WidgetTest: Check types of settings

[VesnaT]

Issue

Fixes #20

Description of changes

Add settings checking into tearDown of WidgetTest.
The change causes the following tests (in biolab/orange3) to fail:

• test_owcolor (savesOrange.widgets.data.owcolor.AttrDesc)
• test_oweditdomain (references OWColor)
• test_owfeaturestatistics (saves np.ndarray, fixed in [FIX] FeatureStatistics: Convert selected rows to list orange3#4375)
• test_owpivot (saves enum)
• test_owpreprocess (saves enum)
• test_owpythonscript (saves Orange.widgets.data.owpythonscript.Script)
• test_owrank (fixed by [FIX] TestOWRank: Setting type should not be np.ndarray orange3#4315)
• test_report (saves enum)
• test_owdistancemap (saves range)
• test_owboxplot (saves Orange.data.filter.Values)
• test_owfile (saves RecentPath, but it does not fail, because parent tearDown is not invoked)

Includes

• Code changes
• Tests
• Documentation

[janezd]

What if this PR showed warnings instead of raising errors for now? It would warn that after 2/1/21 they would change into errors, though.

In its current form, we shouldn't merge it. :)

[janezd]

Why is OWFile not on this list although it stores a list of RecentPath?

@ales-erjavec, did I understand correctly that the idea is to avoid pickling settings?

Context settings are stored as instances of ContextSetting. OWFile stores RecentPath instances. OWColor stores (essentially) namedtuples with some properties... Adapting this would burden the widgets with code that would essentially perform serialization of settings. I wonder whether the benefits of having json outweigh the trouble and the extra code for conversion (with its maintenance and potential bugs).

Or are there other reasons for limiting the allowed types for settings?

[ales-erjavec]

did I understand correctly that the idea is to avoid pickling settings?

Yes.

[VesnaT]

Why is OWFile not on this list although it stores a list of RecentPath?

Because TestOWFile overrides tearDown().

[janezd]

Superseeded by #62.