0.1x deprecate fixes

[technoweenie]

This fixes a few issues with #1054

First: You can't rescue a current Faraday exception (Faraday::ClientError, for example) with its deprecated error class (Faraday::Error::ClientError). This would let us remove all Faraday::Error::* references for Faraday v0.17.1. We should update them too, so that users don't get a bunch of deprecation warnings they can't fix. Even if that's too risky of a change, I still think this behavior is correct and worth including.

Second, the regex for the class name changed so it could grab the class name of a root level class. Only came up because some of the test constants have no namespace.

-/^#<Class:(\w*::\w*)>$/
+/^#<Class:(\w{1}[\w:]*)>$/

Finally, I copied some of the tests in spec/faraday/error_spec.rb to spec/faraday/deprecate_spec.rb, so that those things are still tested once the Faraday::Error::* stuff is all removed for good. This raised some problems with the CustomError and E classes created in error_spec.rb, so I changed those to use Class.new instead.

@BobbyMcWho: I'd really appreciate your thoughts on this :)

[BobbyMcWho]

Overall LGTM, 1 comment on the regex

[olleolleolle]

Curious about the RSpec "not-fail" test cases - the opposite of fail.

Thanks for describing the usage in the description and the tests.

[olleolleolle]

Can we find a way to output a message of success when we pass here?

[technoweenie]

What is a "not-fail" test case? Is that something you see in the results? What would a message of success here look like?

[technoweenie]

I changed those examples to use expect { raise ... }.to raise_error(...) instead. Much shorter, and still tests the correct thing. If I remove the #=== implementation on the inherited meta class that Faraday::DeprecatedClass.proxy_class creates, I get these failures:

  1) Faraday::DeprecatedClass allows rescuing of a current error with a deprecated error
     Failure/Error: expect { raise SampleClass, nil }.to raise_error(SampleDeprecatedClass)

       expected SampleDeprecatedClass, got #<SampleClass: SampleClass> with backtrace:
         # ./spec/faraday/deprecate_spec.rb:44:in `block (3 levels) in <top (required)>'
         # ./spec/faraday/deprecate_spec.rb:44:in `block (2 levels) in <top (required)>'
     # ./spec/faraday/deprecate_spec.rb:44:in `block (2 levels) in <top (required)>'

  2) Faraday::ClientError.initialize maintains backward-compatibility until 1.0 allows rescuing of a current error with a deprecated error
     Failure/Error: expect { raise Faraday::ClientError, nil }.to raise_error(Faraday::Error::ClientError)

       expected Faraday::Error::ClientError, got #<Faraday::ClientError Faraday::ClientError> with backtrace:
         # ./spec/faraday/error_spec.rb:76:in `block (5 levels) in <top (required)>'
         # ./spec/faraday/error_spec.rb:76:in `block (4 levels) in <top (required)>'
     # ./spec/faraday/error_spec.rb:76:in `block (4 levels) in <top (required)>'

[olleolleolle]

I was thinking of "if we get here, assert_equal(true, true, "it works")" kind of thing.

[technoweenie]

Ohh I see. I think the expect syntax works too. I don't think an expect(true).to be_true underneath is necessary.

[olleolleolle]

The raise_error matchers worked out great!

[BobbyMcWho]

Awesome, thanks for catching those cases I missed originally