#478 Win32: unicode support for files with public API for externs

[Hans-Christoph Steiner]

Right now, Pd on Windows does really badly if there are any non-ASCII characters in the path or filename. It makes it freeze for a while, and sometimes crashes it. These patches fix that.

Pd and Tcl/Tk is UTF-8 internally, and UNIXes all use UTF-8 for filenames
and paths. Windows uses UCS-2 everywhere, which is a 16-bit format. The
only place this affects Pd is reading and writing filenames, and printing
to the console. The POSIX-style functions open() and fopen() exist on
Windows, but only work for ASCII filenames. To support Unicode filenames,
we have to convert the UTF-8 to UCS-2, then use Win32-specific functions.

Since any external that opens files will also be affected the same
way, this patch provides a public API: sys_open()/sys_close(), and
sys_fopen()/sys_fclose(). For non-Win32 platforms, they are just
names that point to the normal POSIX versions. On Win32, they are
special functions to handle UTF-8 to UCS-2 conversion.

I have built and run this on Windows XP, Mac OS X 10.6, and Debian/squeeze amd64. These patches are also included in Pd-extended 0.43.4.

[Hans-Christoph Steiner]

patch to apply with 'git am'

[Hans-Christoph Steiner]

patch to apply with 'git am'

[Hans-Christoph Steiner]

patch to apply with 'git am'

[Hans-Christoph Steiner]

patch to apply with 'git am'

[Hans-Christoph Steiner]

I added a fifth patch. Microsoft says that the POSIX close() was deprecated in 2005, and to use
their ISO C++ _close() instead.

http://msdn.microsoft.com/en-US/library/ms235443\(v=vs.80).aspx
http://msdn.microsoft.com/en-US/library/5fzwd5ss\(v=vs.80).aspx

[Hans-Christoph Steiner]

patch to apply with 'git am'

[Hans-Christoph Steiner]

and a couple more for good measure ;)

[Hans-Christoph Steiner]

patch to apply with 'git am'

[Hans-Christoph Steiner]

These are all the externals in Pd-extended that need changes to support Unicode filenames:

http://pure-data.svn.sourceforge.net/viewvc/pure-data?view=revision&revision=16669
http://pure-data.svn.sourceforge.net/viewvc/pure-data?view=revision&revision=16670
http://pure-data.svn.sourceforge.net/viewvc/pure-data?view=revision&revision=16671

Basically, you just need to change open() to sys_open() and fopen() to sys_fopen(). You can remove sys_bashfilename() since its included in those new open functions when needed.

[Miller Puckette]

For some reason I can't apply the first patch (I get:
error: patch failed: src/s_utf8.c:26
error: src/s_utf8.c: patch does not apply
error: patch failed: src/s_utf8.h:1
error: src/s_utf8.h: patch does not apply
--- even though my eyeballs can't see any reason that would fail.

Anyway, I'd like not to have m_pd.h be in the business of pulling in other include files
(except as needed to define its own data structures) as I think that's a threat to future
portability - can this be fixed to leave m_pd.h, d_soundfile.h, etc., alone and just "do the
necessary" to the UTF-8 code?

thanks
Miller

[Hans-Christoph Steiner]

I agree that its good to avoid putting #includes in m_pd.h stdint.h is a stable and widespread as stddef.h, which has been part of m_pd.h for a long time. I put '#include <stdint.h>" because that's the only header that was already included everywhere that needs types defined in stdint.h.

Another option would be to put the int32_t, etc. definitions in s_stuff.h or some other header, and then add that header everywhere its needed.

As for the patches not applying, that's because you accepted 3 patches from Marvin Humpreys that removes lots of code from s_utf8.* and that conflicts with these patches.

[Miller Puckette]

OK -- I'll see if I can unwind the other patch and apply this one - if that works I'll go and fix the
stdint stuff somehow that I can live with :)

[Miller Puckette]

Got 'em & pushed to repo... will re-apply the other patches later.