gh-117607: Speedup os.path.relpath()

[nineteendo]

• Issue: Speedup os.path.relpath() #117607

[nineteendo]

@eryksun, could you run a benchmark for os.path.relpath() on your machine? Otherwise it might undersell the performance improvement.

[eryksun]

@eryksun, could you run a benchmark for os.path.relpath() on your machine? Otherwise it might undersell the performance improvement.

Windows - ntpath - 1.407-1.565 speedup

> .\python_main -m timeit -s "import ntpath" "ntpath.relpath('C:foo', start='C:/Users/account')"
50000 loops, best of 5: 6.7 usec per loop
> .\python_work -m timeit -s "import ntpath" "ntpath.relpath('C:foo', start='C:/Users/account')"
50000 loops, best of 5: 4.28 usec per loop

> .\python_main -m timeit -s "import ntpath" "ntpath.relpath('C:/', start='C:/Users/account')"
50000 loops, best of 5: 4.84 usec per loop
> .\python_work -m timeit -s "import ntpath" "ntpath.relpath('C:/', start='C:/Users/account')"
100000 loops, best of 5: 3.44 usec per loop

Linux - posixpath - 1.137-1.189 speedup

$ ./python_main -m timeit -s "import posixpath" "posixpath.relpath('foo', start='/home/account')"
100000 loops, best of 5: 3.14 usec per loop
$ ./python_work -m timeit -s "import posixpath" "posixpath.relpath('foo', start='/home/account')"
100000 loops, best of 5: 2.64 usec per loop

$ ./python_main -m timeit -s "import posixpath" "posixpath.relpath('/', start='/home/account')"
200000 loops, best of 5: 1.74 usec per loop
$ ./python_work -m timeit -s "import posixpath" "posixpath.relpath('/', start='/home/account')"
200000 loops, best of 5: 1.53 usec per loop

Windows - posixpath - 1.123-1.136 speedup

> .\python_main -m timeit -s "import posixpath" "posixpath.relpath('foo', start='/home/account')"
50000 loops, best of 5: 4.39 usec per loop
> .\python_work -m timeit -s "import posixpath" "posixpath.relpath('foo', start='/home/account')"
100000 loops, best of 5: 3.91 usec per loop

> .\python_main -m timeit -s "import posixpath" "posixpath.relpath('/', start='/home/account')"
100000 loops, best of 5: 3.1 usec per loop
> .\python_work -m timeit -s "import posixpath" "posixpath.relpath('/', start='/home/account')"
100000 loops, best of 5: 2.73 usec per loop

Linux - ntpath - 1.006-1.007 slowdown (no change)

$ ./python_main -m timeit -s "import ntpath" "ntpath.relpath('C:foo', start='C:/Users/account')"
50000 loops, best of 5: 8.63 usec per loop
$ ./python_work -m timeit -s "import ntpath" "ntpath.relpath('C:foo', start='C:/Users/account')"
50000 loops, best of 5: 8.69 usec per loop

$ ./python_main -m timeit -s "import ntpath" "ntpath.relpath('C:/', start='C:/Users/account')"
50000 loops, best of 5: 6.29 usec per loop
$ ./python_work -m timeit -s "import ntpath" "ntpath.relpath('C:/', start='C:/Users/account')"
50000 loops, best of 5: 6.33 usec per loop

[nineteendo]

Could you also check with start not set or a relative path?

[eryksun]

Could you also check with start not set or a relative path?

Sure, except testing ntpath.relpath("C:/") isn't meaningful on POSIX. It fails because the "C:" drive is a different device namespace.

Windows - ntpath - 1.201-1.269 speedup

> .\python_main -m timeit -s "import os,ntpath; os.chdir('C:/Users')" "ntpath.relpath('foo')"
50000 loops, best of 5: 4.84 usec per loop
> .\python_work -m timeit -s "import os,ntpath; os.chdir('C:/Users')" "ntpath.relpath('foo')"
50000 loops, best of 5: 4.03 usec per loop

> .\python_main -m timeit -s "import os,ntpath; os.chdir('C:/Users')" "ntpath.relpath('C:/')"
50000 loops, best of 5: 4.11 usec per loop
> .\python_work -m timeit -s "import os,ntpath; os.chdir('C:/Users')" "ntpath.relpath('C:/')"
100000 loops, best of 5: 3.24 usec per loop

Linux - posixpath - 1.309-1.434 speedup

$ ./python_main -m timeit -s "import os,posixpath; os.chdir('/home')" "posixpath.relpath('foo')"
100000 loops, best of 5: 2.94 usec per loop
$ ./python_work -m timeit -s "import os,posixpath; os.chdir('/home')" "posixpath.relpath('foo')"
100000 loops, best of 5: 2.05 usec per loop

$ ./python_main -m timeit -s "import os,posixpath; os.chdir('/home')" "posixpath.relpath('/')"
100000 loops, best of 5: 2.12 usec per loop
$ ./python_work -m timeit -s "import os,posixpath; os.chdir('/home')" "posixpath.relpath('/')"
200000 loops, best of 5: 1.62 usec per loop

Windows - posixpath - 1.060 slowdown, 1.019 speedup

> .\python_main -m timeit -s "import os,posixpath; os.chdir('C:/Users')" "posixpath.relpath('foo')"
50000 loops, best of 5: 4.01 usec per loop
> .\python_work -m timeit -s "import os,posixpath; os.chdir('C:/Users')" "posixpath.relpath('foo')"
50000 loops, best of 5: 4.25 usec per loop

> .\python_main -m timeit -s "import os,posixpath; os.chdir('C:/Users')" "posixpath.relpath('/')"
100000 loops, best of 5: 3.22 usec per loop
> .\python_work -m timeit -s "import os,posixpath; os.chdir('C:/Users')" "posixpath.relpath('/')"
100000 loops, best of 5: 3.16 usec per loop

Linux - ntpath - 1.004 speedup (no change)

$ ./python_main -m timeit -s "import os,ntpath; os.chdir('/home')" "ntpath.relpath('foo')"
50000 loops, best of 5: 8.23 usec per loop
$ ./python_work -m timeit -s "import os,ntpath; os.chdir('/home')" "ntpath.relpath('foo')"
50000 loops, best of 5: 8.2 usec per loop

[erlend-aasland]

Thanks for running benchmarks, @eryksun 🙏

[nineteendo]

cc @barneygale, @zooba, @serhiy-storchaka

[zooba]

Thanks!