While I do have this problem in Prefix on AIX, this might be a problem for multilib and cross compilation as well (like bug#510248), finding wrong CC/CFLAGS. There's a recent upstream commit already: https://gforge.inria.fr/scm/viewvc.php/trunk/m4/mpc.m4?root=mpc&r1=1451&r2=1450&pathrev=1451 Additionally, the ebuild will need: tc-export CPP Thanks!
In short this can boil down to: --- mpc-1.0.2.ebuild +++ mpc-1.0.2.ebuild @@ -20,10 +20,12 @@ RDEPEND="${DEPEND}" src_prepare() { + sed -i -e 's/for cpp in/& "$CPP"/' configure || die # 516104 elibtoolize # for FreeMiNT, bug #347317 } src_configure() { + tc-export CPP econf $(use_enable static-libs static) }
i don't think forcing people to set CPP by hand is the right answer (using AC_PROG_CPP would be best, but i guess they don't want to deal with that) $CXX -E would be better i think
Agreed - why $CXX -E and not $CC -E ? However, I've just followed upstream here...
(In reply to Michael Haubenwallner from comment #3) for the same reason upstream isn't using the sane answer in the first place -- they're trying to not require a C compiler, only a C++ compiler.
(In reply to SpanKY from comment #4) Doesn't feel like they require a C++ compiler at all. Instead, looks like their idea is to pull CC/CFLAGS from gmp.h first, to guide AC_PROG_CC into the "right" direction... Maybe we just should tc-export CC (besides CFLAGS) to not use their convenience?
Is it still an issue with =dev-libs/mpc-1.1.0-r1? Looks like upstream bug is fixed.
(In reply to Sergei Trofimovich from comment #6) > Is it still an issue with =dev-libs/mpc-1.1.0-r1? Looks like upstream bug is > fixed. No longer a problem for me now.
Honestly, I've almost given up on AIX and don't have a working AIX Prefix any more, and I'm not into multilib or cross compilation, so this bug indeed isn't relevant to me any more. However, when looking at their current revision of the MPC_GMP_CC_CFLAGS macro, there still might be demand to properly set up CPP, because there is no $CC nor $CXX in their cpp candidate list: https://scm.gforge.inria.fr/anonscm/gitweb?p=mpc/mpc.git;a=blob;f=m4/mpc.m4;h=f3c75540b814980c43896922338105bf3289007e;hb=HEAD#l126
I think for multilib ebuilds it was worked around with: """ commit c65f6083898968b5442eeb66afd4298a3b4264e2 Author: Michał Górny <mgorny@gentoo.org> Date: Fri Jun 17 09:30:26 2016 +0200 toolchain-funcs.eclass: Assume CPP="$(tc-getCC) -E" when unset, #582822 """