CompilationError: Error executing in get_native(rsys, odesys, 'cvode')

[jankaWIS]

Hi,

I have general problems (I will point out more once I will be sure the problem is not on my side) with this specific line (4th) in this example notebook but this one I got right now on Win10:

native = get_native(rsys, odesys, 'cvode')

and I am getting:

CompilationError: Error executing 'C:\Users\janka.WISMAIN\AppData\Local\Continuum\anaconda3\Library\mingw-w64\bin\g++.exe -fwrapv -pthread -c -std=c++11 -Wall -Wextra -fPIC -O2 -ffast-math -funroll-loops -fopenmp -fno-strict-aliasing -o .\_cvode_wrapper.obj -DPYCVODES_NO_KLU=1 -DPYCVODES_NO_LAPACK=1 -DANYODE_NO_LAPACK=1 -IC:\Users\janka.WISMAIN\AppData\Local\Continuum\anaconda3\lib\site-packages\numpy\core\include -IC:\Users\janka.WISMAIN\AppData\Local\Continuum\anaconda3\lib\site-packages\pyodesys\native\sources -Ic:\users\janka.wismain\appdata\local\continuum\anaconda3\lib\site-packages\pycvodes\include -IC:\Users\janka.WISMAIN\AppData\Local\Continuum\anaconda3\include .\_cvode_wrapper.cpp' in C:\Users\JANKA~1.WIS\AppData\Local\Temp\tmp0j3jigxs. Command exited with status 1 after givning the following output: .\_cvode_wrapper.cpp:1:0: warning: -fPIC ignored for target (all code is position independent)
 /* Generated by Cython 0.29.14 */
 ^
In file included from .\_cvode_wrapper.cpp:622:0:
c:\users\janka.wismain\appdata\local\continuum\anaconda3\lib\site-packages\pycvodes\include/cvodes_cxx.hpp:19:38: fatal error: sundials/sundials_config.h: No such file or directory
compilation terminated.

What is going on and how can I fix it?

I am using:

CPython 3.7.3
IPython 7.17.0
chempy 0.7.12
chempy._release 0.7.12
sympy 1.6.2
numpy 1.16.2
pytest 6.1.1
scipy 1.5.0
pycodeexport 0.1.2

[bjodah]

Hi @jankaWIS,

I don't have access to a Windows machine so I cannot say for certain, perhaps some other user here has. I did however manage to compile a corresponding python extension (and link to sundials) under windows for a tutorial we gave at SciPy 2017. You can find the configuration over here: https://github.com/sympy/scipy-2017-codegen-tutorial

I haven't touched it since though, so expect some "bit-rot". If you manage to get this working, and want to ensure it stays working, we can look into adding a windows build to the continuous integration setup for ChemPy.

[jankaWIS]

Hi @bjodah,

thanks for responding so fast. What exactly do you mean by trying to do the configuration?
I have run your tests (python3 -m pytest -rs -W ignore::chempy.ChemPyDeprecationWarning --pyargs chempy ) and it complains about a missing package:
anaconda3/lib/python3.8/site-packages/chempy/kinetics/tests/test__native.py:89: Unfulfilled requirements. Missing modules: pygslodeiv2
which is giving me the following error when I tried to install it as recommended on your github using conda:

PackagesNotFoundError: The following packages are not available from current channels:

  - pygslodeiv2

I believe this should done the job (so I should probably open an issue there but since it's connected, I will pose it here):
conda install -c conda-forge pygslodeiv2 pytest

And it is both on Mac and Windows. Which brings me to the second problem I am having which I have I summed up here:
https://stackoverflow.com/questions/64348908/chempy-clang-error-unsupported-option-fopenmp.

[bjodah]

Hi again @jankaWIS,

OK, so there are multiple issues here, I used to have a conda channel with my packages, but this was way back before conda-forge was a thing, I've updated the instructions in pygslodeiv2's README: bjodah/pygslodeiv2@e43faf0

Regarding Windows and MacOS: both are systems which I don't work with regularly, I've tried my best not to exclude the possibility to support these systems, but I would almost be surprised if these things work out-of-the-box. The code-generation, compilation & import of python extension module is inherently coupled to the platform. And it was this I alluded to with the link to that scipy2017codegen repository: there we managed to get that workflow to work on Windows, macOS and Linux.

So if you only want to get some results, I would actually recommend just using a Linux machine (virtual or not). If you want to see this working on macOS and Windows, my guess is that it will take some work, perhaps adding fixes/features to one or more of "my" packages linked in the README.

Back when I wrote pycompilation, meson was not very popular, if I would do this again, I would probably look into using it, at least as a back-end.