What happened to wstring support in open() APIs?

[ribtoks]

Hi

I upgraded exiv2 to 0.28.0 just to find that std::wstring support has been nuked out in 7933ff4 but I could not find a matching PR, issue or at least a discussion that would explain why and what to do now.

std::string just does not work in Windows for non-latin1 paths for Exiv2::ImageFactory::open() (Visual Studio compiler).

What is the intended way to use it now?

These are all the old issues that reference the problem:

• https://dev.exiv2.org/issues/851
• Windows binaries with Unicode paths #679

[cgilles]

I can confirm that digiKam 8.1.0 pre-release based on Exiv2 0.28 is broken under windows to read metadata from images.

https://bugs.kde.org/show_bug.cgi?id=469372
https://bugs.kde.org/show_bug.cgi?id=470566

[neheb]

The intended way to use this is to set UTF-8 as the code page. See: https://learn.microsoft.com/en-us/windows/apps/design/globalizing/use-utf8-code-page

edit: oh I see. digiKam is compiled with MSVCRT instead of UCRT. I don't think that can work.

ping @piponazo

[cgilles]

No, digiKam is cross compiled under Windows using MXE (MinGW env - https://github.com/mxe/mxe) since 10 years now. The MSVC build is just another way to build the application in the KDE infra, it's not the official production release at all, because MSVC is unsafe,slow, and under-productive platform to compile huge code as digiKam, but it's another story...

Exiv2 team, please revert this commit to drop the UNICODE support under Windows. It's not at all a "dead" code... or explain well the work around. Note: I'm surprised to not see an unit-tests crying for the UTF16 long file path under Windows...

[neheb]

No, digiKam is cross compiled under Windows using MXE (MinGW env - https://github.com/mxe/mxe) since 10 years now. The MSVC build is just another way to build the application in the KDE infra, it's not the official production release at all, because MSVC is unsafe,slow, and under-productive platform to compile huge code as digiKam, but it's another story...

Exiv2 team, please revert this commit to drop the UNICODE support under Windows. It's not at all a "dead" code... or explain well the work around. Note: I'm surprised to not see an unit-tests crying for the UTF16 long file path under Windows...

Read what I said again. MSVCRT and UCRT

[maksim-petukhov]

@neheb @piponazo
What was the reason for deleting perfectly working code in 7933ff4? I'd like to advocate for reverting this deletion. The removed code was functional and served a purpose, so it doesn't seem like dead code. It's been quite useful for me, and I believe it could continue to benefit other users as well.

The intended way to use this is to set UTF-8 as the code page. See: https://learn.microsoft.com/en-us/windows/apps/design/globalizing/use-utf8-code-page

I see that the current recommendation is to "set UTF-8 as the code page". It is a bit reckless to ask users of Exiv2 to do that. There are some problems with this approach. First of all, this can perfectly work in tests but there is a reality when you have 150+ external libraries in your project and neither of them mentions that it works flawlessly with relatively new Microsoft invention. There are also loads of your code that relies on assumption "it is not UTF-8 there but something else" you have to check to make sure nothing is broken by forcing UTF-8 code page.

The "UTF-8 as the code page" feature is available only since Windows 1803. For Windows 10 operating systems prior to 1803 (10.0.17134.0), only static linking of UCRT is supported.. Windows 10 version prior 1803 are still supported in LTSC channel and static linking of UCRT is discouraged by Microsoft for security reasons. I don't see a nice way to use 0.28 across all Windows 10 versions.

I'm sharing these insights in the spirit of maintaining the functionality and usability of Exiv2 across a wide range of scenarios. I hope we can discuss this further and consider the potential benefits of restoring the code.

Thank you for your attention and consideration.

[neheb]

@maksim-petukhov more reading here: #2090

[maksim-petukhov]

@neheb I read it. It is still doesn't answer why @piponazo deleted code for EXIV2_ENABLE_WIN_UNICODE. exiv2.exe binary works with "set UTF-8 as the code page" - great. But as I'm trying to point out, there are big projects out there relying on exiv2 library that can't just "set UTF-8 as the code page" because this could break a lot of things.

[cgilles]

And digiKam also depend on Exiv2 in-deep, and we don't have this kind of problem with UTF8 under Windows when it's cross compiled under Linux MinGW. Please revert this code deletion, as the library is not suitable under Windows when it's cross compiled. I second also maksim-petukhov about the topic : "Why to drop working code tested since a vrey long time ?"

NOTE: recommend me to compile whole digiKam with MSVC under Windows is not a solution for digiKam at all...

Gilles Caulier

[kmilos]

NOTE: recommend me to compile whole digiKam with MSVC under Windows is not a solution for digiKam at all...

You have the MSYS2 UCRT64 option (the default now for MSYS2). I think Octave recently switched to UCRT builds also using MXE - @mmuetzel can you please advise the digiKam devs?

[kmilos]

AFAICT, it seems Fedora's MinGW cross toolchain can also target UCRT since F37: https://fedoraproject.org/wiki/Changes/F37MingwUCRT (e.g. https://src.fedoraproject.org/rpms/mingw-zlib/pull-request/3)

Please get in touch with any other distro not supporting and ask for UCRT targets (cf. Arch, Debian, don't know about Ubuntu...)