Safer argument parsing (attempt 2)

[bobjflong]

I couldn't stomach the merge conflicts in #220 (I was working with an old fork).

An issue I've seen crop up is that if a global key is set (which most of your docs use), and then client API keys are used (eg. if using OAuth), then unexpected results can happen. This includes loading customer data from the parent global key.

> require 'stripe'

> Stripe.api_key = "sk_test_foo"
=> "sk_test_foo"

> Stripe::Customer.all
# => loads customers from the global key

> Stripe::Customer.all({}, { api_key: nil })
# => ALSO loads customers from the global key!!

With my changes:

> Stripe::Customer.all({}, { api_key: nil })
TypeError: api_key must be a String

I've also added normalize_opts usage to a bunch of places. I'm sure not all of these are necessary. I've added tests for the static self.foo ones. It is a slight flaw I think that the current "normalization" of opts is spread across lots of resources, when it should be enforced centrally. I have some ideas about how to do that. But I think with a look over, this could be a good step forward and prevent some nasty errors.

[russelldavis]

I couldn't stomach the merge conflicts in #220 (I was working with an old fork).

Ugh, that sucks. Github should really get with the times and show the diff as what the final merge will look like, a la Bitbucket

[russelldavis]

I've also added normalize_opts usage to a bunch of places.

The request method already calls normalize_opts on what's passed in, so I think some of those additions are unnecessary. Not sure off hand about the other ones, but if you think they are, can you move them to a separate PR? (I like to keep each PR as focused as possible.)

[russelldavis]

(Also, note that the Travis tests are currently failing.)

[bobjflong]

@russelldavis I backed out most of those normalize_opts.

I actually think that spec failing is flakiness in the spec itself. Ruby 1.8 doesn't preserve hash ordering so the resulting query string can be different. I've patched the spec here.

[russelldavis]

This looks like another unnecessary call to normalize_opts, since super already calls it.

(After this, I think the PR will be ready to merge -- sorry for the delay.)

[bobjflong]

cool, it's gone

[russelldavis]

@bobjflong thanks for pushing this through!