replace glob with rglob #283

[Keunmo]

PR for fix #283.

Replace Path(some_path).glob('**/*') with Path(some_path).rglob('*/*').

Act same as original code, and it can also find files from symlink.

[sarlinpe]

Thanks for sending this PR. The doc mentions

This is like calling Path.glob() with “**/” added in front of the given relative pattern

So shouldn't the update be

- paths += list(Path(root).glob('**/'+g))
+ paths += list(Path(root).rglob(g))

instead of

- paths += list(Path(root).glob('**/'+g))
+ paths += list(Path(root).rglob('*/'+g))

?

[Keunmo]

Thanks for reviewing @skydes !

I've tested with several cases, and I think current update is more suitable for handling various scenarios involving symlinks.

• Case 1: When dataset path is a symlink, and mapping path and query path are regular directories.
• Case 2: When dataset path is a regular directory, and mapping path and query path are symlinks.
• Case 3: When dataset path, mapping path, and query path are all symlinks.

In these cases, paths += list(Path(root).rglob('*/'+g)) worked well for all three cases, but paths += list(Path(root).rglob(g)) only worked for case 1 and failed for cases 2 and 3. I think it's because when rglob traversal, if a child directory is a symlink, it does not explore within that symlink directory.

So, I believe current approach is more appropriate.

[sarlinpe]

Path(root).rglob('*/'+g) only handles symlinks at the top level, unlike glob.glob which handles arbitrary levels. This is exactly what the cpython issue says. Let's instead use glob, as:

paths += glob.glob((Path(root) / '**' / g).as_posix()), recursive=True)

Can you please confirm that it works and make the changes? Thanks!

[Keunmo]

You're right, glob.glob looks better. I changed it to glob, but it returns list with str, not pathlib.PosixPath. So I convert it to PosixPath like this:

paths += list(map(Path, glob.glob((Path(root) / '**' / g).as_posix(), recursive=True)))

And it works all good. Thanks!