collation

The collation template converts the input into a single locale-aware collation key for the configured LOCALE, rather than into search tokens. A collation key is a transformed form of the string whose byte order matches the locale's sorting rules, so comparing or ordering the keys yields linguistically correct results — for example placing ä where the locale expects it relative to a and z.

Use this template when you need locale-correct sorting or equality over an indexed column. It produces one token per value and is not intended for free-text matching.

Options

Option	Type	Default	Description
LOCALE	string	—	ICU locale

Tokenization

collation emits exactly one token per value: an opaque binary collation key. The key's bytes are not human-readable and you never inspect them directly — their value is that comparing two keys reproduces the locale's sort order. For an en_US locale the key for apple sorts before the key for banana, just as the words do.

Input	LOCALE	Output
apple	en_US.UTF-8	binary collation key (sorts before banana)
banana	en_US.UTF-8	binary collation key (sorts after apple)

The example below builds a collation dictionary and confirms that the key for apple orders before the key for banana:

Query

CREATE TEXT SEARCH DICTIONARY collation_dict (    template = 'collation',    locale = 'en_US.UTF-8');
# A collation dictionary emits one opaque binary collation key per value.# The bytes are not human-readable, but comparing the keys reproduces the# locale's sort order, so `apple` orders before `banana`.
SELECT ts_lexize('collation_dict', 'apple') < ts_lexize('collation_dict', 'banana') AS apple_sorts_before_banana;

Result

 apple_sorts_before_banana--------------------------- t

See also

• keyword — keeps the literal value as one token for exact matching
• norm — normalizes a value for case- or accent-insensitive equality
• CREATE TEXT SEARCH DICTIONARY