I would like to investigate how we can fix api-reference build errors.

------

I have created a travis-ci configuration file which can be used as a reference for running tests. It somehow copies the builders observed on buildbot.twistedmatrix.com

A sample build result can be found here:

Code changes are here. If you find them useful I can refine them into a public patch.

Only python 2.7 and 3.3 builders are enabled... but 2.6 builders can also be enabled.

py-select-gc tests are not enabled since they are very, very slow.

Running all tests on a single process could take about 1 hour.

I guess that running only tests with default reactor and then pyflakes/pydoctor on changed files should cover most errors. 

A developer should run at least the following tests before sending a patch for review:

 * trial with default reactor

 * pyflakes for changed files

 * pydoctor for changed files

If documentation was changed he/she should also run documentation tests.

----

Running pydoctor on my computer takes about 3 minutes... a bit long for my taste.

I hope that pydoctor can be run on a independent python files so that I can create

a separate linter test which will run pyflakes and pydoctor only on changes files.

This should provide a big improvements in speed.

I used travis-ci as it provides  an isolated / reference test environment. Once the configuration is ok, it can be translated into tox [1],  private buildbot [2] or custom script to allow developers to run test on their computers. 

Thanks!

[1] https://github.com/msabramo/python-panci

[2] https://github.com/isotoma/buildbot_travis