test: fix py2 inconsistency with regex ID

[Exirel]

For some reason assertions about compiled regex fail with Python 2.

We don't know why, but at least now we make assertions about the patterns, and not the compiled regex objects. This should prevent the false-negative on Travis CI (I hope).

[Exirel]

Travis seems happy about these tests with Python 2.7.

[dgw]

Well, this will do to get us going again.

We have several PR builds that failed because of this problem, but not all. I'm at a complete loss as to why this is even a thing, to the point where I'm about ready to bring it up in the Travis-CI forums and/or ask for people to look at it in one of freenode's Python channels. Only the Travis environment seems able to trigger this particular problem.

Lest I forget to bring it up on IRC tomorrow (technically later today), I'll mention here: Why does @url add a list of compiled patterns to the callable, but @rule adds the patterns without compiling them? I think we should take a long look at why @url works the way it does, which is different from every other decorator, which attach only simple strings (or the occasional bool) and let the loader code turn those strings into usable regex objects. Probably only the loader should do any pattern compilation.

Separately, maybe we should consider using sets instead of lists?

>>> ['first', 'second'] == ['second', 'first']
False
>>> set(['first', 'second']) == set(['second', 'first'])
True

Wouldn't solve the weird issue that we had to deal with here, but maybe worth a long-term look anyway for other reasons.

That's enough issue-hijacking for now. 😂

[dgw]

@Exirel Thanks for putting this together so quickly! 👌