[Mikhail Arkhipov (MikhailArkhipov)]

Undo #1686. Basically fork right before the PR then cherry pick changes on top. Also removes caching services.

Fixes #1988 and probably some others - to be verified.
Some tests became less stable due to old loop resolution issues.

[Philip May (PhilipMay)]

This is great. Many thanks. I hope #1988 can be fixed soon.

[Jake Bailey (jakebailey)]

There seem to be some regressions in the __all__ handling. Might be something that was over-reverted, since I know the loop dependency stuff was using it to some extent.

[Mikhail Arkhipov (MikhailArkhipov)]

Minimized differences and ported fixes from PRs that were not taken due to compatibility problems with old code.

[Mikhail Arkhipov (MikhailArkhipov)]

What happened to __all__? Yes, loop resolution can definitely affect that. Theoretically it is possible to build off commit right before the loop one and compare if behavior is the same.

[Jake Bailey (jakebailey)]

If you run the whole test suite, many of the __all__ related tests fail.

[Mikhail Arkhipov (MikhailArkhipov)]

Ran full stack, mine's are clean except one import ordering (it uses os and path which have loops, I guess I need to update the test).

[Jake Bailey (jakebailey)]

Those LS results appear to be cached; if I run them I get:

[Mikhail Arkhipov (MikhailArkhipov)]

Those are not cached, they turn this way when you re-run individual tests. HoverSpanCheck was a known problem, typically passes on re-run. In MissingImports SymbolOrdering2 failed, very long test. Fixed.

What fails for you in highlight and imports? Are you on 3.8 or higher?

[Jake Bailey (jakebailey)]

Well, this is embarrassing. I'm still on your undo branch. I didn't see that this PR was being made from a branch on the repo itself rather than a fork. I'll retest.

[Jake Bailey (jakebailey)]

Yeah, now that I run the actual code it passes... 😄

I'll spot check this on some real code.

[Mikhail Arkhipov (MikhailArkhipov)]

This should be off my fork, u2 branch

[Jake Bailey (jakebailey)]

Checked it with a few projects, things appear to work as they do now (just without the loop messages, of course).

[Philip May (PhilipMay)]

Is these anything that can be done on Code comment level so this bug will not be introduced again? Are there regression tests possible?