Filter out malformed nvidia-smi process_name XML tag

[jfennick]

No description provided.

[codecov]

Codecov Report

Merging #1910 (239ce5a) into main (b0c8dda) will decrease coverage by 1.12%.
The diff coverage is 94.44%.

@@            Coverage Diff             @@
##             main    #1910      +/-   ##
==========================================
- Coverage   83.98%   82.86%   -1.12%     
==========================================
  Files          46       46              
  Lines        8190     8200      +10     
  Branches     2174     2138      -36     
==========================================
- Hits         6878     6795      -83     
- Misses        840      903      +63     
- Partials      472      502      +30     

Files	Coverage Δ
cwltool/cuda.py	92.45% <94.44%> (+6.40%)	⬆️

... and 12 files with indirect coverage changes

[jfennick]

FYI, as I discuss in the block comment in the code, to fix the immediate problem the only change necessary is
nvidia-smi -q -x | grep -v process_name
If that's more likely to get merged sooner, then we can do that for now.

But in an attempt to prevent future problems, I have instead removed all tags except the tags that are actually used. Let me know which version you prefer.

[mr-c]

2023-10-03T19:27:36.5351674Z cwltool/context.py:46: initilizer ==> initializer
2023-10-03T19:27:36.5352997Z tests/test_toolargparse.py:183: desription ==> description
2023-10-03T19:27:36.5354037Z tests/test_toolargparse.py:192: desription ==> description
2023-10-03T19:27:36.5354867Z tests/wf/schemadef-bug-1473.cwl:452: cylces ==> cycles
2023-10-03T19:27:36.5441419Z Probable typo foun. Run "make codespell-fix" to accept suggested fixes, or add the word to the ignore list in setup.cfg

[mr-c]

Suggested change

	def cuda_device_count() -> str:
	def cuda_device_count() -> int:

and update the return as well

[tetron]

I would prefer that instead of using a subshell it captures the output and does the substring search in Python. For a simple substring search this really is as simple as something like:

out = subprocess.check_output(["nvidia-smi", "-q", "-x"])
if "cuda_version" not in out:
   raise ...

[tetron]

Actually instead of using minidom, to pull a single tag it would be easier to just use a regular expression to look for <cuda_version>.