Reputation: 4477
The doctest of the following (nonsense) Python module fails:
"""
>>> L = []
>>> if True:
... append_to(L) # XXX
>>> L
[1]
"""
def append_to(L):
L.append(1)
class A(object):
pass
return A()
import doctest; doctest.testmod()
This is because the output after the line marked XXX is <__main__.A object at ...>
(which is returned by append_to
). Of course, I could put this output directly after the line marked XXX but in my case this would distract the reader from what shall be actually tested, namely the side effect of the function append_to
. So how can I suppress that output or how can I ignore it. I tried it with:
"""
>>> L = []
>>> if True:
... append_to(L) # doctest: +ELLIPSIS
...
>>> L
[1]
"""
def append_to(L):
L.append(1)
class A(object):
pass
return A()
import doctest; doctest.testmod()
However, this yields a ValueError: line 4 of the docstring for __main__ has inconsistent leading whitespace: ' ...'
.
What I don't want to do is to change the line append_to(L)
to something like _ = append_to(L)
which would suppress the output, because the doctest is for documentation purposes and to show the reader how the module is supposed to be used.
(In the case being documented, append_to
should be used statement-like and not like a function. Writing _ = append_to(L)
would deviate the reader from this.)
Upvotes: 6
Views: 5146
Reputation: 330
I ended up in this question because I need the +IGNORE_RESULT behavior but for running doc tests on sphinx documentation. The replace
answer posted by @intuited does not work in that case and there was no code for the mentioned "proper" solution. Therefore I post what I ended up with.
As a direct answer to the question the solution would be:
__doc__ = """
>>> L = []
>>> if True:
... append_to(L) # doctest: +IGNORE_RESULT
>>> L
[1]
"""
def append_to(L):
L.append(1)
class A(object):
pass
return A()
if __name__ == "__main__":
import doctest
IGNORE_RESULT = doctest.register_optionflag('IGNORE_RESULT')
OutputChecker = doctest.OutputChecker
class CustomOutputChecker(OutputChecker):
def check_output(self, want, got, optionflags):
if IGNORE_RESULT & optionflags:
return True
return OutputChecker.check_output(self, want, got, optionflags)
doctest.OutputChecker = CustomOutputChecker
doctest.testmod()
For my particular need of doc testing sphinx documentation add to the conf.py
file:
import doctest
IGNORE_RESULT = doctest.register_optionflag('IGNORE_RESULT')
OutputChecker = doctest.OutputChecker
class CustomOutputChecker(OutputChecker):
def check_output(self, want, got, optionflags):
if IGNORE_RESULT & optionflags:
return True
return OutputChecker.check_output(self, want, got, optionflags)
doctest.OutputChecker = CustomOutputChecker
Then test with a command such as:
sphinx-build -M doctest source_dir build_dir source_dir/file.rst
Upvotes: 2
Reputation: 24034
rewrite: This actually does work now; I realized that the "doctest" I had written earlier was actually not being parsed as the module docstring, so the test wasn't passing: it was just not being run.
I made sure to double-check this one.
__doc__ = """
>>> L = []
>>> if True:
... append_to(L) # doctest: +IGNORE_RESULT
>>> L
[1]
""".replace('+IGNORE_RESULT', '+ELLIPSIS\n<...>')
def append_to(L):
L.append(1)
class A(object):
pass
return A()
I'm not sure if this qualifies as more readable or not. Note that there's nothing special about <...>
: it will only work if the actual return value has that form, as it does in this case (i.e. it's <module.A object at 0x...>
). The ELLIPSIS option makes ...
"match any substring in the actual output" ¹. So I don't think there's a way to get it to match the entirety of the output.
update: To do this the "proper" way, it looks like you'd want to call doctest.register_optionflag('IGNORE_RESULT')
, subclass doctest.OptionChecker
, and arrange for an instance of that subclass to be used by the doctest. Presumably this means that running your doctest via $ python -m doctest your_module.py
is not an option.
Upvotes: 7
Reputation: 57474
Please try to give fully self-contained, runnable code; even when you're demonstrating a problem, the code should run on its own to reproduce the problem, so solutions can copy the code directly for demonstrating the answer.
I don't know of a clean solution to this, and I've hit it before; it seems a side-effect of the fuzzy (more bluntly: sloppy) test definitions doctests provide. A workaround is to remember that you can define functions within doctests, so you can contain a whole test as a single function rather than as its individual statements.
def append_to(l):
"""
>>> L = []
>>> def test():
... if True:
... append_to(L) # XXX
>>> test()
>>> L
[1]
>>> def test():
... L = []
... if True:
... append_to(L) # XXX
... return L
>>> test()
[1]
"""
l.append(1)
return object()
if __name__ == "__main__":
import doctest
doctest.testmod()
Upvotes: -2