Re: Install public headers in the standard path?

[Anthony Green <green at moxielogic dot com>]

Thanks for your patch.

If you can believe it, installing certain headers under /usr/lib was a
thing back in the mid/late 90s (gnome, or gtk, or similar did this as
well).  libffi headers are modified at configure time, and the
thinking back then was that the contents of /usr/include should not
vary based on configuration changes (or even on per-host basis).
Also, using pkg-config made it easy to install multiple versions of
libffi as the ABI hadn't settled down yet.

A few years ago I sat down to rewrite the headers to depend solely on
compiler-defined macros, but got lost in the maze of Power
alternatives.  Every other port was fairly straight forward - but
Power was a mess.

I'm OK with moving the headers to /usr/include now, although it would
have been nice to remove the configure-time dependencies from them
first.

AG



On Tue, Nov 15, 2016 at 7:10 AM, Chi-Hsuan Yen <yan12125@gmail.com> wrote:
> Hello libffi enthusiasts,
>
> Currently libffi public headers are installed in
> $PREFIX/lib/libffi-3.x.y/include. The only reliable approach to get
> the include path is via pkg-config. There are some disadvantages:
>
> 1. pkg-config is tricky in terms of cross-compilation. To cross-build
> libffi clients, PKG_CONFIG_LIBDIR, PKG_CONFIG_SYSROOT_DIR,
> PKG_CONFIG_ALLOW_SYSTEM_{CFLAGS,LIBS} may be involved to make it
> working
>
> 2. pkg-config does not come with macOS. Recently CPython is seeking
> for chances to drop bundled libffi copy. [1] However, CPython core
> developers are against including additional tools other than
> XCode/Command Line Tools in CPython build time requirements
>
> There are some more discussions here:
> https://github.com/libffi/libffi/issues/258
>
> As a result, I suggest to move header files to conventional
> $PREFIX/include, so that libffi clients can have simpler build
> systems.
>
> Here's my implementation: https://github.com/libffi/libffi/pull/288
>
> Any ideas?
>
> Best,
>
> Yen, Chi-Hsuan
>
> [1] http://bugs.python.org/issue28491