On Windows, don't strip the Content-Type field from .po files during their compilation.
GNU libintl, by default, converts the l10n strings into the locale encoding, while Subversion requires UTF-8. This conversion can be suppressed by calling bind_textdomain_codeset, but certain old versions of libintl don't have that, so the Unix build system checks for the existence of that function, and if it's not present, strips the Content-Type header from the .po files (which prevents encoding conversion, as well, but makes msgfmt complain).
When building on Windows, this stripping is done unconditionally, but is completely unnecessary:
- On Windows, we only support linking with svn-win32-libintl, which is hacked to disable all encoding conversions. - Even if someone links with with his own version of libintl, it's a safe bet that it will be new enough to support bind_textdomain_codeset, so we can just call that.
This patch removes Content-Type stripping on Windows, which gets rid of msgfmt warnings, as well as simplifies the build system.
* build/generator/build_locale.ezt: Remove the strip-po-charset.py invocation.
* build/generator/gen_win.py: (POFile.__init__): don't store the .spo file name.
* build/strip-po-charset.py: Delete.
* subversion/libsvn_subr/nls.c: (svn_nls_init): Move the bind_textdomain_codeset call out of the #ifdef WIN32 block, so it's executed on Windows, as well.
* subversion/svn_private_config.hw: Indicate that bind_textdomain_codeset is available if NLS is enabled.