ping @w17

from libc++'s implementation:

 18   _LIBCPP_HIDE_FROM_ABI std::string string() const { return string<char>(); }
 17   _LIBCPP_HIDE_FROM_ABI __u8_string u8string() const {
 16     using _CVT = __narrow_to_utf8<sizeof(wchar_t) * __CHAR_BIT__>;
 15     __u8_string __s;
 14     __s.reserve(__pn_.size());
 13     _CVT()(back_inserter(__s), __pn_.data(), __pn_.data() + __pn_.size());
 12     return __s;
 11   }

u8string seems to do the conversion.

oh this is funny.

so I renamed the exiv2 directory to ☺ in the hopes of getting some failures. I got stopped by CMake itself:

CMake Error: Could not read presets from /home/mangix/devstuff/☺:
File not found: /home/mangix/devstuff/☺/CMakePresets.json

HAH. Even meson fails:

meson a;meson compile -C a

ninja: fatal: chdir to 'C:/Users/mangix/scoop/apps/msys2/2023-01-27/home/mangix/devstuff/?/a' - Invalid argument
ninja: Entering directory `C:/Users/mangix/scoop/apps/msys2/2023-01-27/home/mangix/devstuff/?/a'

cd a
ninja

src/lib_convert.a.p/convert.cpp.obj
"ccache" "c++" "-Isrc/lib_convert.a.p" "-Isrc" "-I../src" "-I." "-I.." "-I../include/exiv2" "-I../xmpsdk/include" "-fvisibility=hidden" "-fdiagnostics-color=always" "-D_LIBCPP_HARDENING_MODE=_LIBCPP_HARDENING_MODE_FAST" "-D_FILE_OFFSET_BITS=64" "-std=c++20" "-O0" "-g" "-DEXV_LOCALEDIR=\"C:/Users/mangix/scoop/apps/msys2/2023-01-27/clang64/share/locale\"" "-DEXIV2API=__declspec(dllexport)" -MD -MQ src/lib_convert.a.p/convert.cpp.obj -MF "src/lib_convert.a.p/convert.cpp.obj.d" -o src/lib_convert.a.p/convert.cpp.obj "-c" ../src/convert.cpp
libc++abi: terminating due to uncaught exception of type std::__1::__fs::filesystem::filesystem_error: filesystem error: in __wide_to_char: Illegal byte sequence

When I worked on replacing string paths with filesystem::path for my purposes, one of the challenges was handling actual files vs. other I/O implementations. I needed only files to work with arbitrary characters in file paths, so using filesystem::path in FileIo worked well for that, but I also had to leave the original string path methods in place, so people could still use them to open URLs, etc.

This is an export of the DGML I created when I reworked FileIo to use filesystem::path, which shows the components that would need additional changes, such as BasicIo::path() having to return a temporary because there is no longer a string in FileIo to return a reference to.

The export tool didn't do a good job and some arrows aren't leading to inner boxes, so some comments may look confusing, but it still should give you a good idea of the scope of the change.

I also want to mention that I think the approach or re-testing everything with paths containing various characters is a good idea - see what's working and what isn't before implementing a significant change like this. I would only do it against the test data, like image names with emojis, not with build tools, like project paths. It would be a good idea to introduce all types of characters in these paths, like A, £, ゎ, 🌎, to cover 1-, 2-, 3- and 4-byte characters.

These tests will probably reveal things like calling ReplaceFileA, etc., and calling filesystem::path::string, which will throw exceptions for characters outside of Win1252, unless the locale is set to en_US.UTF-8:

    std::filesystem::path p(u8"A" u8"\U0001F30E" u8"B");
    std::wstring ws = p.wstring();
    std::string ns = p.string();        // will throw unless std::locale("en_US.UTF-8") is set
    std::u8string us = p.u8string();

The locale may present a challenge if people want to run it for non-UTF-8 locales, but still be able to open files with emojis. This may be tricky without an explicit character handling like that in the diagram.

Lastly, Windows 10 reaches the end of support in 2025, so probably isn't as important for this anymore.

Codecov Report

Attention: Patch coverage is 80.00000% with 1 line in your changes missing coverage. Please review.

Project coverage is 63.88%. Comparing base (2e6fb08) to head (8fc087b).
Report is 48 commits behind head on main.

@@            Coverage Diff             @@
##             main    #3264      +/-   ##
==========================================
- Coverage   63.91%   63.88%   -0.04%     
==========================================
  Files         104      104              
  Lines       21752    21714      -38     
  Branches    10636    10631       -5     
==========================================
- Hits        13903    13872      -31     
+ Misses       5641     5634       -7     
  Partials     2208     2208              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

  Actual: it throws Exiv2::Error with description "D:/a/exiv2/exiv2/test/data\Реган.jp2: Failed to open the data source: No such file or directory (errno = 2)".

boo.

Реган.jp2

I cannot figure out the actual character for the first pair bytes, but the subsequent pairs appear to be spelling еган and if that's the case, it means it interprets UTF-8 as Latin1, which would point to an instance of std::string interpreted without setting the locale to UTF-8.

Реган.jp2

I cannot figure out the actual character for the first pair bytes, but the subsequent pairs appear to be spelling еган and if that's the case, it means it interprets UTF-8 as Latin1, which would point to an instance of std::string interpreted without setting the locale to UTF-8.

it gets better. that's from the CMake CI. The meson one properly prints Реган but still fails to open.

Alright so far the unit tests succeed on UCRT64.

MSVC is hilarious:

Actual: it throws Exiv2::Error with description "D:/a/exiv2/exiv2/unitTests/../test/data\\u0420\u0435\u0433\u0430\u043d.jp2: Failed to open the data source: No such file or directory (errno = 2)".

wonder what happened there.