fix broken ui tests

[Icxolu]

While working on #5741 I noticed that some of our ui tests are broken, indicating that they are not running in CI. Looking in more detail it seem we either run without any features, which means ui tests are not running since they require macros or we are running full. Basically all tests that require the absence of a feature is not running in CI.

Quite a few of are gated on #[cfg(not(feature = "experimental-inspect"))] as the output changes with the feature enabled. We could invert the cfg, but this would decrease the readability of the output quite a bit. And for some of them this would not work, as they specifically test errors that we emit when a feature is not enabled.

Currently untested are (as far as I can tell):

• invalid_property_args
• invalid_pyclass_args
• invalid_pyfunction_argument
• invalid_async
• missing_intopy
• invalid_cancel_handle
• duplicate_pymodule_submodule

I already fixed a few simple ones, but I'm not sure how to best handle the rest. We could remove experimental-inspect and experimental-async from full and have separate runs, but that would probably increase CI time significantly and also feels more like a band aid rather than a fix. Maybe we should have a separate job that just runs test_compile_error with more feature combinations? An additional run with just macros would fix quite a few of these, but not all I believe. Ideas welcome.

[Tpt]

Thank you!

Aside: I opened #5913 going into this direction a while back but running all the tests and not only the UI tests.

[mejrs]

Wouldnt switching to ui_test help with this?

[davidhewitt]

Wouldnt switching to ui_test help with this?

I just updated #5863, let's see if that now builds green 👀

[Icxolu]

I think as long as we don't change the runs, it will have the same problem, no? A lot of these #5863 conditionally skips, but that condition is currently always true.