Suppress Redundant Configurable Validation

[rmorshea]

Closes #390

Summary

Configurables currently make a redundant series of setattr calls on themselves to override config values with those that have been explicitly hard-coded in the constructor. In the past this was required due to the timing of descriptor initialization via instance_init, which occurred in HasTraits.__init__. Now though, descriptor initialization occurs in HasTraits.__new__ via setup_instance. Thus we can remove the redundant setattr in Configurable.__init__ and simply use super().__init__(**kwargs) instead.

Todo

1. I'm encountering a test failure complaining that a warning for a mistyped section did not occur. However, pytest shows that stderr received a call with the exact warning required by the test. This is specific to Python 3 since it requires TestCase.assertLogs.
2. Add a test case.

cc: @jasongrout

[minrk]

While you're at it, can you make sure there's a test case for what this is supposed to resolve?

class C(Configurable):
    x = Unicode(config=True)

cfg = Config()
cfg.C.x = 'from config'

c = C(x='from init')
assert c.x == 'from init'

[rmorshea]

@minrk sure. Any thoughts on the test failure I'm getting here? I don't know what's causing it.

[minrk]

The test is verifying that constructor keyword-args (in this case 'log') are loaded prior to config. This PR explicitly removes that behavior, which isn't quit right.

We have these requirements:

1. traits passed to the constructor load before config
2. traits passed to the constructor have higher priority than config

This is handled somewhat clumsily in master by loading all traits passed as kwargs twice - before and after config (hence #390). I've updated the PR to only re-initialize those kwargs that are actually overridden by config, which should make it a smaller change and still fix #390.

[minrk]

TODO: remove print before merge

[rmorshea]

@minrk creative fix! 👍