1. Clone the pypy repository
2. get the latest translated PyPy from the nightly build
3. run the failing test from inside a clone with a pre-built PyPy, e.g.:
4. pick something you think you can fix
5. add an app level test that highlights the *specific* failure (that will be in somewhere like pypy/objspace/std/test/test_something.py )
6. running app level tests with py.test is fast - so you can see when you fix it. Running the test you have written (on top of CPython) will be a command like:
7. *eventually* run a full translate so you can see that you didn't break pypy and if you really fixed the problem.