Created attachment 561346 [details]
This happens for both =dev-python/numpy-1.14.5 and =dev-python/numpy-1.15.4 and probably other versions as well.
I have checked and both lapack and cblas are installed and selected.
Created attachment 561348 [details]
Would you using gcc-8 by any chance?
x86_64-pc-linux-gnu-gcc -march=native -O2 -pipe /var/tmp/portage/dev-python/numpy-1.15.4/temp/tmpW2Yg3q/var/tmp/portage/dev-python/numpy-1.15.4/temp/tmpW2Yg3q/source.o -L/usr/lib64 -lrefblas -lrefcblas -o /var/tmp/portage/dev-python/numpy-1.15.4/temp/tmpW2Yg3q/a.out
/usr/lib/gcc/x86_64-pc-linux-gnu/8.2.0/../../../../x86_64-pc-linux-gnu/bin/ld: warning: libgfortran.so.4, needed by /usr/lib64/librefblas.so, not found (try using -rpath or -rpath-link)
libgfortran.so.4 is provided by gcc-7 but gcc-8 provides libgfortran.so.5 which means that blas and lapack have to be rebuilt.
As you have correctly identified from the log, I do indeed use `x86_64-pc-linux-gnu-8.2.0`. Your suggestion also solved the issue :) thanks!
The only remaining question for me is: why did neither `emerge @preserved-rebuild` nor `revdep-rebuild` pick up on this?
That particular one is not picked up for a number of reasons. It is a hard case. But I have been fielding that problem for some months.
It started when moving to gcc-7 where libgortran moved from so.3 to so.4. And now with gcc-8 where it moved to so.5.
For a long time gfortran stayed at so.3 so it was not a problem. Part of the issue is FORTRAN itself which is underdefined compared to C or C++. Unlike those there is no "forward" compatibility in the language spec.
Can confirm, reinstalling lapack and blas after upgrade to gcc 8 fixed issues with numpy for me.
Sorry that we did not foresee this during the gfortran transition. But let's close it as obsolete for now.
Sorry, it is not obsolete. There is still potential breakage in the future.
(In reply to Horea Christian from comment #3)
> As you have correctly identified from the log, I do indeed use
> `x86_64-pc-linux-gnu-8.2.0`. Your suggestion also solved the issue :) thanks!
> The only remaining question for me is: why did neither `emerge
> @preserved-rebuild` nor `revdep-rebuild` pick up on this?
I am having the same question.
This is finally fixed in bug 646090.