DefectsParser hangs on Windows

[shubhmsme93]

Trying to parse multiple defect calculations using DefectsParser doesn't progress on Windows at all

and just hangs. For example, please see the attached screenshot showing parsing of the 'CdTe' folder downloaded from https://github.com/SMTG-Bham/doped. This is on a Windows 10 machine (32 GB RAM, i9, 8 cores). The same works fine on a Linux VM installed on the same Windows guest. I have tried both with and without the processes argument without any success. Likely an issue with how multi-processing is handled on Windows? Any workaround would be appreciated!

[kavanase]

Hi Shubham! Hmm, if processes is set to 1, then there should be no multi-processing attempted;
https://github.com/SMTG-Bham/doped/blob/main/doped/analysis.py#L1008-L1016
So I'm not sure what the issue is here, but something else related to Windows it seems. Unfortunately it's not so easy to debug here with no error messages, and I don't have Windows so can't reproduce easily.

Most Windows users we know use WSL (so similar to your Linux VM I assume), and that should have no issues. Is it possible for you to use them, or you need to run on native Windows?

[shubhmsme93]

Hi Seán, thanks for your reply. Unfortunately, there are no error messages and it just gets stuck. Its the same issue with even parsing just one defect with DefectParser. Actually, I am trying to call doped inside our company's simulation software workflows (MedeA) for defect calculations done with VASP. The workflows work seamlessly on Linux OS. 95% of our users use Linux machines, so no issues there. However, there still maybe a handful of users wanting to use Windows OS. The issue with WSL is that our software workflows do not work robustly on there and thus a Linux VM is a better solution. But a Linux VM may not always be possible for those users working with native Windows. Thus, the need to run DefectsParser on native Windows. I understand that you don't have Windows at the moment and may not be able to look into this issue immediately. But in case if you ever get a chance to test this on Windows in the future, it would be very helpful. I'll also keep playing with it in the meanwhile to see if I find any solution.

[kavanase]

Ok, thanks for the info!

If you are able to debug at all, even just finding where the issue is coming from, that would be very useful. I'm not sure if some of the dependent packages, like vise/pydefect, are incompatible with Windows, which could potentially cause these issues. (If so, their use can be avoided if needs be, they are not crucial to parsing).

[shubhmsme93]

Disabling the vise package made it to work! However, this also automatically disabled the pydefect package which seems like is needed to apply the eFNV correction scheme. Please see the pydefect and eFNV related warning messages in the snippet below. Is it possible to make pydefect not depend on vise to still be able to use eFNV correction scheme and all pydefect related functionalities (seems like it is also needed to compute carrier concentrations; pydefect >> analyzer >> concentration?)

In [15]: dp = DefectsParser("dir500/971/Stage_1/", subfolder="Stage_1", bulk_path="dir500/984/Stage_1/", bulk_band_gap_vr="dir500/982/Stage_1/BSTR1 vasprun.xml", dielectric=9, proc
    ...: esses=1)
Parsing Iter_000007/Stage_1: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 7/7 [00:05<00:00,  1.89it/s]analysis.py:1781: UserWarning: Warning(s) encountered when parsing Iter_000003 at C:\MD\Jobs\dir500\971\Stage_1\Iter_000003\Stage_1:
Detected atoms far from the defect site (>4.19 Å) with major displacements (>0.5 Å) in the defect supercell. This likely indicates a mismatch between the bulk and defect supercell definitions (-> see troubleshooting docs) or an unconverged supercell size, both of which could cause errors in parsing. The mean displacement of the following species, at sites far from the determined defect position, is >0.5 Å: ['Li'], with displacements (Å): {'Li': [np.float64(1.16)]}
analysis.py:1787: UserWarning: Defects: {'Iter_000006', 'Iter_000007', 'Iter_000001', 'Iter_000005', 'Iter_000002', 'Iter_000004', 'Iter_000003'} each encountered the same warning:
Got this error message when attempting to apply finite-size charge corrections:
To use the Kumagai (eFNV) charge correction, you need to install pydefect. You can do this by running `pip install pydefect`.
-> Charge corrections will not be applied for this defect.
analysis.py:1787: UserWarning: Defects: {'Iter_000006', 'Iter_000007', 'Iter_000001', 'Iter_000005', 'Iter_000002', 'Iter_000004', 'Iter_000003'} each encountered the same warning:
Projected eigenvalues/orbitals parsing failed with error: ImportError('To perform eigenvalue & orbital analysis, you need to install pydefect. You can do this by running `pip install pydefect`.')

[kavanase]

Ah ok! This is useful to know, but I'm not sure what we can do from here.
pydefect (and thus vise) are only used for the eFNV charge correction and for eigenvalue analysis. The latter is certainly not crucial, but the former is for accurate formation energies in highly-anisotropic systems (otherwise the FNV correction can be used). pydefect is not needed for concentrations or any thermodynamic analyses.

I'm not sure if it's a straightforward fix for vise to be made compatible with Windows. Avoiding the problematic code in vise might not be possible, but I'm trying something now with the develop branch

[shubhmsme93]

Thanks for the explanation and for further looking into it! eFNV is indeed crucial for accurate image-charge correction, as you said, for highly-anisotropic systems.