[Fix bug] dpdata.gaussian.log plugin can't locate the coords

[Zhang-Zhiyuan-zzy]

At present, the plugin reading gaussian log file identifies the coords context by title: "Z-Matrix orientation:" and "Input orientation:". In general, it works fine. However, the log file, sometimes, might only record the "Standard orientation:" coords. In the case, the coords can't be localized by the current plugin and will raise the UnboundLocalError.

The following is the bug-raised code:

import dpdata
ls = dpdata.LabeledSystem('pair_3.log', fmt='gaussian/md')

The error message:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/pub/conda3/envs/hpu/lib/python3.9/site-packages/dpdata/system.py", line 281, in __init__
    self.from_fmt(
  File "/home/pub/conda3/envs/hpu/lib/python3.9/site-packages/dpdata/system.py", line 318, in from_fmt
    return self.from_fmt_obj(load_format(fmt), file_name, **kwargs)
  File "/home/pub/conda3/envs/hpu/lib/python3.9/site-packages/dpdata/system.py", line 1162, in from_fmt_obj
    data = fmtobj.from_labeled_system(file_name, **kwargs)
  File "/home/pub/conda3/envs/hpu/lib/python3.9/site-packages/dpdata/plugins/gaussian.py", line 23, in from_labeled_system
    return GaussianLogFormat().from_labeled_system(file_name, md=True)
  File "/home/pub/conda3/envs/hpu/lib/python3.9/site-packages/dpdata/plugins/gaussian.py", line 15, in from_labeled_system
    return dpdata.gaussian.log.to_system_data(file_name, md=md)
  File "/home/pub/conda3/envs/hpu/lib/python3.9/site-packages/dpdata/gaussian/log.py", line 49, in to_system_data
    coords_t.append(coords)
UnboundLocalError: local variable 'coords' referenced before assignment

The system and envoriments:

• Operate System: Ubuntu 22.04 lts
• Python: 3.9.0
• dpdata version: 0.2.15

This error can be resolved by making a little modification to a portion of the code: 1) Add a new conditional branch to recognize "Standard orientation:" 2) read the Standard coordinates when meet the flag == 16

The modified code like the following (The modified have been marked by '# TODO: Modified'). This modification would not make changes for the general case, where the "input orientation:" or "Z-Matrix orientation:" is in the text, because of the "Input orientation:" and "Z-Matrix orientation:" are always on the above of "Standard orientation:"

The following modification has been tested on my machine and works fine in both the general case and the above bug-raised case.

[Zhang-Zhiyuan-zzy]

@njzjz In certain situations, it is possible that we may only require the molecular conformer without the need to align the atoms' coordinate orientation with their force vectors. In my particular case, all I need is the molecular conformers and the corresponding energies. Would it be feasible to incorporate a non-default flag to allow for this less stringent operation?

[njzjz]

This needs more discussion. Currently in dpdata.LabeledSystem, force is required. dpdata was initially designed for machine learning potentials training, where force is necessary. It's unclear if reading data without force will benefit users. Besides, it's not hard to let Gaussian print input orientation using geom=printinputorient keyword.

[Zhang-Zhiyuan-zzy]

Thanks for your response, I get it.