While I hit the problem on a uclibc system, the issue is in numpy and I've opened a bug upstream. The problem is that their commit c15f7747 did a bunch of code cleanups that moved #ifndef HAVE_CEXP@C@ ... #endif in numpy/core/src/npymath/npy_math_complex.c.src to cover a function definition that's called outside the #ifndef. The problem is triggered on any systems where HAVE_CEXP@C@ is not defined (for some substituion of @C@). While the upstream bug is being fixed, can we have an epatch_user added to dev-python/numpy-1.10.4 which is currently being stabilized so that it doesn't cause havoc with uclibc systems. I have a temporary workaround.
commit ac9ac2fafdadacecb793e0087e5ef82ea7247fc5 Author: Justin Lecher <jlec@gentoo.org> Date: Thu Feb 4 11:09:36 2016 +0100 dev-python/numpy: Add epatch_user() call Gentoo-Bug: https://bugs.gentoo.org/show_bug.cgi?id=573824 Package-Manager: portage-2.2.27 Signed-off-by: Justin Lecher <jlec@gentoo.org> https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=ac9ac2fafdadacecb793e0087e5ef82ea7247fc5
(In reply to Justin Lecher from comment #1) > commit ac9ac2fafdadacecb793e0087e5ef82ea7247fc5 > Author: Justin Lecher <jlec@gentoo.org> > Date: Thu Feb 4 11:09:36 2016 +0100 > > dev-python/numpy: Add epatch_user() call > > Gentoo-Bug: https://bugs.gentoo.org/show_bug.cgi?id=573824 > > Package-Manager: portage-2.2.27 > Signed-off-by: Justin Lecher <jlec@gentoo.org> > > > https://gitweb.gentoo.org/repo/gentoo.git/commit/ > ?id=ac9ac2fafdadacecb793e0087e5ef82ea7247fc5 Upstream looks like they have a huge backlog of bugs. I'm pretty sure of my analysis, but the correct fix may not be easy. I don't think upstream is going to get around to this or any bug any time soon. Carrying epatch_user() forward might be wise for this package.
8 months later this is still not addressed upstream, at least the bug ticket is still open and a new uclibc-ng hardened system still breaks with numpy. Can we include a patch for this to fix it until upstream has fixed the bug?
(In reply to N. "ng0" Gillmann from comment #3) > 8 months later this is still not addressed upstream, at least the bug ticket > is still open and a new uclibc-ng hardened system still breaks with numpy. > > Can we include a patch for this to fix it until upstream has fixed the bug? Fine with me, @blueness, will you add it? If possible, don't revbump the ebuild, as 99% of users aren't affected it by this.
This patch still applies to numpy-1.10.4 and it seems to compile the graph for git now. https://github.com/numpy/numpy/issues/7182#issue-131293676
In the mentioned issue upstream says its fixed. Can you confirm?