Need your help with MinGW Issue 17: --color options don't work (produce garbage)

Here are the facts:

'git branch --color' produces garbage:
$ git branch --color
  devel←[m
  dima←[m
  dmitryk←[m
* ←[32mmaster←[m
  mob←[m
  next←[m

'git branch --color | cat' produces expected colored output.

I've traced it down to printf statement in gdb and it sends the right esc-sequence. Where should I look next?

(little bit) more info here: http://code.google.com/p/msysgit/issues/detail?id=17

On 15/08/07, Dmitry Kakurin wrote:

Windows doesn't recognise the *nix printf colour codes.

Piping through cat will be going through cygwin/mingw emulation, translating the colour codes to the correct API calls.

You need to call the SetConsoleTextAttribute Win32 API. For example:

#ifdef defined(WIN32) || defined(WIN64)

typedef WORD color_t;

color_t red = FOREGROUND_INTENSITY | FOREGROUND_RED; color_t green = FOREGROUND_INTENSITY | FOREGROUND_GREEN; color_t blue = FOREGROUND_INTENSITY | FOREGROUND_BLUE;

color_t white = red | green | blue;

void set_color( color_t color )
{
    SetConsoleTextAttribute(GetStdHandle(STD_OUTPUT_HANDLE), color );
}

#else

typedef const char * color_t;

color_t red = ...; ...

void set_color( color_t color ){ printf( color ); }

#endif

That way, you can do things like:
    set_color( red );
    printf( ... );
    set_color( blue );

This is not as pretty as the existing codebase, so another possibility would be to create wrappers around the console output functions (i.e. printf) and call SetConsoleTextAttribute there. This way, you can restore the old colour when a restore settings sequence is intercepted. It is also possible to reuse the GetStdHandle return value.

NOTE: There isn't a GetConsoleTextAttribute in the Windows API, but
Google found this:

#if ( (defined(WIN32) || defined(_WINDOWS)) && !defined(__CYGWIN__) ) && defined(_CONSOLE)

static WORD GetConsoleTextAttribute(HANDLE Console)
{
    CONSOLE_SCREEN_BUFFER_INFO ConsoleInfo;
    GetConsoleScreenBufferInfo(Console, &ConsoleInfo);
    return ConsoleInfo.wAttributes;
}

#endif

HTH,
- Reece

On 8/15/07, Reece Dunn <msclrhd@googlemail.com> wrote:

That's my question. If there is a way to build cat.exe to do this kind of emulation under MinGW then I should be able to do the same for git.exe. I hope I just need to #define something while building Git. But what is it?

On 15/08/07, Dmitry Kakurin <dmitry.kakurin@gmail.com> wrote:

You will need to implement cat (or something similar) on MinGW that will process the *nix colour codes and then pass that to SetConsoleTextAttributes. For example:

    int ch = 0;
    while(( ch = getch()) != EOF )
    {
        if( /*ch is part of a colour sequence*/ )
        {
            SetConsoleTextAttribute( GetStdHandle(...),
win_color_from_color_sequence(...));
        }
        else putc( ch );
    }

I don't believe that MinGW has got this working for cat, that is why one that supports colours on Windows needs to be written.

- Reece

On 8/15/07, Reece Dunn <msclrhd@googlemail.com> wrote:

MSys' cat does support colors even when I run it from cmd.exe. I've looked at sources for cat on mingw.sourceforge.com and (as expected) there is no code to handle esc-sequences. That would be insane - too many utilities like ls, diff, cat, less etc. do that. So there is obviously a library for this. I've tried to "parse" makefiles for cat, but they are so huge and I know about gcc/make on Unix so little that I was unable to find the secret sauce. I've also asked this question on MinGW list so I'm waiting for reply.

Hi,

On Wed, 15 Aug 2007, Reece Dunn wrote:

Hmm. Somehow I doubt that this hack works _outside_ of the Windows console. I.e. if you call git in rxvt, it will fail, if you ssh into Windows, it will fail.

Ciao, Dscho

Johannes Schindelin wrote:

I'd say that since Windows doesn't have the concept of terminfo or terminal types, and since SSH is mostly only supported through cygwin-style "hacks", (or F-Secure, but I have no experience with that) that we should not be _too_ concerned about that.

Users that *do* need to use rxvt or SSH should simply disable the color mode, or alternatively, use the cygwin version. Color, while useful, is hardly critical functionality.

My R0.02.

Regards,

Rogan

Rogan Dawes <lists@dawes.za.net> writes:

Heh, that almost suggests that the native Windows command.com support can disable the color without upsetting anybody ;-)

Junio C Hamano wrote:

If that is the easiest way, yes.

I *do* think that trying to get color working on the DOS cmdline is a worthy goal. I just don't think it is worth overcomplicating the issue and trying to handle rare corner cases such as someone SSHing into a Windows box, or running CMD inside rxvt.

One approach that might be more compatible with the existing escape code approach is to define the appropriate ANSI color escape sequences. Ah, a bit more googling shows that cmd.exe doesn't handle ANSI sequences in Windows NT and later.

Regards,

Rogan

Junio C Hamano wrote:

Following my other comment in this regard, another reason that colour in the cmd line tools is less important, is because my impression of the intent of the Windows port (end goal, that is), is that the user will be able to use a Windows GUI front-end to do whatever it is they want to.

And so the GUI front-end will be responsible for implementing the color anyway, most likely through pattern matching and a state machine.

Regards,

Rogan