dgilford/tcpyPI

Setting specific package versions may not be the best idea.

aramirezreyes opened this issue · 2 comments

aramirezreyes commented

Currently setup.py shows:

    install_requires=[
        "xarray==2022.3.0",
        "numba==0.55.1",
        "numpy==1.22.3",
    ],

Unfortunately this will most certainly conflict with existing environments. For example #44. A more sensible option would be to specify compatible versions using ">=" instead of "==".

dgilford commented

Hi @aramirezreyes ! Thanks for this idea. I've incorporated it into the most recent fix for this installation problem. Please see the commit at 36b4164. Would you be willing to test on your end to check that it works?

burlen commented

I had this problem too. we work around by telling pip --no-dependencies when installing tcpyPI. In our case we are currently using numpy 1.19. tcpyPI seems to work fine with that version of numpy. We use numba 0.53.1 (no reason except that was current when we wrote the install script). We do not use xarray. I wonder what specifically do you need from numpy version 1.22.3? Would it be possible to relax the numpy version requirement slightly? If there is no specific features from numpy/numba then removing the version number will increase compatibility.