> From: Bruno Haible <br...@clisp.org> > Cc: Eli Zaretskii <e...@gnu.org> > Date: Thu, 03 May 2018 05:42:17 +0200 > > On a recent mingw (both 32-bit and 64-bit), I'm seeing this test failure: > > FAIL: test-localename > ===================== > > ../../tests/test-localename.c:158: assertion 'strcmp (name, "de_DE.UTF-8") == > 0' failed > > Once this is worked around, there is another one here: > > /* Check that gl_locale_name_thread always returns NULL. */ > ASSERT (gl_locale_name_thread (LC_CTYPE, "LC_CTYPE") == NULL); > > It is due to the localename.c change from 2014-07-15 [1]. While the general > approach (to call GetThreadLocale only as a fallback, because GetThreadLocale > does not consider changes made by 'setlocale') is reasonable, the exact > placement of the new code is odd: Fixing the test failures in the tests > (attached: fix-tests.diff) makes the code even stranger. > > [1] https://lists.gnu.org/archive/html/bug-gnulib/2014-07/msg00003.html > > The better fix is this one:
Sorry, I've read the text and and diffs several times, and I still cannot see the light. Can you explain the rationale for moving code between gl_locale_name_thread and gl_locale_name_posix. I don't think I understand why that matters, given that both are called from gl_locale_name. I also don't understand why having a thread-level code inside a function whose name ends with "_thread" makes less sense than having non-Posix code in a function whose name ends with "_posix". As for the testing part, it sounds like you kept the changes I made? And yet you call doing that "odd"? What am I missing? And how all that is related to the failures which are the trigger for your proposal? Sorry for being so dumb this afternoon, and thanks in advance for explaining this.
