Andrey, I tested the numeric-locale TAP patch from attachment 189686 on native Windows ARM64 and emulated x64, against master 8e82d6610e7.
Using a local frontend fixture to select German_Germany.1252 for each process, the original pg_test_timing and psql tests fail on the decimal-comma output on both architectures. With your patch, both pass: 18 timing and 134 psql subtests. The timeouts and checkpoint logical-slot recovery tests also pass, two subtests each, along with all three setup groups. C process-locale timing/psql controls pass as well. This uses a controlled CRT process locale, not a Windows account with its regional settings changed. The fixture is local only. Other pending Windows changes were present in the integration builds, so these results establish the before/after TAP behavior in that setup. I found no blocker in the test changes. Happy to provide the detailed logs if useful. Best, Harrison Original discussion: https://www.postgresql.org/message-id/[email protected]
