Sure,
> In any case, could you please check with nm if libm.so has "cabsl"?
~~ nm /lib/libm.so |grep cab
000000000000d516 T __c99_cabs
000000000000d4ff T __c99_cabsf
000000000001957b T __c99_cabsl
000000000000d52d T cabs
000000000000d51b T cabsf
so, no.
>
> Still trying to figure out if it is a problem of the base system or
> the gfortran configuration.
> But those things linking and symbol things are not exactly my
> expertise, so i'm just trying around.
>
> I also tried gcc8 gfortran by setting it in mk.conf and compiling
> math/blas as suggested.
> gcc8 gets installed and applied, but yields the same cabsl problem again:
>
> bash-5.0# bmake install
> => Bootstrap dependency digest>=20010302: found digest-20190127
> => Checksum SHA1 OK for lapack-3.9.0.tgz
> => Checksum RMD160 OK for lapack-3.9.0.tgz
> => Checksum SHA512 OK for lapack-3.9.0.tgz
> ===> Installing dependencies for blas-3.9.0
> => Tool dependency bsdtar-[0-9]*: found bsdtar-3.4.3
> => Tool dependency cmake>=2.8.1nb1: found cmake-3.17.3
FWIW my cmake is 3.18.1.
I have set up a new system on an empty SSD: NetBSD9.1, latest pkgsrc, sparc64.
The problem is still present and blas-3.9.0 / lapack-3.9.0 will not compile.
I tried to compile the old blas-3.7.1nb1 / lapack-3.7.1nb1 and it will finish without a problem.
The main difference is that the old versions use gmake, while the new version use cmake (i am using cmake 3.19.2).
Any ideas if that could be the cause?