Here's another mystery from Windows + MinGW. Although "fairywren" is
green, that is because it lacks ICU, which would activate extra tests.
CI is green too, but the optional CI task "Windows - Server 2019,
MinGW64 - Meson" has ICU and it is now failing if you trigger it[1]
after commit 35eeea62, in initdb/001_initdb:
[05:43:49.764] | 146/305 - options --locale-provider=icu --locale=und
--lc-*=C: no stderr FAIL
... because it logs a warning to stderr:
WARNING: no usable system locales were found
I can only assume there was some extra dependency on setlocale()
global state changes in the removed code. I don't quite get it, but
whatever the reason, it's less than helpful to have different
compilers taking different code paths on our weirdest OS that most of
us don't use, so I propose to push this change to take the regular
MSVC code path for MinGW too, when looking up code pages. Somehow,
this fixes that, though it'd probably take someone with a local MinGW
setup to dig into what exactly is happening there.
(There are plenty more places where we do something different for
MinGW. I suspect they are all obsolete problems. We should probably
just harmonise everything and see what breaks now that we have a CI
system, but that can be for another day.)
That warning is from pg_import_system_locales(), which is new-ish
(v16) on that OS. It was recently discovered to trigger a
pre-existing problem[2]: the simple setlocale() save/restore pattern
doesn't work in general on Windows, because some local names are
non-ASCII, and the restore can fail (abort in the system library due
to bad encoding, because the intermediate setlocale() changed the
expected encoding of the locale name itself). So it's good that we
aren't doing that anymore in this location; I'm just thinking out loud
about whether that phenomenon could also be somehow connected to this
failure, though I don't see it. Another realisation is that my
pg_localeconv_r() patch, which can't avoid a thread-safe setlocale()
save-and-restore on that OS (and might finish up being the only one
left in the tree by the time we're done?), had better use wsetlocale()
instead to defend itself against that particular madness.
[1] https://cirrus-ci.com/task/5928104793735168
[2]
https://www.postgresql.org/message-id/CA%2BhUKG%2BFxeRLURZ%3Dn8NPyLwgjFds_SqU_cQvE40ks6RQKUGbGg%40mail.gmail.com
From 21acdecdeec65bc1f847c58d5afffd390e826c73 Mon Sep 17 00:00:00 2001
From: Thomas Munro <thomas.mu...@gmail.com>
Date: Wed, 14 Aug 2024 09:19:13 +1200
Subject: [PATCH] Harmonize MinGW CODESET lookup with MSVC.
Historically, MinGW environments lacked some Win32 calls, so we took a
different code path in win32_langinfo(). Somehow, the code change in
commit 35eeea62 (removing setlocale() calls) caused one particular
001_initdb.pl test to fail on MinGW + ICU builds, because
pg_import_system_collations() found no collations. It might take a
MinGW user to discover the exact reason.
Updating that function to use the same code as MSVC seems to fix that
test, so lets do that. (There are plenty more places that test for MSVC
unnecessarily, to be investigated later.)
While here, also rename the helper function win32_langinfo() to
win32_get_codeset(), to explain what it does less confusingly.
---
src/port/chklocale.c | 9 +++------
1 file changed, 3 insertions(+), 6 deletions(-)
diff --git a/src/port/chklocale.c b/src/port/chklocale.c
index 99e27ed6de..9506cd87ed 100644
--- a/src/port/chklocale.c
+++ b/src/port/chklocale.c
@@ -193,7 +193,7 @@ static const struct encoding_match encoding_match_list[] = {
#ifdef WIN32
/*
- * On Windows, use CP<code page number> instead of the nl_langinfo() result
+ * On Windows, use CP<code page number> instead of CODESET.
*
* This routine uses GetLocaleInfoEx() to parse short locale names like
* "de-DE", "fr-FR", etc. If those cannot be parsed correctly process falls
@@ -203,12 +203,10 @@ static const struct encoding_match encoding_match_list[] = {
* Returns a malloc()'d string for the caller to free.
*/
static char *
-win32_langinfo(const char *ctype)
+win32_get_codeset(const char *ctype)
{
char *r = NULL;
char *codepage;
-
-#if defined(_MSC_VER)
uint32 cp;
WCHAR wctype[LOCALE_NAME_MAX_LENGTH];
@@ -233,7 +231,6 @@ win32_langinfo(const char *ctype)
}
}
else
-#endif
{
/*
* Locale format on Win32 is <Language>_<Country>.<CodePage>. For
@@ -336,7 +333,7 @@ pg_get_encoding_from_locale(const char *ctype, bool write_message)
freelocale(loc);
#else
- sys = win32_langinfo(ctype);
+ sys = win32_get_codeset(ctype);
#endif
if (!sys)
--
2.39.2
