An extension to expose functionality from ICU to PostgreSQL applications.
It requires PostgreSQL version 10 or newer, configured with ICU (--with-icu).
Note: this text is in GitHub Flavored Markdown format. Please see the version on github if it's rendered weirdly elsewhere.
The Makefile uses the PGXS infrastructure to find include and library files and determine the install location.
Build and install with:
$ make
$ (sudo) make install
icu_char_name
icu_character_boundaries
icu_collation_attributes
icu_compare
icu_confusable_strings_check
icu_default_locale
icu_is_normalized
icu_line_boundaries
icu_locales_list
icu_normalize
icu_number_spellout
icu_replace
icu_sentence_boundaries
icu_set_default_locale
icu_sort_key
icu_spoof_check
icu_strpos
icu_transform
icu_transforms_list
icu_unicode_version
icu_version
icu_word_boundaries
These functions work in both Unicode and non-Unicode databases.
Returns the version of the ICU library linked with the server.
Returns the version of the Unicode standard used by the ICU library linked with the server.
Returns a table-type list of available ICU locales with their main properties
(country code and name, language code and name, script, direction).
When translations are available, the country and language names are
localized with the default ICU locale, configurable with
icu_set_default_locale()
. Set it to en
to force english names.
Examples:
=# SELECT * FROM icu_locales_list() where name like 'es%' limit 5;
name | country | country_code | language | language_code | script | direction
--------+---------------+--------------+----------+---------------+--------+-----------
es | | | Spanish | spa | | LTR
es_419 | Latin America | | Spanish | spa | | LTR
es_AR | Argentina | ARG | Spanish | spa | | LTR
es_BO | Bolivia | BOL | Spanish | spa | | LTR
es_CL | Chile | CHL | Spanish | spa | | LTR
=# SELECT name,country FROM icu_locales_list() where script='Simplified Han';
name | country
------------+---------------------
zh_Hans |
zh_Hans_CN | China
zh_Hans_HK | Hong Kong SAR China
zh_Hans_MO | Macau SAR China
zh_Hans_SG | Singapore
This list is obtained independently from the collations declared to
PostgreSQL (found in pg_collation
).
Lists the attributes, version and display name of an ICU collation,
returned as a set of (attribute,value)
tuples. The collator
argument must designate an
ICU collator and accepts
several different syntaxes. In particular,
a locale ID or (if ICU>=54)
language tags
may be used.
Note that this argument is not a reference to a PostgreSQL collation, and
that this function does not depend on whether a corresponding
collation has been instantiated in the database with CREATE COLLATION
.
To query the properties of an already created PostgreSQL
ICU collation, refer to pg_collation.collcollate
(which corresponds
to the lc_collate
argument of CREATE COLLATION).
=# SELECT a.attribute,a.value FROM pg_collation
JOIN LATERAL icu_collation_attributes(collcollate) a
ON (collname='fr-CA-x-icu');
attribute | value
-------------+-------------------
displayname | français (Canada)
kn | false
kb | true
kk | false
ka | noignore
ks | level3
kf | false
kc | false
kv | punct
version | 153.80.33
icu_collation_attributes()
is useful to check that the settings embedded into a
collation name activate the intended options, because ICU parses them
in a way that non-conformant parts tend to be silently ignored,
and because the interpretation
somewhat depends on the ICU version (in particular, pre-54 versions do
not support options expressed as BCP-47 tags). It may be also
useful to search existing collations by their properties. When
exclude_defaults
is set to true
, attributes that have their
default value are filtered out, to put in evidence the specifics
of collations. For instance, to find the only collations that use
shifted
for the Alternate
attribute:
=# SELECT collname,collcollate,a.attribute,a.value FROM pg_collation
JOIN LATERAL icu_collation_attributes(collcollate,true) a
ON (attribute='ka') ;
collname | collcollate | attribute | value
-------------+-------------+-----------+---------
th-x-icu | th | ka | shifted
th-TH-x-icu | th-TH | ka | shifted
(2 rows)
By default there is no filtering (exclude_defaults
= false) so that
all attributes known by the function as well as the collation version
number are reported.
Example of checking a collation without any reference to pg_collation
:
=# SELECT * FROM icu_collation_attributes('fr-u-ks-level2-kn');
attribute | value
-----------+----------
kn | true
kb | false
kk | false
ka | noignore
ks | level2
kf | false
kc | false
version | 153.64
icu_collation_attributes()
will error out if ICU is unable to open
a collator with the given argument.
Returns the binary sort key (type: bytea
) corresponding to the
string with the given collation.
See http://userguide.icu-project.org/collation/architecture#TOC-Sort-Keys
When a collator
argument is passed, it is interpreted as an ICU
locale independently of the persistent collations instantiated in
the database.
When there is no collator
argument, the collation associated
to string
gets used to generate the sort key. It must be
an ICU collation or the function will error out. This form
with a single argument is faster due to Postgres keeping its
collations "open" (in the sense of ucol_open()/ucol_close()
) for the
duration of the session, whereas the other form with the
explicit collator
argument does open and close the ICU collation
for each call.
Binary sort keys may be useful to circumvent a core PostgreSQL
limitation that two strings that differ in their byte representation
are never considered equal by deterministic collations (see for instance this thread in the pgsql-bugs mailing-list for a discussion of this problem in relation with the ICU integration).
With PostgreSQL 12 or newer versions, the "deterministic" property can be set
to false
by CREATE COLLATION
to request that string comparisons with these collations skip the tie-breaker.
With older versions, "deterministic" is always true
.
You may order or rank by binary sort keys, or materialize them in a unique index to achieve at the SQL level what cannot be done internally by persistent collations, either because PostgreSQL is not recent enough or because you don't want or lack the permission to instantiate nondeterministic collations.
The function is declared IMMUTABLE to be usable in indexes, but please be aware that it's only true as far as the "version" of the collation doesn't change. (Typically it changes with every version of Unicode). In short, consider rebuilding the affected indexes on ICU upgrades.
To simply compare pairs of strings, consider icu_compare()
instead.
Example demonstrating a case-sensitive, accent-sensitive unique index:
=# CREATE TABLE uniq(name text);
=# CREATE UNIQUE INDEX idx ON uniq((icu_sort_key(name, 'fr-u-ks-level1')));
=# INSERT INTO uniq values('été');
INSERT 0 1
=# INSERT INTO uniq values('Ête');
ERROR: duplicate key value violates unique constraint "idx"
DETAIL: Key (icu_sort_key(name, 'fr-u-ks-level1'::text))=(\x314f31) already exists.
=# insert into uniq values('Êtes');
INSERT 0 1
Compare two strings with the given collation. Return the result as a signed integer, similarly to strcoll(), that is, the result is negative if string1 < string2, zero if string = string2, and positive if string1 > string2.
When a collator
argument is passed, it is taken as the ICU
locale (independently of the collations instantiated in the database)
to use to collate the strings.
When there is no collator
argument, the collation associated
to string1
and string2
gets used for the comparison.
It must be an ICU collation and it must be the same for the two
arguments or the function will error out. With PostgreSQL 12 or newer,
it can be nondeterministic, but whether it is nondeterministic
or deterministic will not make any difference in the result of icu_compare
,
contrary to comparisons done by PostgreSQL core with the equality operator.
The two-argument form is significantly faster due to Postgres keeping its
collations "open" (in the sense of ucol_open()/ucol_close()
) for the
duration of the session, whereas the other form with the
explicit collator
argument does open and close the ICU collation
for each call.
Example: case-sensitive, accent-insensitive comparison:
=# SELECT icu_compare('abcé', 'abce', 'en-u-ks-level1-kc-true');
icu_compare
-------------
0
=# SELECT icu_compare('Abcé', 'abce', 'en-u-ks-level1-kc-true');
icu_compare
-------------
1
With two arguments and a collation determined by the COLLATE clause:
=# SELECT icu_compare('Abcé', 'abce' COLLATE "fr-x-icu");
icu_compare
-------------
1
With an implicit Postgres collation:
=# CREATE COLLATION mycoll (locale='fr-u-ks-level1', provider='icu');
CREATE COLLATION
=# CREATE TABLE books (id int, title text COLLATE "mycoll");
CREATE TABLE
=# insert into books values(1, $$C'est l'été$$);
INSERT 0 1
=# select id,title from books where icu_compare (title, $$c'est l'ete$$) = 0;
id | title
----+-------------
1 | C'est l'été
Sets the default ICU locale for the session, and returns a canonicalized
version of the locale name. The POSIX syntax (lang[_country[@attr]]
)
is accepted. Call this function to change the output language of
icu_locales_list()
.
This setting should not have any effect on PostgreSQL core
functions, at least as of PG version 10.
Warning: passing bogus contents to this function may freeze the backend with older versions of ICU (seen with 52.1).
Returns the name of the default ICU locale as a text. The initial value is automatically set by ICU from the environment.
Break down the string into its characters and return them as a set of text.
This is comparable to calling regexp_split_to_table
with an empty regexp,
with some differences, for instance:
- CRLF sequences do not get split into two characters.
- Sequences with a base and a combining character are kept together.
- Legacy and extended grapheme clusters are extracted as one result per grapheme.
Example (the "e" followed by the combining acute accent U+0301 may be rendered as an accented e or differently depending on your browser):
=# SELECT * FROM icu_character_boundaries('Ete'||E'\u0301', 'fr') as chars;
chars
-------
E
t
é
See Boundary Analysis in the ICU User Guide and UAX #29 (Unicode Text Segmentation) for more information.
Break down the string into words and non-words constituents,
and return them in a set of (tag, contents) tuples.
tag
has values from the UWordBreak enum defined in ubrk.h indicating the nature of the piece of contents.
The current values are:
UBRK_WORD_NONE = 0,
UBRK_WORD_NUMBER = 100,
UBRK_WORD_LETTER = 200,
UBRK_WORD_KANA = 300,
UBRK_WORD_IDEO = 400, /* up to 500 */
(strictly speaking, any number between the lower and the upper bounds may be counted, as these numbers are meant to be intervals inside which new subdivisions may be added in future versions of ICU).
Example:
=# SELECT * FROM icu_word_boundaries($$I like O'Reilly books, like the japanese 初めてのPerl 第7版.$$ , 'en');
tag | contents
-----+----------
200 | I
0 |
200 | like
0 |
200 | O'Reilly
0 |
200 | books
0 | ,
0 |
200 | like
0 |
200 | the
0 |
200 | japanese
0 |
400 | 初めて
400 | の
200 | Perl
0 |
400 | 第
100 | 7
400 | 版
0 | .
or to count words in english:
=# SELECT count(*) FROM icu_words_boundaries($$piece of text$$, 'en_US')
WHERE tag=200;
Split the string into pieces where a line break may occur, according
to the Unicode line breaking algorithm defined in UAX #14,
and return them in a set of (tag, contents) tuples.
tag
has values from the ULineBreakTag enum defined in ubrk.h indicating the nature of the break.
The current values are:
UBRK_LINE_SOFT = 0,
UBRK_LINE_HARD = 100, /* up to 200 */
(strictly speaking, any number between the lower and the upper bounds may be counted, as these numbers are meant to be intervals inside which new subdivisions may be added in future versions of ICU).
Example:
=# SELECT *,convert_to( contents, 'utf-8') from icu_line_boundaries(
$$Thus much let me avow--You are not wrong, who deem
That my days have been a dream;
Yet if hope has flown away
In a night, or in a day,$$
, 'en');
tag | contents | convert_to
-----+----------+------------------
100 | +| \x0a
| |
0 | Thus | \x5468757320
0 | much | \x6d75636820
0 | let | \x6c657420
0 | me | \x6d6520
100 | avow-- +| \x61766f772d2d0a
| |
0 | You | \x596f7520
0 | are | \x61726520
0 | not | \x6e6f7420
0 | wrong, | \x77726f6e672c20
0 | who | \x77686f20
100 | deem +| \x6465656d0a
| |
0 | That | \x5468617420
0 | my | \x6d7920
0 | days | \x6461797320
0 | have | \x6861766520
0 | been | \x6265656e20
0 | a | \x6120
100 | dream; +| \x647265616d3b0a
| |
0 | Yet | \x59657420
0 | if | \x696620
0 | hope | \x686f706520
0 | has | \x68617320
0 | flown | \x666c6f776e20
100 | away +| \x617761790a
| |
0 | In | \x496e20
0 | a | \x6120
0 | night, | \x6e696768742c20
0 | or | \x6f7220
0 | in | \x696e20
0 | a | \x6120
0 | day, | \x6461792c
Split the string into sentences, according the Unicode text segmentation
rules defined in UAX #29,
and return them in a set of (tag, contents) tuples.
tag
has values from the USentenceBreakTag enum defined in ubrk.h indicating the nature of the break.
The current values are:
UBRK_SENTENCE_TERM = 0,
UBRK_SENTENCE_SEP = 100, /* up to 200 */
(strictly speaking, any number between the lower and the upper bounds may be counted, as these numbers are meant to be intervals inside which new subdivisions may be added in future versions of ICU).
Example:
=# SELECT * FROM icu_sentence_boundaries('Mr. Barry Sheene was born in 1950. He was a motorcycle racer.',
'en-u-ss-standard');
tag | contents
-----+-------------------------------------
0 | Mr. Barry Sheene was born in 1950.
0 | He was a motorcycle racer.
Note: "Mr." followed by a space is recognized by virtue of the locale as an abbreviation of the english "Mister", rather than the end of a sentence.
Return the spelled out text corresponding to the number expressed in the given locale.
Example:
=# SELECT loc, icu_number_spellout(1234, loc)
FROM (values ('en'),('fr'),('de'),('ru'),('ja')) AS s(loc);
loc | icu_number_spellout
-----+-------------------------------------------
en | one thousand two hundred thirty-four
fr | mille deux cent trente-quatre
de | eintausendzweihundertvierunddreißig
ru | одна тысяча двести тридцать четыре
ja | 千二百三十四
(Note: the german output uses U+00AD (SOFT HYPHEN) to separate words. Github's markdown to HTML conversion seems to remove them, so in the above text the spellout might appear like a single long word.)
Return the Unicode character name corresponding to the first codepoint of the input.
Example:
=# SELECT c, to_hex(ascii(c)), icu_char_name(c)
FROM regexp_split_to_table('El Niño', '') as c;
c | to_hex | icu_char_name
---+--------+---------------------------------
E | 45 | LATIN CAPITAL LETTER E
l | 6c | LATIN SMALL LETTER L
| 20 | SPACE
N | 4e | LATIN CAPITAL LETTER N
i | 69 | LATIN SMALL LETTER I
ñ | f1 | LATIN SMALL LETTER N WITH TILDE
o | 6f | LATIN SMALL LETTER O
Return a boolean indicating whether the argument is likely to be an attempt at confusing a reader. The implementation is based on Unicode Technical Reports #36 and #39 and uses the ICU default settings for spoof checks.
Example:
=# SELECT txt, icu_spoof_check(txt) FROM (VALUES ('paypal'), (E'p\u0430ypal')) AS s(txt);
txt | icu_spoof_check
--------+-----------------
paypal | f
pаypal | t
(Note: The second character in the second row is U+0430 (CYRILLIC SMALL LETTER A) instead of the genuine ASCII U+0061 (LATIN SMALL LETTER A))
Return a boolean indicating whether the string arguments are visually confusable with each other, according to data described in Unicode Technical Report #39. The settings and comparison levels are ICU defaults. For strictly identical strings, it returns true.
Example:
=# SELECT txt, icu_confusable_strings_check('phil', txt) AS confusable
FROM (VALUES ('phiL'), ('phiI'), ('phi1'), (E'ph\u0131l')) AS s(txt);
txt | confusable
------+------------
phiL | f
phiI | t
phi1 | t
phıl | t
Return a string with some transformations applied. This function essentially calls ICU's utrans_transUChars().
The first argument is the string to transform, and the second is the transformation
to apply, expressed as a sequence of transforms and filters
(see the ICU user guide on transforms
and the output of icu_transforms_list()
mentioned below).
Examples:
Transliterate:
=# select icu_transform('Владимир Путин', 'Cyrl-Latn'); -- just 'Latin' would work here too
icu_transform
----------------
Vladimir Putin
Transform Unicode names into the corresponding characters:
=# select icu_transform('10\N{SUPERSCRIPT MINUS}\N{SUPERSCRIPT FOUR}'
'\N{MICRO SIGN}m = 1 \N{ANGSTROM SIGN}',
'Name-Any');
icu_transform
---------------
10⁻⁴µm = 1 Å
Remove diacritics (generalized "unaccent") through Unicode decomposition.
=# select icu_transform('1 Å', 'any-NFD; [:nonspacing mark:] any-remove; any-NFC');
icu_transform
---------------
1 A
Generate hexadecimal codepoints for non-ASCII characters:
=# select icu_transform('Ich muß essen.', '[:^ascii:]; Hex');
icu_transform
---------------------
Ich mu\u00DF essen.
Return the list of built-in transliterations or transforms, as a set of text,
corresponding to "Basic IDs" in ICU documentation.
The initial set of transforms are transliterations between scripts
(like Katakana-Latin
or Latin-Cyrillic
), but they're
supplemented with functionalities related to accents, casing,
Unicode composition and decomposition with combining characters
and other conversions.
Values from this list are meant to be used individually as the 2nd argument of
icu_transform()
, or assembled with semi-colon separators to form
compound transforms, possibly with filters added to limit the set of characters
to transform.
Like strpos(text,text)
in Postgres core, except that it uses the
linguistic rules of collator
to search substring
in string
,
and that it supports nondeterministic collations seamlessly.
When the substring is not found, it returns 0. Otherwise, It returns
the 1-based position of the first match of substring
inside
string
, or 1 if substring
is empty.
When collator
is not passed, the collation of the
arguments is used. As with the other functions in this extension, the
two-argument form is faster since it can keep the ICU collation open
across function calls.
Example:
-- Search in names independently of punctuation, case and accents
=# select name from addresses where
icu_strpos(name, 'jeanrene', 'fr-u-ks-level1-ka-shifted') > 0
name
------------------
jean-rené dupont
Jean-René Dupont
jeanrenédupont
Like replace(string text, from text, to text)
in Postgres core,
except it uses the linguistic rules of collator
to search
substring
in string
instead of a byte-wise comparison. It also
supports nondeterministic collations to search from
as a substring.
It returns strings
with all substrings that match from
replaced by to
.
When collator
is not passed, the collation of the arguments is used,
which is faster because the ICU collation can be kept open across
function calls.
Example:
-- Collation comparing independently of punctuation, case and accents
=# CREATE COLLATION ciaipi (provider = icu, locale = 'und-u-ks-level1-ka-shifted');
-- Replace names matching 'jeanrene' by a placeholder
=# select s.n, icu_replace(n, 'jeanrene', '{firstname}' collate "ciaipi")
from (values('jeanrenédupont'),('Jean-René Dupont')) as s(n) ;
n | icu_replace
-------------------+---------------------
jeanrenédupont | {firstname}dupont
Jean-René Dupont | {firstname} Dupont
Return string
transformed into the Unicode normalized form
,
which must be nfc
, nfkc
, nfd
, or nfkd
(upper case or mixed
case variants are accepted). Returns NULL if any input argument is NULL.
The database must use an Unicode encoding, which means UTF-8 in practice.
See the Unicode Annex UAX #15
for an introduction on Unicode normal forms.
Example:
=# select icu_normalize('éte'||E'\u0301', 'nfc') = E'ét\u00E9';
?column?
----------
t
Return true if string
is in the Unicode normalized form
,
which must be nfc
, nfkc
, nfd
, or nfkd
(upper case or mixed
case variants are accepted). Returns false otherwise, or NULL if
any input argument is NULL. The database must use an Unicode encoding,
which means UTF-8 in practice.
Example:
=# SELECT icu_is_normalized('ét'||E'\u0301', 'nfc');
icu_is_normalized
-------------------
f
=# SELECT icu_is_normalized('ét'||E'\u0301', 'nfd');
icu_is_normalized
-------------------
t
This project is licensed under the PostgreSQL License -- see LICENSE.md.