bpo-46337: Urllib.parse scheme-specific behavior without reliance on URL scheme by oldaccountdeadname · Pull Request #30520 · python/cpython

Some features in urllib are dependent on schemes, (i.e., preserving the
netloc in url joining). Prior to this patch, this was governed by the
uses_* lists (uses_relative, uses_netloc, uses_params) which hard code
these attributes for certain schemes. Providing an enum interface and a
'constructor' that allows overrides makes this mechanism a bit more
flexible for future modifications.

This allows the callers of urljoin and urlparse to add guaranteed scheme
classes to the url regardless of the actual scheme, which may not be in
the default uses_* lists of schemes. This call-time behavior is done
through an optional parameter that preserves backwards compatibility.

A test case is added for this, and requires the change present in
test_urlparse.checkJoin.

urljoin will not treat `..` as moving up one directory rather than
moving up one file, thus causing the doctests to fail due to a missing
trailing slash. Both changes are of the form:

http://example.org/post/x -> http://example.org/post/x/

Additionally, the my-protocol example's expected output had the wrong
scheme.

Minor style things:
+ _scheme_classes' docstring's summary was made explicit.
+ _scheme_classes was prepended with and followed by two newlines.