Wrong gfortran path from nc-config #48
Comments
|
@jsalort, indeed, I don't think |
|
The prefix replacement should have done this? I wonder why nc-config ended up with the build env and not the host. We could try to merge the build and host envs. |
|
@beckermr, thanks, I didn't realize files like |
|
prefix replacement should be done on everything. If not, one can force it. |
|
@beckermr, would you be able to make a PR to force this? Or point me to the right place in the documentation to figure out how to do it myself? |
|
try setting |
|
@xylar I am not sure how to try. $ conda update -c conda-forge netcdf-fortranbut this does not update netcdf-fortran. I have 4.5.2 version, nompi_h45d7149_104 build, according to: $ conda list netcdf-fortran
# packages in environment at /home/jsalort/anaconda3/envs/py38:
#
# Name Version Build Channel
netcdf-fortran 4.5.2 nompi_h45d7149_104 conda-forgeAccording to $ conda search -c conda-forge netcdf-fortran --infothere is a newer nompi_hfef6a68_106 dated 2020-05-16. So I tried this: $ conda install -c conda-forge netcdf-fortran=4.5.2=nompi_hfef6a68_106but it fails with conflicts. |
|
@jsalort, could you create a new conda environment instead to test with? My guess is that your conflicts might relate to mixing packages from (see https://conda-forge.org/) |
|
Thank you for your answer @xylar. I now use strict channel priority. However, it seems that my problem is that the new build of netcdf-fortran is not compatible with h5py somehow. I am not sure if this issue belongs here or not, but in practise I cannot benefit from the new build because of this. $ conda create -n test -c conda-forge python=3.8
$ conda activate test
$ conda config --add channels conda-forge
$ conda config --set channel_priority strict
$ conda install netcdf-fortran
$ nc-config --fc
/home/jsalort/anaconda3/envs/test/bin/x86_64-conda_cos6-linux-gnu-gfortran
$ conda install h5py
Collecting package metadata (current_repodata.json): done
Solving environment: done
## Package Plan ##
environment location: /home/jsalort/anaconda3/envs/test
added / updated specs:
- h5py
The following NEW packages will be INSTALLED:
h5py conda-forge/linux-64::h5py-2.10.0-nompi_py38h513d04c_102
libblas conda-forge/linux-64::libblas-3.8.0-14_openblas
libcblas conda-forge/linux-64::libcblas-3.8.0-14_openblas
liblapack conda-forge/linux-64::liblapack-3.8.0-14_openblas
libopenblas conda-forge/linux-64::libopenblas-0.3.7-h5ec1e0e_6
numpy conda-forge/linux-64::numpy-1.18.4-py38h8854b6b_0
six conda-forge/noarch::six-1.14.0-py_1
The following packages will be DOWNGRADED:
hdf5 1.10.6-nompi_h3c11f04_100 --> 1.10.5-nompi_h3c11f04_1104
libnetcdf 4.7.4-nompi_h84807e1_104 --> 4.7.4-nompi_h9f9fd6a_101
netcdf-fortran 4.5.2-nompi_hfef6a68_106 --> 4.5.2-nompi_h45d7149_104
Proceed ([y]/n)? y
Preparing transaction: done
Verifying transaction: done
Executing transaction: done
$ nc-config --fc
/home/conda/feedstock_root/build_artifacts/netcdf-fortran_1585602845013/_build_env/bin/x86_64-conda_cos6-linux-gnu-gfortran |
|
@jsalort I can reproduce this under linux and I'll look into it as soon as I have a chance. |
|
@jsalort, it looks like we're in the middle of hdf5 migrating to a new version. This fix went in after If you keep an eye on that pull request, you should be able to take advantage of the fix once that happens. Alternatively, we could do a special build of |
|
@xylar Thank you for the investigation. It seems that there are PR under review, so I guess it will not take too long before the migration is complete? In that case, I can certainly wait. No need to make a special build. Thanks again. |
|
@jsalort, I'm assuming this got resolved when conda-forge/h5py-feedstock#72 got merged. Please commend and re-open if not. |
|
It seems this problem reappeared with version 4.5.4: But it works with 4.5.3: |
|
This only affects the nompi version. The mpich and openmpi version of 4.5.4 seem to work. |
|
I do not unfortunately. :/ |
|
I think the underlying problem is that only |
|
I think the analogous line should be |
Issue:
nc-configreports the following gfortran compilerwhich breaks Makefile which relies on
nc-config --fcto give the actual Fortran compiler that should be used.I would expect it to return a bare
x86_64-conda_cos6-linux-gnu-gfortran, or the path to the actual compiler in the active environment, e.g. in my case,/home/jsalort/anaconda3/envs/py38/bin/x86_64-conda_cos6-linux-gnu-gfortran.But maybe I am wrong, and
nc-configjust always return the absolute path in the system in which it was built. But then, it hardly seems useful.Environment (
conda list):Details about
condaand system (conda info):The text was updated successfully, but these errors were encountered: