gh-122885: Improve os.scandir (and os.walk) performance on Windows

[Michael-K-Stein]

Improve os.scandir (and os.walk) performance on Windows

Use native NT syscalls to implement os.scandir rather than WinAPI and POSIX functions.

• Issue: Performance improvement for os.scandir on Windows #122885

[ghost]

The following commit authors need to sign the Contributor License Agreement:

• [email protected]

Click the button to sign:

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[9001]

when I tried this patchset just now, I'm seeing a severe performance degradation. This could of course be due to some particularities of my rig, so I'll include all the relevant details I can think of.

git-head\python.exe -m timeit -s "import os" "for _ in os.scandir('.'): pass"
1000 loops, best of 5: 200 usec per loop

pr122886\python.exe -m timeit -s "import os" "for _ in os.scandir('.'): pass"
500 loops, best of 5: 523 usec per loop

when testing a real-world application, a webserver which uses scandir:

python.exe c:\users\ed\Downloads\copyparty-sfx.py -q
oha --ipv4 --insecure -z11s --fps 4 http://127.0.0.1:3923/?ls

git-head: 680 req/s
pr122886: 595 req/s

when doing a similar test with oha against python -m http.server there was no difference between git-head and this pr, and ditto when running copyparty with --no-scandir (400 req/s).

my environment is windows 10 ltsc (version 1809, build 17763.5576), and I built cpython with VS-buildtools 2019 v16.11.37 using build.bat (no arguments)

the folder I'm listing is cpython\PCbuild after building, so 132 files and 6 dirs