Robert Bradshaw wrote: > On Oct 1, 2009, at 5:06 AM, Dag Sverre Seljebotn wrote: > >> Stefan Behnel wrote: >>> Dag Sverre Seljebotn wrote: >>> >>>> I now have code sitting on my drive for a doctesthack directive. >>>> What it >>>> does is work around the limitations of the doctest module by >>>> automatically creating a global __test__ dictionary mapping >>>> descriptive >>>> names to function and method docstrings. >>>> >>>> 1) Can I push it? >>>> 2) Can I make it the default? It never does anything if a __test__ >>>> already exists, so it should be fully backwards compatible. The only >>>> downside I can see is a little bit of extra memory used. >>>> >>>> Line numbers are embedded into the test names (the keys in >>>> __dict__). >>>> >>>> Note: This does not solve the problem of doctest being unable to >>>> print >>>> line numbers for class docstrings. Also the test procedure ends up >>>> flowing a little bit different from the Python module (as each >>>> function >>>> is a seperate testcase, and not simply part of the module test). >>>> As the >>>> name says, it is a hack. >>>> >>> Now that I know what kind of problems this does *not* fix and what >>> new >>> problems it introduces, what about some more detail what kind of >>> problem >>> this actually fixes? >>> >> What it fixes is that I can take a simple pyx file like this: >> >> def func(x): >> """ >>>>> func(3) >> """ >> ... >> >> and then test it like I would a Python module: >> >> python -c 'import doctest; import mymod; doctest.testmod(mymod)' >> >> doctest does not pick up the function docstring automatically as it >> reports negative to inspect.isfunction, however putting those >> functions >> in the __test__ dict works around it. >> >> Arguments in favor: >> - Lowers the barrier. It is the kind of thing which took me by >> surprise >> when I first met it; letting this be the default could save half an >> hour >> for every new Cython user. >> - This also allows very easily writing function-wise doctests in the >> Cython test suite >> >> An approach which requires a custom script for doctesting Cython files >> is here: >> >> http://wiki.cython.org/FAQ#HowcanIrundoctestsinCythoncode. >> 28pyxfiles.29.3F >> >> But I happen to like this approach better as it lower the barriers and >> makes peoples test scripts simpler without Cython-specific logic. > > I actually wasn't aware that this was such an issue, as Sage has its > own doctesting framework (e.g. it need to take care of preprocessing, > optional doctests, etc.) I think the right thing to do is convince > them to fix the doctest module. Of course, that won't help past and > current version of Python. That being said, this looks like a sound > solution, and I'm in favor of it going in, and probably even being on > by default.
It's possible to rather easily fix doctest to pick up the functions, but to my knowledge there's no way for doctest to retrieve the line numbers -- which has made me reluctant to submit anything. My hack works around it by embedding that information in the test name, so it says something like: Doctest failed at line ?: somemodule.__tests__.somefunction (line 35) I guess we could try to agree with them on a new convention though -- something like: Any module can optionally provide the dict __testlinenumbers__, mapping functions and classes to integers. This dict, if present, is checked before any of the conventional methods for retrieving line numbers. -- Dag Sverre _______________________________________________ Cython-dev mailing list [email protected] http://codespeak.net/mailman/listinfo/cython-dev
