Closed JacksonBurns closed 1 year ago
Related, #219 attempted to do this but we dropped it because it caused backwards compatibility issues with Python. Will attempt to resolve here.
mordred
requires an older version of networkx to function properly with Python 3.9.
The fixed version of networkx
is incompatible with moving the version of numpy up -- there is a PR open to fix this on the repo but it's stale. Looks like mordred
has basically been abandoned at this point.
I think that, in order to support future Python versions as well as M1 Macs, we would need to remove mordred
from the default install. We can make it an optional install, i.e. pip install aimsim[mordred]
but clarify that such an install is only supported on Python 3.8
The option that I would prefer would be to fork mordred
, fix the dependency issue, and then include the source code in AIMSim
with the appropriate attribution.
Option 3 is, of course, to just drop mordred
altogether.
Thoughts @himaghna ?
The fixed version of
networkx
is incompatible with moving the version of numpy up -- there is a PR open to fix this on the repo but it's stale. Looks likemordred
has basically been abandoned at this point.I think that, in order to support future Python versions as well as M1 Macs, we would need to remove
mordred
from the default install. We can make it an optional install, i.e.pip install aimsim[mordred]
but clarify that such an install is only supported on Python 3.8 The option that I would prefer would be to forkmordred
, fix the dependency issue, and then include the source code inAIMSim
with the appropriate attribution. Option 3 is, of course, to just dropmordred
altogether. Thoughts @himaghna ?
I think for the short term we might include modred as an optional install with tags (such as -md or some such). Long term I agree moving it into our base code makes the most sense.
Tests failing
@JacksonBurns this should be accompanied by a beta release, likely 2.0.0b0
so that other changes can continue to be made.
Also see #247 when reevaluating Python minor version compatibility and dependency version requirements.
Attempting to resolve #240 by allowing wider range of numpy versions since the latest builds have better support.