fixing malparsed windows paths

[tuix]

Fixes #4277

[marc-h38]

The posix= parameter is designed to manage nuances between different Unix shells, so it's very mysterious how this helps with Windows and why?!

I realize this is still just a draft, but some... draft comments and a commit message draft would not hurt.

Generally speaking, shlex is NOT meant to help prepare Windows commands ("outputs"). According to its documentation, it's meant as a convenience to help parsing inputs to the program, like a simple configuration file or some parameters.
https://docs.python.org/3/library/shlex.html

Longer example of shlex abuse and Windows compatibility story in zephyrproject-rtos/west@94f8a04

[ferkulat]

@marc-h38
"The posix= parameter is designed to manage nuances between different Unix shells, so it's very mysterious how this helps with Windows and why?!"
When the posix=false is not passed on windows, shlex is eating up back slashes:

And here with posix=False:

here is some related issue:
#mesonbuild/meson#5726

[marc-h38]

The posix= parameter is designed to manage nuances between different Unix shells,

Correction: posix=True means backwards-compatibility=False. This could/should be better documented. I found the evidence here:
python/cpython#38314 (comment)

so it's very mysterious how this helps with Windows and why?!

When the posix=false is not passed on windows, shlex is eating up back slashes:

Right, escaping rules are different between posix and non-posix:
https://docs.python.org/3/library/shlex.html#parsing-rules

However this is just one difference among a long list of other differences. shlex is effectively two pretty different parsers in one. Those other differences should be understood and evaluated in this particular context and the commit message should tell that story. When not, the "Pragrammatic Programmer" book names this "programming by coincidence".

Or maybe I should say "called" in the past tense now since AI is writing all the code ;-)

[marc-h38]

here is some related issue: #mesonbuild/meson#5726

Apparently, the fix there was to stop (ab)using shlex for cases it was not designed to handle, and replace it with something else entirely.

[ferkulat]

Just found by chance that #5017 is another tentative fix.

no, #5017 is a fix for back slashes in the "file" field of compile_database.json. but not for failing to extract the compiler path, like this fix