[adding autoconf and gnulib; replies can drop cygwin]
On 08/23/2010 10:37 PM, Yaakov (Cygwin/X) wrote:
On Tue, 2010-08-24 at 01:31 +0200, Angelo Graziosi wrote:
Trying to build Emacs-trunk on Cygwin with GCC>= 4.5.1, I have found a
problem if also the package 'libelf0' is installed.
[snip]
and the following test in 'configure' script is true:
[snip]
i.e. SVR4 is _defined_. But... 'src/xfaces.c' includes indirectly via
'/usr/include/X11/Xos.h' the header '/usr/include/X11/Xarch.h', which has:
[snip]
i.e. finding SVR4 defined, it searches for 'byteorder.h' in
'/usr/include/sys' and not in '/usr/include/asm', where it lives (*as on
GNU/Linux*[1]), and this leads to the above failure.
This is coming from autoconf's AC_FUNC_GETLOADAVG, which tries to find
several different getloadavg(3) implementations (which Cygwin doesn't
provide), otherwise it points to an AC_LIBOBJ from gnulib. On Solaris,
the gnulib getloadavg uses kvm_open(3) and friends, and libkvm requires
libelf, hence the check for the latter.
According to gnulib, AC_FUNC_GETLOADAVG is obsolescent (in part because
it uses AC_LIBOBJ in a manner that is not easily extensible by gnulib).
If emacs is using modern gnulib, then it is _not_ using
AC_FUNC_GETLOADAVG from autoconf, but rather the more up-to-date
gl_GETLOADAVG from the gnulib module getloadavg.
Either way:
AFAICS the proper solution is, in _AC_LIBOBJ_GETLOADAVG:
if test $ac_have_func = no&& test "$ac_cv_lib_elf_elf_begin" = yes;
then
ac_have_func=yes
AC_DEFINE(SVR4, 1, [Define to 1 on System V Release 4.])
fi
to add<< && test "$ac_cv_lib_kvm_kvm_open" = yes>> to the conditional.
Yes, autoconf's AC_FUNC_GETLOADAVG should be updated in this manner, and
that should also be reflected into gnulib's gl_GETLOADAVG. Thanks for
the analysis.
--
Eric Blake ebl...@redhat.com +1-801-349-2682
Libvirt virtualization library http://libvirt.org