]> git.ipfire.org Git - thirdparty/gcc.git/commit
libstdc++: Fix wstring conversions in filesystem::path [PR95048]
authorJonathan Wakely <jwakely@redhat.com>
Fri, 11 Nov 2022 15:22:02 +0000 (15:22 +0000)
committerJonathan Wakely <jwakely@redhat.com>
Fri, 11 Nov 2022 17:16:42 +0000 (17:16 +0000)
commitb331bf303bdc1edead41e2b3d11d1a7804b433cf
treecc1835b8ef964957683d95879280460744d84020
parentdab5d73959cfc8f03cba548777adda9a798e1f0e
libstdc++: Fix wstring conversions in filesystem::path [PR95048]

In commit r9-7381-g91756c4abc1757 I changed filesystem::path to use
std::codecvt<CharT, char, mbstate_t> for conversions from all wide
strings to UTF-8, instead of using std::codecvt_utf8<CharT>. This was
done because for 16-bit wchar_t, std::codecvt_utf8<wchar_t> only
supports UCS-2 and not UTF-16. The rationale for the change was sound,
but the actual fix was not. It's OK to use std::codecvt for char16_t or
char32_t, because the specializations for those types always use UTF-8 ,
but std::codecvt<wchar_t, char, mbstate_t> uses the current locale's
encodings, and the narrow encoding is probably ASCII and can't support
non-ASCII characters.

The correct fix is to use std::codecvt only for char16_t and char32_t.
For 32-bit wchar_t we could have continued using std::codecvt_utf8
because that uses UTF-32 which is fine, switching to std::codecvt broke
non-Windows targets with 32-bit wchar_t. For 16-bit wchar_t we did need
to change, but should have changed to std::codecvt_utf8_utf16<wchar_t>
instead, as that always uses UTF-16 not UCS-2. I actually noted that in
the commit message for r9-7381-g91756c4abc1757 but didn't use that
option. Oops.

This replaces the unconditional std::codecvt<CharT, char, mbstate_t>
with a type defined via template specialization, so it can vary
depending on the wide character type. The code is also simplified to
remove some of the mess of #ifdef and if-constexpr conditions.

libstdc++-v3/ChangeLog:

PR libstdc++/95048
* include/bits/fs_path.h (path::_Codecvt): New class template
that selects the kind of code conversion done.
(path::_Codecvt<wchar_t>): Select based on sizeof(wchar_t).
(_GLIBCXX_CONV_FROM_UTF8): New macro to allow the same code to
be used for Windows and POSIX.
(path::_S_convert(const EcharT*, const EcharT*)): Simplify by
using _Codecvt and _GLIBCXX_CONV_FROM_UTF8 abstractions.
(path::_S_str_convert(basic_string_view<value_type>, const A&)):
Simplify nested conditions.
* include/experimental/bits/fs_path.h (path::_Cvt): Define
nested typedef controlling type of code conversion done.
(path::_Cvt::_S_wconvert): Use new typedef.
(path::string(const A&)): Likewise.
* testsuite/27_io/filesystem/path/construct/95048.cc: New test.
* testsuite/experimental/filesystem/path/construct/95048.cc: New
test.
libstdc++-v3/include/bits/fs_path.h
libstdc++-v3/include/experimental/bits/fs_path.h
libstdc++-v3/testsuite/27_io/filesystem/path/construct/95048.cc [new file with mode: 0644]
libstdc++-v3/testsuite/experimental/filesystem/path/construct/95048.cc [new file with mode: 0644]