[syahra712]

UPath.match() passes the pattern straight to full_match(), so a relative pattern has to cover the whole path. pathlib.PurePath.match() matches relative patterns from the right instead:

>>> from pathlib import PurePosixPath
>>> PurePosixPath("/x/a/b").match("a/b")
True
>>> from upath import UPath
>>> UPath("s3://bucket/x/a/b").match("a/b")
False   # before
True    # after

The same thing happens with match("b"), match("*/b") and so on, and for every protocol (memory, s3, gcs, http, ...).

Fix: match() tries full_match() first, so nothing that matched before stops matching. If that fails and the pattern is relative (it has no leading / and no ://), it compares the pattern's segments against the path's last parts, using fnmatchcase on each segment. The anchor (bucket/, https://host/) is left out of that comparison, as pathlib does. The ** → * handling is unchanged.

Tests: added test_match to the shared BaseTests in upath/tests/cases.py, so it runs for every implementation. DataPath gets an override, as it already does for test_full_match. The new test fails on main and passes with this change. These local runs pass: memory, local, data, s3, zip, tar, test_relative.py, test_core.py and the pathlib suites (1588 + 736). pre-commit is clean.

AI-assisted (Claude): I used it to help read the code and draft the test. I reviewed and ran everything myself.