On Sat, Dec 27, 2025 at 2:45 AM Jonathan Wakely <[email protected]> wrote:
>
> On Wed, 24 Dec 2025 at 17:01, 王靖淏 <[email protected]> wrote:
> >
> > On Mon, Dec 22, 2025 at 4:37 AM Jonathan Wakely <[email protected]>
> > wrote:
> > >
> > >
> > >
> > > On Sun, 21 Dec 2025, 19:44 Wang Jinghao, <[email protected]>
> > > wrote:
> > >>
> > >> Ignoring FORMAT_MESSAGE_IGNORE_INSERTS when retrieving system
> > >> error messages from FormatMessage may result in unexpected error
> > >> messages and
> > >> trigger new exceptions.
> > >
> > >
> > > For those like me who have no idea what this is about, this explains it:
> > > https://devblogs.microsoft.com/oldnewthing/20071128-00/?p=24353
> > >
> > > Can we add a test with error code 87?
> > I discovered that `std::local` does not affect the language used by
> > the system error information obtained by `FormatMessage`. It seems
> > that `void test3()` in
> > `libstdc++-v3\testsuite\19_diagnostics\error_category\system_category.cc`
> > cannot pass errors in non-English language environments.
> > >
> > >
> > >> This patch changes the
> > >> std::system_category.message(int) to be the same as the flags in
> > >> gcc/plugin.cc:win32_error_msg()
> > >>
> > >> libstdc++-v3/ChangeLog:
> > >>
> > >> * src/c++11/system_error.cc:
> > >
> > >
> > > This changelog entry is missing a description.
> > I'am sorry I forgot. ; (
> >
> > libstdc++-v3/ChangeLog:
> >
> > * src/c++11/system_error.cc (std::system_category): Add flag
> > FORMAT_MESSAGE_IGNORE_INSERTS and
> > FORMAT_MESSAGE_MAX_WIDTH_MASK to
> > dwFlag parameter of the FormatMessage function.
> >
> > >> ---
> > >> libstdc++-v3/src/c++11/system_error.cc | 4 +++-
> > >> 1 file changed, 3 insertions(+), 1 deletion(-)
> > >>
> > >> diff --git a/libstdc++-v3/src/c++11/system_error.cc
> > >> b/libstdc++-v3/src/c++11/system_error.cc
> > >> index b9a0b2c158f..ea7b5cd1226 100644
> > >> --- a/libstdc++-v3/src/c++11/system_error.cc
> > >> +++ b/libstdc++-v3/src/c++11/system_error.cc
> > >> @@ -162,7 +162,9 @@ namespace
> > >> char* buf = nullptr;
> > >> auto len
> > >> = FormatMessage(FORMAT_MESSAGE_FROM_SYSTEM
> > >> - | FORMAT_MESSAGE_ALLOCATE_BUFFER,
> > >> + | FORMAT_MESSAGE_ALLOCATE_BUFFER
> > >> + | FORMAT_MESSAGE_IGNORE_INSERTS
> > >> + | FORMAT_MESSAGE_MAX_WIDTH_MASK,
> > >> nullptr,
> > >> i,
> > >> LANG_USER_DEFAULT,
> > >> --
> > >> 2.52.0
> > >>
> >
> > I found that the existing `test03` function in
> > `libstdc++-v3/src/c++11/system_error.cc`, used for testing
> > `std::system_error_category.message(int)`, doesn't work correctly in
> > non-English language environments. This is likely because neither
> > `std::locale` nor `std::setlocale` affects the language of the error
> > message obtained by `FormatMessage()`.
>
> We pass LANG_USER_DEFAULT to FormatMessage, but I don't think that is
> related in any way to the C locale.
>
> I think the testcase could use SetThreadPreferredUILanguages to set
> en-US as the language that will be used by FormatMessage.
>
> > Furthermore, in non-English
> > environments, such as Chinese, system error messages end with "."\r\n"
> > instead of ".\r\n", which prevents `__builtin_memcmp(buf + len - 3,
> > ".\r\n", 3) {in libstdc++-v3\src\c++11\system_error.cc:178}` from
> > correctly removing the trailing "\r\n".
>
> Let's do this instead:
>
> if (len > 2 && !__builtin_memcmp(buf + len - 2, "\r\n", 2)) [[likely]]
> len -= 2 + (buf[len - 1] == '.');
^
Is buf[len - 1] == '.' always false here?
>
>
>
> >
> > This is my first time, and I'm not sure if I need to resubmit a patch.ಥ_ಥ
