Consider removing `allow_any_instance_of` from the controller spec scaffold templates

[cupakromer]

Many of us believe that allow_any_instance_of is a smell. While it can be useful for legacy applications I do not feel we should be promoting it's use.

[JonRowe]

However as #1010 proves this isn't necessarily the easiest replacement.

[cupakromer]

Based on the comments in that issue, it seems the sentiment is to move away from providing a mockist approach for the controller spec scaffold. Personally, I'm mixed on this at the moment. I do tend to favor that approach, however, in light of recent debates on the issue I need to consider all the options a bit.

I'm very open to hearing other's opinions.

[myronmarston]

To be clear, I have I think that a mockist approach to controller testing is fine. I just think it needs to drive you creating a domain-specific API and is a poor fit foot generated specs (which can only use built in AR APIs). Also, there IMO there is greater potential danger from overmocking than from undermocking and using a poor example of mocking in the generators can lead people to overmocking.

[xaviershay]

Also, there IMO there is greater potential danger from overmocking than from undermocking and using a poor example of mocking in the generators can lead people to overmocking.

This.

[cupakromer]

I'm not sure I completely agree with not stubbing new. However, I have seen a bit of a tendency for over mocking recently. So I'm fine with dropping mocking directly and just going with a more integration approach.

Something else to note: the generated controller specs do not work properly when respond_with is used. This is very confusing and difficult to debug. The root cause is because the internal code for respond_with checks for the existence of errors on the provided object. If errors are present, it re-renders the normal templates, otherwise, it redirects.

Also, recently I've seen an influx in newer rails users who have started using InheritedResources; which uses respond_with under the hood. So they all hit these strange spec failures, but the app runs fine.

[JonRowe]

Also, recently I've seen an influx in newer rails users who have started using InheritedResources; which uses respond_with under the hood. So they all hit these strange spec failures, but the app runs fine.

This is also an argument for less mocking, as they shouldn't be mocking things they don't own ;)

[xaviershay]

I have an idea for doing this that I'm going to try tonight ... if that doesn't work I'm going to lobby to make this not release blocking.

[xaviershay]

side note: I've never used these generators, do you? Or is it for new people?

[JonRowe]

I know people do, I don't fathom why once you've done it a few times though :/