On Thu, May 20, 2010 at 13:12 +0200, Hans Meine wrote: > On Thursday 20 May 2010 12:49:31 holger krekel wrote: > > On Thu, May 20, 2010 at 12:43 +0200, Hans Meine wrote: > > > [...] py.test fails with > > > > > > E ImportError: No module named something.test_foo > > > > > > [...] > > > > > > Obviously, this is a bug that should be fixed, but I would already happy > > > if you could tell me a workaround (i.e. different invocation of py.test, > > > without changing the directory name of course). > > > > The error message is a bit obscure but scrolling a bit up you see > > that py.test tries to import 'io.something.test_foo' > > Can I circumvent this? (IIRC, for instance nosetests differentiates between > full filenames and importable module names.)
Could you elaborate on nose's logic a bit? > > - i.e. you have created a package 'io' and that overlaps with the standard > > library's 'io' package. > > Ah, bad. But I won't change that now, after using the name myself for > several > years already. Who needs stdlib's io anyway.. ;-p Hehe :) However, it only works as long as no pluguin and no code you are using depends on the standard 'io' module. py.test itself uses a fallback in case it can't "from io import StringIO" but in general a user package named 'io' will lead to breakage, latest when using Python3. > Another workaround is to prepend my main package directory to PYTHONPATH > (tested, works -> fine). You could also try to not have __init__.py files in test directories - as long as all tests have unique names that's fine (and py.test should warn you if there is a conflict). > Thanks for the explanation, > Hans have fun, holger > PS: No need to CC me anymore, the list turned out to be subscribers-only. :-/ P.S.: yeah, sorry. otherwise a lot of spam-sorting would be needed to let the occassional post through. _______________________________________________ py-dev mailing list py-dev@codespeak.net http://codespeak.net/mailman/listinfo/py-dev
