Open ikrommyd opened 5 days ago
This is the same problem as described in #15309: if the BLAS version that is shipped by ROOT is different from the one that is shipped by other python packages, there will be a collision of different BLAS libraries.
For this case of using ROOT with Conda, the solution would be either to make sure the Conda build uses the same BLAS library as NumPy, using the BLA_VENDOR
flag for CMake, of nothing else works set tmva-cpu=OFF
for the Conda builds, as this is the component that makes ROOT vendor a BLAS library.
Well this arised particularly by trying to have a conda environment with both root
and zfit
. zfit depends on tensorflow and therefore tensorflow would either be installed from conda-forge or from pypi. I'm not really sure I understand your proposed solution. Can I do all that without compiling root myself and while grabbing it from conda? If so, at which point of the installation?
cc @jonas-eschle
Also, I'd like to mention that I'm getting this on x86 Linux and I used lxplus to reproduce. If I do mamba create -n root-bug root=6.32 tensorflow
on my apple silicon macbook, then it works fine.
@guitargeek thanks, so as it seems to be a conda-specific dependency issue then, we can take it there: https://github.com/conda-forge/root-feedstock/issues/253
Check duplicate issues.
Description
Importing
tensorflow
and thenROOT
results in a segmentation fault with ROOT 6.32. The other way around works fine.Reproducer
ROOT version
6.32/0
Installation method
Conda:
mamba create -n root-bug root=6.32 tensorflow
Operating system
Linux
Additional context
Full seg fault trace: