Re: [rdiff-backup-users] Building rdiff-backup on Windows

[Josh Nisly]

Andrew, the whole point of my patches is to make so that rdiff-backup 
builds and runs without cygwin, i.e. natively. Knowing this might make 
some of the below make more sense.

Andrew Ferguson wrote:
> > Thanks,
> > Josh Nisly
> > --- setup.py    2008-01-03 09:36:49.000000000 -0600
> > +++ setup.py    2008-04-02 17:58:36.984375000 -0500
> > @@ -40,6 +40,9 @@
> >             libdir_list = [os.path.join(LIBRSYNC_DIR, 'lib')]
> >         if '-lrsync' in LIBS:
> >             libname = []
> > +elif os.name == 'nt':
> > +    incdir_list = ['.']
> > +    lflags_arg = ['/NODEFAULTLIB:libcmt.lib']
> >
> > setup(name="rdiff-backup",
> >       version=version_string,
>
>
> Can you explain why this patch is necessary for Windows? When I last 
> built rdiff-backup under Cygwin/windowsXP (July/August timeframe), it 
> was not necessary to do that. Have things changed such that librsync 
> can only be built in the current directory? Is the libcmt.lib piece a 
> Vista thing, perhaps?
>
> Personally, I don't commit patches that are only to make it build in 
> an individual user's environment (eg, I didn't commit logic to handle 
> librsync installed by the Fink package management system, etc.), so I 
> just wanted to check.
Under a normal non-cygwin Windows build, there is no place to look for 
the librsync library, so you have to specify a location somehwere. It 
seemed that looking in the current directory was the most generic place. 
Another option would be allowing the --librsync-dir option to work on nt 
platforms (we only support this flag on posix platforms.)

The libcmt.lib is a Microsoft compiler thing. Again, it could be made a 
flag, although it will always have to be that option, because of the 
mess that Microsoft has made of their libraries.
> > --- cmodule.c    2008-01-03 09:36:48.000000000 -0600
> > +++ cmodule.c    2008-04-01 20:08:15.484375000 -0500
> > @@ -24,7 +24,9 @@
> > #include <Python.h>
> > #include <sys/types.h>
> > #include <sys/stat.h>
> > +#if !defined(MS_WIN64) && !defined(MS_WIN32)
> > #include <unistd.h>
> > +#endif
> > #include <errno.h>
>
> Looks good, but I don't have a system to test on. Again, I don't 
> remember the need to comment that out when I last built it. Do you not 
> have Cygwin installed, per chance? I believe you should have a 
> unistd.h file....
I believe that unistd.h is a unix or posix library. At any rate, windows 
doesn't supply it.
> > @@ -46,6 +48,7 @@
> > /* This code taken from Python's posixmodule.c */
> > #undef STAT
> > #if defined(MS_WIN64) || defined(MS_WIN32)
> > +#  define lstat _stati64
> > #    define STAT _stati64
> > #    define FSTAT _fstati64
> > #    define STRUCT_STAT struct _stati64
> > @@ -69,6 +72,16 @@
> > #define S_ISFIFO(mode)        (((mode) & S_IFMT) == S_IFIFO)
> > #endif
>
> Ah, good catch, but what we should do is change the calls to 'lstat' 
> in the body of cmodule.c to be calls to LSTAT and then make that block:
Ok, I'll update the patch. Thanks for the feedback.
> #undef STAT
> #undef LSTAT
> #if defined(MS_WIN64) || defined(MS_WIN32)
> #    define LSTAT _stati64
> #    define STAT _stati64
> #    define FSTAT _fstati64
> #    define STRUCT_STAT struct _stati64
> #else
> #    define LSTAT lstat
> #    define STAT stat
> #    define FSTAT fstat
> #    define STRUCT_STAT struct stat
> #endif
>
> (not that STAT and FSTAT are used in cmodule.c any more, but at least 
> it keeps the correct mappings there for the future).
>
> > +#if defined(MS_WIN64) || defined(MS_WIN32)
> > +#define S_ISSOCK(mode) (0)
> > +#define S_ISFIFO(mode) (0)
> > +#define S_ISLNK(mode) (0)
> > +#define S_ISCHR(mode) (0)
> > +#define S_ISBLK(mode) (0)
> > +#define sync _flushall
> > +#endif
> > +
>
> Do these types not exist on Windows? (socket, fifo, symlink, character 
> device, block device) ?
No, they do not exist. On Windows, mode only specifies whether a path is 
a file or directory, plus (sometimes) some permission information. See 
http://msdn2.microsoft.com/en-us/library/14h5k7ff(VS.80).aspx.
> The sync() call in my_sync() should operate like the LSTAT/STAT/FSTAT 
> calls ... it should be in caps, SYNC(), and then have the definition 
> of SYNC in the LSTAT, etc. block above.
I'll change this too.
> Andrew

JoshN