Due to a python-ideas discussion about reducing boilerplate for __file__-relative path calculations, I was running "./python -X importtime -S -c 'import pathlib'" and noticed three potential candidates where it may be worthwhile deferring the imports until the modules are actually needed:

- re (used in _WildcardSelector)
- fnmatch (used in PurePath.match and _WildcardSelector)
- urllib.parse (used in PurePath.as_uri, by way of self._flavour.make_uri)

Using an optimised Python 3.7 on an SSD with warm disk caches, commenting out those 3 imports reduced my typical import times for pathlib from 12-13 ms to 7-8 ms)

Also it's a smaller win, but it might be worth considering whether we can avoid the import of 'nt' on posix platforms and vice-versa.

Did you mean the idea about module-level alias to dirname(__file__)? I think this is a dead horse. See issue33277.

Personally I support moving imports of rarely used modules into functions that need them. But this is against PEP 8 and should be discussed more widely. Perhaps this rule should be weaken, we have to state what exceptions are permitted.