Re: [RFC] Do not treat '\' as escape character on MinGW Windows hosts

[Joel Brobecker <brobecker at adacore dot com>]

I didn't get much feedback on this. I thought it would have stirred
a little more controversy ;-). Since the feedback that we did get
went both ways, we'll just keep that an AdaCore-specific feature.

On Wed, Apr 21, 2010 at 04:33:54PM -0400, Joel Brobecker wrote:
> Hello,
> 
> With the MinGW debugger, it is currently not possible to use a path
> that follows the Windows convention.  For instance:
> 
>         (gdb) file c:\foo\bar.exe
>         c:foobar.exe: No such file or directory.
> 
> This is because the routine that parses arguments treats all backslashes
> as an escape character.  With a MinGW tool, this does not make sense
> when the argument is a path, since the canonical directory separator
> on Windows is a backslash...
> 
> What the user has to do, at this point, is escape every backslash,
> which can be quite painful when the path starts getting longer...
> 
>         (gdb) file c:\\foo\\bar.exe
>         Reading symbols from c:\foo\bar.exe...done.
> 
> What we have done, at AdaCore, is disabling the special nature of
> the backslash character when the host is MinGW (we left cygwin alone,
> since cygwin users most likely expect a unix-y type of behavior).
> It's a incompatible change, and as you'll see with the attached patch,
> it changes the behavior for all arguments, not just path names.
> Polling the few Windows users we have a AdaCore, they all seem to agree
> that they did not expect backslash to be a special character in any
> context, so the change seemed right to them.
> 
> Before officially submitting this patch for inclusion (including
> formal testing), I wanted to see what the feeling towards this sort
> of change was...
> 
> Note that GDB uses a wrapper around this routine, mostly to add
> an out-of-memory exception raised when the routine returns null.
> One possible way of achieving the same result, while limiting the
> change to GDB alone, would be to modify the gdb_buildargv routine
> to escape all backslashes before calling libiberty's buildargv.
> But I think that other tools should also consider this change as
> beneficial.
> 
> -- 
> Joel

> diff --git a/libiberty/argv.c b/libiberty/argv.c
> index 3084248..b5cf71f 100644
> --- a/libiberty/argv.c
> +++ b/libiberty/argv.c
> @@ -177,7 +177,8 @@ returned, as appropriate.
>  
>  */
>  
> -char **buildargv (const char *input)
> +char **
> +buildargv (const char *input)
>  {
>    char *arg;
>    char *copybuf;
> @@ -189,6 +190,15 @@ char **buildargv (const char *input)
>    char **argv = NULL;
>    char **nargv;
>  
> +/* On Windows hosts, the backslash character should not be treated
> +   as an escape character.  Define a constant bs_is_escape whose value
> +   is non-zero when the backslash is an escape character.  */
> +#if !defined (__MINGW32__)
> +  const int bs_is_escape = 1;
> +#else
> +  const int bs_is_escape = 0;
> +#endif
> +
>    if (input != NULL)
>      {
>        copybuf = (char *) alloca (strlen (input) + 1);
> @@ -234,12 +244,12 @@ char **buildargv (const char *input)
>  		}
>  	      else
>  		{
> -		  if (bsquote)
> +		  if (bs_is_escape && bsquote)
>  		    {
>  		      bsquote = 0;
>  		      *arg++ = *input;
>  		    }
> -		  else if (*input == '\\')
> +		  else if (bs_is_escape && (*input == '\\'))
>  		    {
>  		      bsquote = 1;
>  		    }


-- 
Joel