2017-06-16 22:10 GMT+02:00 Artyom Beilis via Boost :

On Fri, Jun 16, 2017 at 6:12 PM, Andrzej Krzemienski via Boost wrote:

...

Hi Everyone, I admit I am not quite familiar with the problem, but I understand that as one of the features, nowide offers a replacement for std::fstream that can be constructed with its string types. At the same time we have boost::filesystem that offers its own replacement for std::fstream that can be constructed with filesystem::path. Now, if I want to use `filesystem::path`s in my program (to be able to tell just any string from a filesystem path), can I still use the benefits of `nowide` library?

Yes of course. There is an integration between nowide and filesystem to make sure it considers narrow API to be UTF-8.

Also note the nowide::fstream works on MinGW as well as filesystem.fstream calls std::fstream and only MSVC version has open(wchar_t const *).

...

Also, in the docs for nowide::ifstream, we read, "Same as std::basic_ifstream<char> but accepts UTF-8 strings under Windows." What about other systems? What does it accept on Linux? ascii?

Under Linux it accepts "char *" in whatever encoding it is considered. See: http://cppcms.com/files/nowide/html/index.html#qna

...

In documentation for `nowide::args`, we read, "args is a class that fixes standard main() function arguments and changes them to UTF-8 under Microsoft Windows." Does it write to the input strings in-place? is it even legal in C++?

It replaces values of argc and argv and points them to other location not modifying the original values.

Ok. It makes sense :)

...

It "fixes", which implies that otherwise the args are "broken". How are args in function main() broken? (other than not being UTF-8)?

That main(argc,argv) receives parameters converted from native UTF-8 internal API to current locale's codepage - generally not being able to represent the all the required charset (since Windows does not support UTF-8 as native locale)

But given that what main() receives is already broken (Windos already could not handle a name containing letters from two code pages), how can you recover from this loss of information? Regards, &rzej;

Take a look to the code :-) I use WinAPI to retrieve the original Utf-16 args. I don't relay on the original strings.

This is interesting. You can retreive more than what was given! I think you should document this, not only in the code it-self. Frédéric

Actually it is cppcms.com/files/nowide/html/classboost_1_1nowide_1_1args.html

sorry, yes, everything is there. Frédéric

...

It "fixes", which implies that otherwise the args are "broken". How are args in function main() broken? (other than not being UTF-8)?

That main(argc,argv) receives parameters converted from native UTF-8 internal API to current locale's codepage - generally not being able to represent the all the required charset (since Windows does not support UTF-8 as native locale)

This is incorrect. You have been able to set the Windows console to UTF-8 for many years. Just issue `chcp 65001`, your console is now in UTF-8 and UTF-8 strings will present to argv. Indeed it is possible to set UTF-8 consoles globally as the default, but lots of stuff hard assumes Latin1 input and gets very upset if it sees UTF-8. In particular, MSVCRT, though maybe the VS2015 rewrite of MSVCRT has fixed that. I do remember .NET programs ran great with UTF-8 input though, as do NT kernel programs. The blocker is MSVCRT. Niall -- ned Productions Limited Consulting http://www.nedproductions.biz/ http://ie.linkedin.com/in/nialldouglas/