It seems like there's a history of issues here and different workarounds. Some keywords would be "GCC_4.6.0", "cblas", possibly "libgfortran".
I was running into the issue with the most current port revision and after a bunch of make / make clean cycles I've gotten it to compile and work just by making changes to /usr/ports/math/py-numpy/files/site.cfg 1) Changing the variable "library_dirs" to be equal to /usr/lib/local/gcc49 , removing the other colon-separated alternatives. Not removing the other alternatives resulted in the build failing. 2) Adding /usr/local/lib/gcc49/include to the "include_dirs" variable. I did not have to remove the other variables. It seems like the build system is preferentially trying to link against one of the openblas.so files in /usr/local/lib as opposed to (say) /usr/local/lib/gcc49 . However the one in /usr/local/lib has incompatible versions on my system, and so the compile fails. Trying this again*, the build also works compiled against the corresponding gcc6 folders instead of gcc49. I am using the pkg-provided gcc6. Hopefully this information helps wiser people solve the problem for others. Joe *: Trying again in this case means make clean, changing things then make deinstall, make reinstall, and runnin numpy.test(label='fast', verbose=10) in python3.6. _______________________________________________ freebsd-python@freebsd.org mailing list https://lists.freebsd.org/mailman/listinfo/freebsd-python To unsubscribe, send any mail to "freebsd-python-unsubscr...@freebsd.org"
