Original Launchpad bug 521612: https://bugs.launchpad.net/ipython/+bug/521612
Reported by: fdo.perez (Fernando Perez).
No idea why, but querying np.ma.compress with ? or %pdoc double-prints the docstring, with the second copy indented.
See this numpy thread for some context:
That object has a runtime-generated docstring, but I checked and its doc is OK (no duplication there). So the bug is definitely ours.
The object implements a .getdoc() method. IPython.core.oinspect.getdoc() reckons that, if an object has a .getdoc() method as well as a normal docstring, the result of calling it should be treated as extra information, and concatenated together.
Are there actually any cases where .getdoc() adds info to the normal docstring, or can we simplify it to only check .getdoc() if it doesn't get a normal docstring. Otherwise, one possible way round this is to use textwrap.dedent, then compare the two, and only show both if they differ. Or perhaps, if they're both over a certain length, only show one.
@takluyver, agreed: we should probably change our behavior and only call getdoc if we didn't get a docstring. Should be an easy fix.
I wonder if we should possibly prefer .getdoc() over a normal docstring, on the basis that it might offer more customised information? I don't know how many or what objects implement getdoc.
That's a good point. I think it's fairly rare, so anyone implementing it must have had a good reason to do it, and we might as well honor it. Since using both is definitely a bug, I agree we could go with letting getdoc take precedence, unless what it returns is not a string, case in which we fall back to __doc__.
If object has a getdoc() method, override its normal docstring.