQL_ASCII is handled separately */ continue; /* C/POSIX are already in the catalog */
PG_CMD_PRINTF2("INSERT INTO tmp_pg_collation (locale, encoding) VALUES ('%s', %d);", PG_CMD_PRINTF2("INSERT INTO tmp_pg_collation (locale, encoding) VALUES ('%s', %d);",
escape_quotes(localebuf), enc); escape_quotes(localebuf), enc);
@ -1651,10 +1655,6 @@ setup_collation(void)
escape_quotes(alias), escape_quotes(localebuf), enc); escape_quotes(alias), escape_quotes(localebuf), enc);
} }
for (i = PG_SQL_ASCII; i <= PG_ENCODING_BE_LAST; i++)
PG_CMD_PRINTF2("INSERT INTO tmp_pg_collation (locale, encoding) VALUES ('C', %d), ('POSIX', %d);",
i, i);
/* Add an SQL-standard name */ /* Add an SQL-standard name */
PG_CMD_PRINTF1("INSERT INTO tmp_pg_collation (collname, locale, encoding) VALUES ('ucs_basic', 'C', %d);", PG_UTF8); PG_CMD_PRINTF1("INSERT INTO tmp_pg_collation (collname, locale, encoding) VALUES ('ucs_basic', 'C', %d);", PG_UTF8);

2
src/include/catalog/catversion.h
View File

@ -53,6 +53,6 @@
*/ */
/* yyyymmddN */ /* yyyymmddN */
#define CATALOG_VERSION_NO 201103191 #define CATALOG_VERSION_NO 201103201
#endif #endif

11
src/include/catalog/pg_collation.h
View File

@ -58,8 +58,19 @@ typedef FormData_pg_collation *Form_pg_collation;
#define Anum_pg_collation_collcollate 5 #define Anum_pg_collation_collcollate 5
#define Anum_pg_collation_collctype 6 #define Anum_pg_collation_collctype 6
/* ----------------
* initial contents of pg_collation
* ----------------
*/
DATA(insert OID = 100 ( default PGNSP PGUID -1 "" "" )); DATA(insert OID = 100 ( default PGNSP PGUID -1 "" "" ));
DESCR("database's default collation"); DESCR("database's default collation");
#define DEFAULT_COLLATION_OID 100 #define DEFAULT_COLLATION_OID 100
DATA(insert OID = 950 ( C PGNSP PGUID -1 "C" "C" ));
DESCR("standard C collation");
#define C_COLLATION_OID 950
DATA(insert OID = 951 ( POSIX PGNSP PGUID -1 "POSIX" "POSIX" ));
DESCR("standard POSIX collation");
#define POSIX_COLLATION_OID 951
#endif /* PG_COLLATION_H */ #endif /* PG_COLLATION_H */

2
src/include/port.h
View File

@ -155,6 +155,8 @@ extern int pg_strcasecmp(const char *s1, const char *s2);
extern int pg_strncasecmp(const char *s1, const char *s2, size_t n); extern int pg_strncasecmp(const char *s1, const char *s2, size_t n);
extern unsigned char pg_toupper(unsigned char ch); extern unsigned char pg_toupper(unsigned char ch);
extern unsigned char pg_tolower(unsigned char ch); extern unsigned char pg_tolower(unsigned char ch);
extern unsigned char pg_ascii_toupper(unsigned char ch);
extern unsigned char pg_ascii_tolower(unsigned char ch);
#ifdef USE_REPL_SNPRINTF #ifdef USE_REPL_SNPRINTF

26
src/port/pgstrcasecmp.c
View File

@ -13,6 +13,10 @@
* *
* NB: this code should match downcase_truncate_identifier() in scansup.c. * NB: this code should match downcase_truncate_identifier() in scansup.c.
* *
* We also provide strict ASCII-only case conversion functions, which can
* be used to implement C/POSIX case folding semantics no matter what the
* C library thinks the locale is.
*
* *
* Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
* *
@ -123,3 +127,25 @@ pg_tolower(unsigned char ch)
ch = tolower(ch); ch = tolower(ch);
return ch; return ch;
} }
/*
* Fold a character to upper case, following C/POSIX locale rules.
*/
unsigned char
pg_ascii_toupper(unsigned char ch)
{
if (ch >= 'a' && ch <= 'z')
ch += 'A' - 'a';
return ch;
}
/*
* Fold a character to lower case, following C/POSIX locale rules.
*/
unsigned char
pg_ascii_tolower(unsigned char ch)
{
if (ch >= 'A' && ch <= 'Z')
ch += 'a' - 'A';
return ch;
}