patch suggestion: Fix citext_utf8 test's "Turkish I" with ICU collation provider
Hackorum builds and tests every patch posted to the lists, not only commitfest submissions. This is Hackorum's own CI rather than the PostgreSQL project's, and it is still under testing - please report anything that looks wrong.
You can run a PostgreSQL built from this patch straight from Docker, with no checkout and no build:
docker run --rm -p 5432:5432 ghcr.io/hackorum-dev/postgres-patch:t46818psql -h localhost -U postgresBuilt from patchset v1 (message #1), October 06, 2026 at 11:29 AM.
Every patchset is also pushed to a branch of our PostgreSQL fork, so you can check out the same tree CI built. Without a PostgreSQL checkout:
git clone --branch t46818_1 https://github.com/hackorum-dev/postgres.gitIn a checkout you already have, add the fork once:
git remote add hackorum https://github.com/hackorum-dev/postgres.gitthen, for this patchset and every later one:
git fetch hackorum t46818_1 && git checkout t46818_1Patchset v1 (message #1) is on t46818_1
Hello, hackers.
In current master, as well as in REL_15_STABLE, installcheck in
contrib/citext fails in most locales, if we use ICU as a locale provider:
$ rm -fr data; initdb --locale-provider icu --icu-locale en-US -D data
&& pg_ctl -D data -l logfile start && make -C contrib/citext
installcheck; pg_ctl -D data stop; cat contrib/citext/regression.diffs
...
test citext ... ok 457 ms
test citext_utf8 ... FAILED 21 ms
...
diff -u
/home/ashutosh/pg/REL_15_STABLE/contrib/citext/expected/citext_utf8.out
/home/ashutosh/pg/REL_15_STABLE/contrib/citext/results/citext_utf8.out
---
/home/ashutosh/pg/REL_15_STABLE/contrib/citext/expected/citext_utf8.out
2022-07-14 17:45:31.747259743 +0300
+++
/home/ashutosh/pg/REL_15_STABLE/contrib/citext/results/citext_utf8.out
2022-10-21 19:43:21.146044062 +0300
@@ -54,7 +54,7 @@
SELECT 'i'::citext = 'İ'::citext AS t;
t
---
- t
+ f
(1 row)The reason is that in ICU lowercasing Unicode symbol "İ" (U+0130
"LATIN CAPITAL LETTER I WITH DOT ABOVE") can give two valid results:
- "i", i.e. "U+0069 LATIN SMALL LETTER I" in "tr" and "az" locales.
- "i̇", i.e. "U+0069 LATIN SMALL LETTER I" followed by "U+0307 COMBINING
DOT ABOVE" in all other locales I've tried (including "en-US", "de",
"ru", etc).
And the way this test is currently written only accepts plain latin "i",
which might be true in glibc, but is not so in ICU. Verified on ICU
70.1, but I've seen this on few other ICU versions as well, so I think
this is probably an ICU's feature, not a bug(?).
Since we probably want installcheck in contrib/citext to pass on
databases with various locales, including reasonable ICU-based ones,
I suggest to fix this test by accepting either of outputs as valid.
I can see two ways of doing that:
1. change SQL in the test to use "IN" instead of "=";
2. add an alternative output.
I think in this case "IN" is better, because that allows a single
comment to address both possible outputs and to avoid unnecessary
duplication.
I've attached a patch authored mostly by my colleague, Roman Zharkov, as
one possible fix.
Only versions 15+ are affected.
Any comments?
--
Anton Voloshin
Postgres Professional, The Russian Postgres Company
https://postgrespro.ru