Remaining --list-tests time breaks down as follows:
13% in ensure_generator (plus a bit more due to sorting), which I don't see how to improve without either changing our definition of a "method" or changing how fixtures are run, either of which would be fairly disruptive.
48% in discover_from, a decent chunk of which is spent in the inspect module determining whether something is a "real" function, which again would be disruptive to change.
41% doing imports, which is a different problem altogether. (I don't believe this time overlaps with the above two, but it's hard to tell.)
That's just over 100% accounted for, so I don't think I can do much better from here. :)
~10% faster.
Remaining
--list-teststime breaks down as follows:ensure_generator(plus a bit more due to sorting), which I don't see how to improve without either changing our definition of a "method" or changing how fixtures are run, either of which would be fairly disruptive.discover_from, a decent chunk of which is spent in theinspectmodule determining whether something is a "real" function, which again would be disruptive to change.That's just over 100% accounted for, so I don't think I can do much better from here. :)