glob('*.txt') is much slower than it should be on zip files with many recursive files

[jaraco]

glob('*.txt') is much slower than it should be on zip files with many recursive files.

This is because it uses ._descendants() (which lists all recursive descendants) when there's no need, as this glob should only need to read the top-level .iterdir():

zipp/zipp/__init__.py

Line 373 in ee6d711

for child in self._descendants()

For large zip files, this can easily make a 1000x speed difference, which is surprising when you're writing code that aims to work on both pathlib.Path and zipfile.Path.

Originally posted by @nh2 in #98 (comment)