Once upon a time [1] I had propagated the idea of refactoring marvin to make test case writing simpler. At the time, there weren't enough people writing tests using marvin however. Now as focus on testing has become much more important for the stability of our releases I would like to bring back the discussion and to review the refactoring of marvin which I've been doing in the marvin_refactor branch.
The key goal of this refactor was to simplify test case writing. In doing so I've transformed the library from its brittle hand-written nature to a completely auto-generated set of libraries. In that sense, marvin is much closer to cloudmonkey now. The two important changes in this refactor are: 1. data represented in an object-oriented fashion presented as factories 2. test case writing using entities and their operations rather than a sequence of disconnected API calls. To see the full nature of this proposal I've updated the spec I put up on the wiki: https://cwiki.apache.org/confluence/x/RI3lAQ For a quick comparison I wrote a test for the VPC vm's lifecycle in tools/marvin/marvin/test/test_vpc_life_cycle.py which one can compare with the existing tests for vpc under test/integration/component/test_vpc_vm_life_cycle.py These changes being 'architectural' so to speak and in a way disruptive even I would like to merge this at the beginning of the upcoming cloudstack release. This is only a small part of a larger change for marvin which will be moving to a more BDD like implementation [2] where tests are written using a gherkin-like language. But that will come later. I've also tried to disconnect marvin from depending on CloudStack's build and repo. This will help split marvin from CloudStack which I will discuss in a seperate thread. [1] http://markmail.org/message/4tsscn6zvtfsskuj [2] http://pythonhosted.org/behave/ -- Prasanna., ------------------------ Powered by BigRock.com
