Replace mocking in generated controller specs with integration tests.

[xaviershay]

This was a terrible example and should never have been here. This
replacement isn't great, but at least is not obviously bad.

Unfortunately we cannot provide default implementations of these because
we don't have sufficient information about the data model. Specs that
require this are marked pending.

I gave up on the default approach: it doesn't work at all for models
with no attributes, and is a crapshoot if they do.

@cupakromer @JonRowe @myronmarston

Fixes #1008

[JonRowe]

This gets my 👍 I'd much rather we make people think about their specs than try to blindly use scaffolded code, I'd actually be ok with them failing rather than being skipped when they're not setup, but I think some people will argue that's not very friendly to beginners.

[cupakromer]

I don't have time right now, but I plan on reviewing this tomorrow AM; US EST time.

[myronmarston]

This LGTM, FWIW.

[cupakromer]

Overall, I think it looks pretty good. I want to sit on it just for a bit (no longer than today) to think it over. I would prefer to use pending instead of skip for the attributes. However, this provides some challenges based on how those let variables are used.

Several specs simply use them to create! initial objects. Which causes them to fail, even though they could pass with an empty attribute hash. However, providing an empty attribute hash causes Rails strong params to complain that the params are empty. :-/

[xaviershay]

I used skip because pending would cause a failure in the ones that don't have any assertions yet.

create! should be provided with valid_attributes, the empty hash issue you describe is part of the reason I removed example_valid_attributes.

[cupakromer]

I don't really have a good alternative at the moment. We can always adjust it in dot release since it's only a generator.

ship it