defect analysis as an add-on package

[jmmshn]

Hi @mkhorton @shyuep,
I'm working on adding some defect analysis code that will have a few new dependencies like (dscribe and mp-pyrho)
I'm wondering if it makes sense to separate the analysis.defects as an add-on package since I heard we are worried about additional dependencies.

As far as I can tell the defect code is used only in one other places strucutre_matcher.py so it should not be so disruptive, also I think a few of the charge density-based interstitial tests are pretty time-consuming so that is a possible plus as well

[mkhorton]

Also sk-image as pointed out by @acrutt

[shyuep]

I definitely support moving defect analysis to a add-on package. As far as I know, the only place where the structure matcher uses it is in the DefectComparator. That can definitely be either moved back to analysis.defects or alternatively, can be surrounded by a try-catch to inform users to install the pymatgen-analysis-defects package if they want to use it.

[shyuep]

FYI, pls refer to https://github.com/materialsproject/pymatgen-addon-template for how to make the addon.

[mkhorton]

I already pointed this out to @jmmshn in separate conversation, but there is an issue with numba (a dep of dscribe and others) being fundamentally incompatible with the latest pymatgen due to not supporting the latest numpy.

It's tricky, because numba is used by a lot of packages (eg, I know the developers of py4dstem want to use pymatgen but also want to use numba and it's causing them issues).

[mkhorton]

In the context of this conversation, having defects be a separate package will at least mean these version incompatibly issues aren't affecting pymatgen core, but they will still have to be dealt with.

[shyuep]

I would recommend that everyone who needs to work with packages requiring old numpy to just use pymatgen 2022.0.x (specifically 2022.0.17). As far as I know, all those still work. I am open to ideas as to how to solve the numpy issue. I have no idea why binary incompatibilities exist between different versions of numpy. It has already happened a few times in the past one year.