Postgres Pro
Downloads
Home   >   mailing lists

Thread: ICU warnings during make installcheck and text_extensions test

ICU warnings during make installcheck and text_extensions test

From
Oleg Tselebrovskiy
Date:
Greetings, everyone!

When you try to run installcheck using a cluster that was initialized 
with ICU (./initdb -D ../data --locale-provider=icu 
--icu-locale='en_US_POSIX') and NO_LOCALE=1 you get a warning:

    # +++ regress install-check in src/test/regress +++
    # using postmaster on Unix socket, default port
    WARNING:  could not convert locale name "C" to language tag: 
U_ILLEGAL_ARGUMENT_ERROR
    WARNING:  ICU locale "C" has unknown language "c"
    HINT:  To disable ICU locale validation, set the parameter 
"icu_validation_level" to "disabled".

This happens with PostgreSQL 16, 17 and master

While this case is somewhat superficial (you need ICU and NO_LOCALE), 
when you try to run installcheck of test_extensions test module (using a 
cluster initialized with ICU) you will get the same warnings due to 
NO_LOCALE=1 in Makefile

There could be an argument "if you are using ICU and don't want warnings 
just set icu_validation_level=disabled in postgresql.conf", but then 
installcheck fails because of collate.icu.utf8 like this:

    CREATE COLLATION testx (provider = icu, locale = 
'@colStrength=primary;nonsense=yes'); DROP COLLATION testx;
    -WARNING:  could not convert locale name 
"@colStrength=primary;nonsense=yes" to language tag: 
U_ILLEGAL_ARGUMENT_ERROR
    CREATE COLLATION testx (provider = icu, locale = 'nonsense-nowhere'); 
DROP COLLATION testx;
    -WARNING:  ICU locale "nonsense-nowhere" has unknown language 
"nonsense"
    -HINT:  To disable ICU locale validation, set the parameter 
"icu_validation_level" to "disabled".
    CREATE COLLATION test4 FROM nonsense;

And we definitely want to check warnings there

So for now I propose adding icu_validation_level=disabled to pg_regress 
when we were passed NO_LOCALE=1 (patch is attached)

Also since locale 'C' isn't converted to anything with ICU since 
f3a01af, maybe we want to somehow handle NO_LOCALE with ICU in a special 
way? Maybe only during tests?

On another none, test test_extensions from 
src/test/modules/test_extensions fails during installcheck when the 
cluster was initialized with ICU locale. It was already reported at [1]. 
This test fails on current REL_17_STABLE (2530367) and master (2a5e709)

One way to fix it is to just skip the test if we find ICU locale 
provider (patch attached bellow), but I'm not sure this is the optimal 
way. In [1] there was an attempt to replace \dx+ with function with 
collation-independent output

[1] 
https://www.postgresql.org/message-id/a2fd691f8d9f1325cec1283b161d6a8e@postgrespro.ru
Attachment