Python: don't build curses and readline modules for host

[escalade]

#301 broke compilation of the curses module for some. We need netbsd-curses:host and a patch to fix this:

/root/LibreELEC.tv/build.LibreELEC-Generic.x86_64-8.0-devel/Python-2.7.11/Modules/_cursesmodule.c: In function 'PyCursesWindow_EchoChar': /root/LibreELEC.tv/build.LibreELEC-Generic.x86_64-8.0-devel/Python-2.7.11/Modules/_cursesmodule.c:808:18: error: dereferencing pointer to incomplete type 'WINDOW {aka struct __window}' if (self->win->_flags & _ISPAD)

[lrusak]

You don't need to unpack

just do this

pre_build_target() {
  mkdir -p $PKG_BUILD/.$TARGET_NAME
  cp -RP $PKG_BUILD/* $PKG_BUILD/.$TARGET_NAME
}

pre_build_host() {
  mkdir -p $PKG_BUILD/.$HOST_NAME
  cp -RP $PKG_BUILD/* $PKG_BUILD/.$HOST_NAME
}

[escalade]

OK, changed.

[escalade]

Removed -fPIC from Python, as it gets added where appropriate already.

[TheUlpio]

@escalade your latest patchset builds flawlessly, no issues with mc, so i'm happy with it. Thank you

[escalade]

Thanks for the feedback, I guess this is good to go then. Builds fine both for Generic and RPi.

[escalade]

There's one issue left (Assumed value of MB_LEN_MAX wrong): http://pastebin.com/tJcF2TJU

This apparently occurs both on Gentoo and Ubuntu. I can't reproduce on Arch, so this is probably yet another case where the system headers mess things up.

Relevant part of stdlib.h:

{
  /* We would have to include <limits.h> to get a definition of MB_LEN_MAX.
     But this would only disturb the namespace.  So we define our own
     version here.  */
#define __STDLIB_MB_LEN_MAX     16
#if defined MB_LEN_MAX && MB_LEN_MAX != __STDLIB_MB_LEN_MAX
# error "Assumed value of MB_LEN_MAX wrong"
#endif
  if (__bos (__s) != (size_t) -1 && __STDLIB_MB_LEN_MAX > __bos (__s))
    return __wctomb_chk (__s, __wchar, __bos (__s));
  return __wctomb_alias (__s, __wchar);
}

[escalade]

The above issue where MB_LEN_MAX is defined somewhere to something else than 16 is pretty much the same issue that has been worked around in the patch attached to this PR, where WINDOW_HAS_FLAGS gets defined somewhere, when it shouldn't. I'm guessing this is another case of the system headers being used instead of those in the toolchain.

Do we even need to build the readline/ncurses modules for the host? If not, a simple fix might be to keep it disabled only for the host build.

[escalade]

As compiling the modules for the host build is problematic, it's better to simply disable them for host. I've tested and it works well. Commits changed, please revise and test.

[lrusak]

👍 looks good