[ClusterLabs Developers] [PATCH] Add support for GNU/Hurd

[Svante Signell]

On Wed, 2016-02-24 at 09:40 -0600, Ken Gaillot wrote:
> On 02/24/2016 06:15 AM, Jan Pokorný wrote:
> > 
> I added this comment to the pull request:
> 
> * The leading null byte in sun_path (the "+ 1" when using it) indicates
> an abstract socket (see unix(7)). If Hurd doesn't support those, it'll
> need a separate block -- we shouldn't change the behavior for Linux/Cygwin.

Now I see why you use the + 1 in sun_path, thanks. However, the second patch was
derived under GNU/Linux, where the sun_path was found to be NULL, as by the
comment. It FTBFS without the unlink statement with the second patch. The
unpatched code in Debian version 0.17.2.real-4 FTBFS with gcc-4.9.1-16, as
written in Debian bug #803766. How is the abstract socket selected in favor of
the regular socket in GNU/Linux?

> * I don't think the unlink() will make sense with abstract sockets.

Same here, the unlink was is needed with the latest gcc: 5.3.1-8 :( Otherwise
ipc.test fails.

> * Not necessary with this request, but it would be nice if we could have
> ./configure define semantic macros like HAVE_ABSTRACT_SOCKET or
> something. Or at least #define them ourselves so we don't have to repeat
> "defined(...) || ..." all the time.

That would definitely be needed, especially if abstract sockets are not used in
GNU/Linux, as I probably experienced, see above.