[Distutils] Distribute testing

ssteinerX@gmail.com ssteinerx at gmail.com
Sat Dec 19 18:33:40 CET 2009


	We recently talked about incorporating the testing scheme I started in distutils-buildbot directly into the release path for Distutils to avoid issues like the ones with the most recent release.

	I am putting this out on the list for comments; I'm NOT interested in bikeshedding this to death.  Anyone else is welcome to do their own project if they feel strongly enough about it.  I'm going to take opinions into account but I want to have the tool built before the next release, not the next millennium release of Python.

	The way I started with the project in distutils-buildbot (http://bitbucket.org/tarek/distutils-buildbot/ smoke-and-mirrors branch, bin/smoke2.py, bin/smoke2.cfg, and bin/smoke_base_products.cfg) was to assume a brand new system e.g. a newly initialized Cloud Server and to build out everything necessary to run the tests under Python 2.4, 2.5, 2.6, 2.7dev, 3.1, and 3.2dev.  

	While this approach will be fine when we move it to building on-demand buildbots, it is much too time consuming both in development time and real-time processing to use for this release cycle.  It would be much better (for now) to just make sure you're running it on a machine that can do the job.

	The main goal is to come up with a simple way to run as many full-scope tests i.e. from install through full test-suite on as many large products, with as many versions of Python as practical.

	What we had originally started was to build and run the test suites for (spec'd in bin/smoke_base_products.cfg):


	I had stopped where it reads the configuration files and knows on what versions of Python to test which products.

	At this point, I'd like to move this into Distribute proper and get it into the "pre-release" test process.


	First, we need to decide where to put the testing sub-project.  I'm asuming it can just live in the /tests subdirectory, maybe with a subdir to hold this whole project as its own module.

	Here's what I'd like to see it do:
	Run pre-test checks i.e. verify that the machine is capable of running the tests:

		1>	Assume that `python2.4`, `python2.5`, `python2.6`, `python3.1` will invoke a properly 	set up Python interpreter.  This is beyond the scope of this project; might be taken up at a later time.

		2>	Create a simple way to verify installed utilities before proceeding with tests (anyone have one of these around?).
		Something like:
			ASSERT(`nosetests -V` == `nosetests version 0.11.1`)

		3>	Allow a certain amount of corrective action to be able to be taken to satisfy the assertions above (like, install nose version 0.1.11 from PyPi, for example) for each supported version of Python.

		for each product to be tested:
			for each Python version this product is to be tested on:
				install the product
				run its test suite
				record any failures

	Pretty simple, but will avoid any of the sorts of issues we just had with the most recent release, is very easy to add new products to (just add a configuration section), and can be easily moved out to the buildbots when it does what we want.

	Comments etc. gladly accepted as long as there are no bikeshedding tools (hammers, paint-brushes, Erlang) in hand.


aka/Steve Steiner


More information about the Distutils-SIG mailing list