Add more pathlib benchmarks

[zmievsa]

A small extension on current benchmarks related to pathlib.

My current implementation is very barebones and dumb because I am quite new to writing pyperformance benchmarks and benchmarks in general. Any and all opinions on how to improve my work are welcome!

[mdboom]

I have no comment one way or the other about these changes, but we should rename the benchmark to pathlib2 if we do this so that comparisons won't be misleading.

[zmievsa]

@mdboom could you help me understand what you mean by "so that comparisons won't be misleading"?

[mdboom]

@mdboom could you help me understand what you mean by "so that comparisons won't be misleading"?

When a benchmark is changed significantly like this, previously run baseline benchmarks are no longer meaningful to compare against, and could lead someone to make the wrong decision based on the comparison.

[barneygale]

Sorry for taking so long to look at this.

Thanks for putting this together. I'm already using it locally, with some modifications :)

IIUC, pyperformance isn't designed for function-by-function benchmarking. I think we might want to merge some of these test cases. Overall, I think I'm aiming for:

• pathlib: previous code for this benchmark per @mdboom's comment
• pathlib_construct: cover PurePath(), Path(), joinpath(), fspath(path)
• pathlib_normalize: cover drive, root, anchor, parts, name, suffix, suffixes, stem, with_name(), with_stem(), with_suffix(), relative_to(), is_relative_to(), parent, parents, is_reserved(), match()
• pathlib_string: cover str(path), as_posix(), bytes(path), path.as_uri(), repr(path).
• pathlib_compare: cover a == b, hash(a), a < b
• pathlib_fs: cover absolute(), stat(), open(), touch(), mkdir(), unlink(), rmdir(), exists(), is_dir(), is_file(),
• pathlib_fs_walk: cover iterdir(), glob(), rglob(), walk()

(pyperformance folks, please correct me if I'm doing this wrong)

This PR could add pathlib_construct and pathlib_normalize.

We need more realistic test cases for path construction. We need a variety of paths: POSIX and Windows, absolute and relative, short and (some) long. I'd imagine that the average path length falls off quite rapidly, so most of our paths should be <5 components long, with very few >10. It would be good to generate some realistic-looking file extensions too (the benchmark already does something like this, but for concrete paths).