BUG #19474: LIKE with nondeterministic collations mis-handle literal backslashes in patterns containing escape

PG Bug reporting form <noreply@postgresql.org>

From: PG Bug reporting form <noreply@postgresql.org>
To: pgsql-bugs@lists.postgresql.org
Cc: zxwsbg12138@gmail.com
Date: 2026-05-09T02:22:23Z
Lists: pgsql-bugs
The following bug has been logged on the website:

Bug reference:      19474
Logged by:          Bowen Shi
Email address:      zxwsbg12138@gmail.com
PostgreSQL version: 18.3
Operating system:   centos
Description:        

After commit 85b7efa1cdd63c2fe2b70b725b8285743ee5787f ("Support LIKE with
nondeterministic collations"), LIKE on a nondeterministic collation can
return an incorrect result when the pattern contains a literal backslash.

The problem appears to be in MatchText() in
src/backend/utils/adt/like_match.c. In the nondeterministic-collation path,
when a pattern substring contains escape processing, the code builds an
unescaped copy of the substring. In that logic, a backslash that should
remain as a literal character can be dropped, so the substring compared by
pg_strncoll() is not the same as the original SQL pattern semantics.

As a result, a LIKE pattern that should match a string containing a literal
backslash can incorrectly return false.

SQL reproduction:

CREATE COLLATION ignore_accents (
    provider = icu,
    locale = 'und-u-ks-level1',
    deterministic = false
);

SELECT 'back\slash' COLLATE ignore_accents LIKE 'back\slash%' ESCAPE '#';

Expected result:
 t

Actual result:
 f

The same pattern works as expected without the nondeterministic collation
semantics.

A table-based reproduction:

CREATE COLLATION ignore_accents (
    provider = icu,
    locale = 'und-u-ks-level1',
    deterministic = false
);

CREATE TABLE like_test (val text);
INSERT INTO like_test VALUES ('back\slash');

SELECT val
FROM like_test
WHERE val COLLATE ignore_accents LIKE 'back\slash%' ESCAPE '#';

Expected result:
 one row: back\slash

Actual result:
 zero rows

This seems to be caused by the unescape logic in like_match.c for
nondeterministic collations, where a pattern fragment containing backslashes
is copied incorrectly before calling pg_strncoll().





Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →
  1. Support LIKE with nondeterministic collations