gh-121445: C implementation for fnmatch

[picnixz]

• Update configure script.
• C implementation
  • use of lru_cache
  • use of os.path.normcase
  • fnmatch.fnmatch
  • fnmatch.fnmathcase
  • fnmatch.translate
• Make it work on Windows CI/CD.
• Add documentation
  • C documentation
  • Python documentation
  • NEWS entry
• Increase test coverage

• Issue: Add C accelerator for fnmatch  #121445

[picnixz]

I forgot that MSVC didn't have an implementation of fnmatch... I'll add a guard for Windows so that it uses the pure Python implementation.

[picnixz]

So, everything should be ready. Documentation and NEWS entry need to be written and I also need to know how I can make it work on Windows.

@zooba I know you are well-versed on Windows platforms so could you help me there please if you have time? How can I specify the files to be compiled correctly...?

[zooba]

Easiest way is to look at any other similar module, find all the references to it, and set yours up the same way. Looks like you'll have a bit of rearranging to do here (e.g. header goes somewhere under Include).

Filename matching should also be a subset of regex, so it may be easier and more efficient to write our own comparison function. That would also then be fully portable, as well as being easier to handle compatibility.

[picnixz]

Easiest way is to look at any other similar module, find all the references to it, and set yours up the same way. Looks like you'll have a bit of rearranging to do here (e.g. header goes somewhere under Include).

Thank you for that (actually, that's how I managed to do it for the linux build...). I actually never used MSVC so I don't know how to compile nor what I should actually specify but I'll do it the old way (trial and error).

Filename matching should also be a subset of regex, so it may be easier and more efficient to write our own comparison function. That would also then be fully portable, as well as being easier to handle compatibility.

For now, I've just replicated the Python implementation which translates UNIX shell patterns to RE patterns. Strictly speaking, I don't think the translation is needed when comparing filenames (since we would anyway use a subset of regex indeed), but I didn't want to introduce a new algorithm (which I more or less gave up to describe after 10 minutes...)

[picnixz]

Mmmh... so the improvements are.... not improvements =/ Well I'll just forget about using interned strings actually.

EDIT 2: mmh, actually it didn't change at all. So what happened between yesterday and today for 'fnmatch.translate' to be a bit slower... (maybe the merge?)

[picnixz]

I think I'll leave the implementation like that for now and open it for review. At least, we have 1.7x performance gain for fnmatch.translate (but let's say we only have 1.5x to be very conservative).