Re: UNC path syntax

[Jim Balter <jqb at netcom dot com>]

Ed Huott wrote:
> 
> In message <34035660.1323@elementalsoftware.com>, Tom Yonkman writes:
> >     Just started using the gnu-win32 package and wonder if ls and friends
> >     can handle the UNC path syntax in some way? This would allow access to
> >     machines in my "Network Neighborhood", as the explorer so quaintly
> >     puts it.
> >
> 
> No.  Nothing built with cygwin32 can deal with UNC path names.  The
> cygwin32 approach requires that you map a drive to any remote (network
> neighborhood) volumes.  Inside the bash shell, a mapped network drive
> such as N: can then be referenced using the path "//N".  This syntax
> can, unfortunately, collide with the most logical choice for
> referencing a network share, which would be something like:
> "//ComputerName/SharedDrive".
> 
> There is a unix-like volume mounting scheme that lets you acces a
> drive as a sub-directory (i.e. N: ==> /DriveN) something like the old
> DOS "join" command, but this works *only* for apps built with
> cygwin32.
> 
> Lack of UNC support can be a real drawback if you're dealing in an
> environment that uses a mix of both native (Windows) and cygwin32
> apps.

You can refer to UNC paths with, e.g., "\\xxx\y", but not
"//xxx/y", which gets turned into "c:\xxx\y".  You can use names like
"\\.\a:", but not "\\.\tape0", which gets turned into "c:\tape0".
And you cannot use UNCs in mounts; they get a \ appended that win32
can't handle.  I've attached a patch that fixes the latter two.
Among other things, this will allow you to read a local tape drive
on WinNT via tar, dd, etc. by using /dev/st0 if you fix the bogus mount
table entries (the proper name is "\\.\tape0", not "\\.\tape0:"):

umount /dev/st0
mount '\\.\tape0' /dev/st0
umount /dev/st1
mount '\\.\tape1' /dev/st1

Due to other bugs, this still does not allow you to write to local tape
via cygwin, however.

--
<J Q B>
*** path.cc.orig	Sun Mar 23 02:15:12 1997
--- path.cc	Sun Mar 23 03:15:33 1997
***************
*** 208,214 ****
  
    /* The rule is :'s can't appear in [our] posix path names so this is a safe
       test; if ':' is present it already be in win32 form.  */
!   if (strchr (src_path, ':') != NULL)
      {
        debug_printf ("conv_to_win32_path: %s already win32\n", src_path);
        if (win32_path != NULL)
--- 208,216 ----
  
    /* The rule is :'s can't appear in [our] posix path names so this is a safe
       test; if ':' is present it already be in win32 form.  */
!   // JQB: hack in test for \\ as well
!   if (strchr (src_path, ':') != NULL ||
!       src_path[0] == '\\' && src_path[1] == '\\')
      {
        debug_printf ("conv_to_win32_path: %s already win32\n", src_path);
        if (win32_path != NULL)
***************
*** 305,313 ****
  	{
  	  int j = mount[i].devicelen;
  	  memcpy (p, mount[i].device, j);
! 	  if (pathbuf[mount[i].pathlen] != '/')
  	    p[j++] = '\\';
! 	  strcpy (p + j, pathbuf + mount[i].pathlen);
  	  backslashify (p, p, trailing_slash_p);
  	}
        if (win32_path != NULL && ! got_rel_p && win32_path != p)
--- 307,321 ----
  	{
  	  int j = mount[i].devicelen;
  	  memcpy (p, mount[i].device, j);
! 
! 	  // JQB: add test for empty tail, so, for instance,
! 	  // \\.\tape0 doesn't turn into \\.\tape0\,
! 	  // which win32 doesn't like.  Still turn a: into a:\ though.
! 	  char *tail = pathbuf + mount[i].pathlen;
! 	  if ((*tail || tail[-1] == ':') && *tail != '/')
  	    p[j++] = '\\';
! 	  strcpy (p + j, tail);
! 
  	  backslashify (p, p, trailing_slash_p);
  	}
        if (win32_path != NULL && ! got_rel_p && win32_path != p)