Integration tests #85
Conversation
This an extra effort to deal with the lxml-backend, but it's more important to define a consistent behaviour. It's noteworthy that the keys of the attributes could also consist of two-value tuples with namespace and local name.
This is a revert of a2c7d55 that would lead to crashes when documents have a funky mixture of several default namespaces. Due to that removal the hashing of Namespace objects isn't needed anymore and get removed as well. Where the funk is, the element wrapper cache comes around to bang. Bless the funk when the effing cache is gone.
as lxml's implementation doesn't represent a default namespace in etree._Attrib (that's the backend data structure of TagAttributes).
Explicitly declared namespace weren't considered before.
funkyfuture
added
bug
|
looks awesome! i'd need some time to set it up and check it out but i dont see why this shouldnt be merged. lets just set required coverage down to 96% to make the ci stop complaining! |
|
as an aside, i very much enjoy modeling a dsl for integration test scenarios (as you already did) and then extract scenario specs out of the test suite code and into yaml files early on. really adds to developing speed and overall perceived comfort in my experience. maybe keep something like this in mind for consideration sometime later? |
note: python3.7 is the only environment in which 97% is not reached currently.
since it only affects Python 3.7:
it's okay for now though. |
i can't really follow. feel free to open an issue with a practice oriented sketch. the next addition test-wise will be the comparison of a pretty serialisat in conjunction with collapsed whitespace. |
personally, I can make my peace with a 96% coverage minimum, but don't let me stop you applying the gate selectively for all versions except 3.7 because I have no idea how one would do that. |
well, it's a regression. but getting rid of the lxml backend will result in something very different code-volume-wise and 99% will be the target then. and these tests increase edge case coverage a lot, that's way more important than a stupid number.
something with cli arguments i guess. but it's simple enough now. |
|
thanks for the feedback. i hope the other branches rebased themselves when i arrive home. |
This adds an initial framework for tests against larger sets of real-world corpora.
A mixed bag of fixes for discovered bugs comes with it.
Let me know if we should look at it together in person.
This PR has higher priority than #82. I still need to find the opportunity to create new graphics for that.
I'll continue to work on serialization based on the additions here.