fix union default - no need for builtin dynamic default

[rmorshea]

Closes #344

No more need for dynamic default in Union as a result of #332.

Also fixes a test case that was inheriting from HasTraits instead of TestCase, which masked #344.

[jakevdp]

Awesome – thanks!

[SylvainCorlay]

@rmorshea the reversion of the trait types validation is a backward incompatible change (we rely on the current order in bqplot).

What is the reason for the inversion?

[rmorshea]

@SylvainCorlay, since validation prefers the first given TraitType, it seemed reasonable to prefer the default value of the first given TraitType as well.

I'll put that in a separate PR though since I didn't think about compatibility when including that here.

ignore deleted comments - I second guessed myself

[jakevdp]

If validation order is an important part of the API, it might be worth putting in a unit test that will catch if that ever breaks.

[SylvainCorlay]

Ah now I understand. Sorry I was mistaken. I agree that it would be more logical for it to be reversed.

[SylvainCorlay]

If validation order is an important part of the API, it might be worth putting in a unit test that will catch if that ever breaks.

It is just that the first trait type that validate is the type that will be used.

So if a given value works for two of the types, it might be validated as one or the other depending of the order that is used.

Then, in the case of e.g. widgets, the serialization might be different for these two types.

[rmorshea]

it might be worth putting in a unit test that will catch if [union validation order] ever breaks.

#347

[rmorshea]

Found a more fundamental bug. Will rework this shortly.