Benchmark of transfer_nested_constants option.

[xaviershay]

Shows a non-trivial performance impact, though low absolute time for the common case.

I'm inclined to leave the default value as false, reconsidering if it happens to cause a lot of user confusion.

WDYT @myronmarston?

Resolves #395.

[xaviershay]

cc @pat as an FYI because you were involved in the original discussion.

[coveralls]

Coverage decreased (-0%) when pulling 8baa9b8 on xaviershay:issue-395 into c8139a9 on rspec:master.

[myronmarston]

What do you think about adding a config option that specifies a default for transfer_nested_constants?

[pat]

@xaviershay appreciate the cc, though you both know this topic far better than I do, so I'm happy to defer to your judgement.

[JonRowe]

I think given the performance difference its worth defaulting as false and then adding a RSpec.configuration option for those who feel otherwise.

[xaviershay]

Added configuration option.

[coveralls]

Coverage decreased (-0%) when pulling 8140de9 on xaviershay:issue-395 into c8139a9 on rspec:master.

[myronmarston]

You can use the "with isolated configuration" shared context to manage the configuration isolation. Just include the context in your example group:

describe "the example group" do
  include_context "with isolated configuration"

  # ...
end

[myronmarston]

This looks great. Nice and simple. Just one suggestion that will cleanup the spec a bit.

[xaviershay]

@myronmarston oh neat. I added a small refactoring of the verifying double specs as well to use it too.

[coveralls]

Coverage decreased (-0%) when pulling 26dbe69 on xaviershay:issue-395 into 80d80d9 on rspec:master.

[myronmarston]

This looks good. Nice to see that you were able to cleanup some other specs, too :).

I just kicked the errored builds (I suspect github's DDOS problems today are causing build issues).

[xaviershay]

(it's green now too.)

[myronmarston]

Maybe I've been doing it for too long, but it doesn't bother me that much. It is unambiguous, in that there aren't any other concept it could be referencing. You'd likely have to look up the documentation anyway, since it's not an obvious concept. Also, having it exactly match the option name is a nice property, even if it doesn't otherwise have the context. Two names for the same thing is possibly more confusing.

Works for me.