MPIBackend platform logic

[gtdang]

Was inspecting the MPIBackend class and found this logic below.

hnn-core/hnn_core/parallel_backends.py

Lines 650 to 655 in ccef332

	# Split the command into shell arguments for passing to Popen
	if 'win' in sys.platform:
	use_posix = True
	else:
	use_posix = False
	self.mpi_cmd = shlex.split(self.mpi_cmd, posix=use_posix)

Isn't this setting posix=True for Windows and Mac (Darwin), and False for Linux? sys.platform docs

Although somehow MPI seems to be working OK on Linux despite the incorrect flag here?

[gtdang]

Tested this shlex.split function with paths with forward and back slashes. The flag doesn't impact posix path parsing, which explains why linux is still functioning with this error. Windows paths would be broken though.

shlex.split('/test/path/posix', posix=True)
Out[13]: ['/test/path/posix']

shlex.split('/test/path/posix', posix=False)
Out[14]: ['/test/path/posix']

shlex.split(r'\test\path\non_posix', posix=True)
Out[15]: ['testpathnon_posix']

shlex.split(r'\test\path\non_posix', posix=False)
Out[16]: ['\\test\\path\\non_posix']

[rythorpe]

Maybe I'm missing something here, but don't we want posix=True only for Windows? If so, the Linux flag is correct as-is.

[gtdang]

posix path is Unix style, you’d want it True for Mac and Linux. The shlex.split function defaults with True, you only want to pass it posix=False for Windows so it can parse backslashes correctly. See the test examples above.

[rythorpe]

Oh I get it now. It's a good thing I asked because I had that all mixed up in my head 🙃 Thanks @gtdang!