direct access for performance and typechecking

[dimbleby]

Revisiting #2149, which seems to have stalled or maybe got into some merge conflict horror.

As it goes, I can't reproduce anything like the performance gains reported over there, which is a shame because I'd really have liked to!

• in the example of the original report, resolution is simply much faster in the first place. It's hard to tell whether that's because poetry has changed, or because the web of dependencies has changed such that this particular example is easier to resolve. I am seeing times nearer 2 seconds than 200 seconds.
• in some examples I do have with slower resolves, I just don't see meaningful gains.

However: this surely can't hurt performance...

Even if this doesn't turn out to be a big performance win though, I think that it is still good discipline. In particular, removing the __setattr__() and __getattr_() tightens up the typechecking: we no longer have to wait to runtime to find that we have gone wrong.

Actually I'd have been a little scared to make this change without good mypy coverage: but that and the test suite makes me pretty confident that this is safe.

[dimbleby]

just doing a review of my open MRs and I think this one is good to merge. (But also it's not urgent; it can wait until after 1.2.0)

[github-actions]

This pull request has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.