ElementMaker as class variables are interpreted as abstract methods

[Hynek Schlawack]

So this is kind of a fun one:

>>> import abc

>>> from lxml.builder import ElementMaker

>>> EM = ElementMaker()

>>> EM.foo()
<Element foo at 0x11066a388>

>>> class C(metaclass=abc.ABCMeta):
... x = EM

>>> C()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: Can't instantiate abstract class C with abstract methods x
Can't instantiate abstract class C with abstract methods x

Details:

Python : sys.version_info(major=3, minor=6, micro=1, releaselevel='final', serial=0)
lxml.etree : (3, 8, 0, 0)
libxml used : (2, 9, 4)
libxml compiled : (2, 9, 4)
libxslt used : (1, 1, 29)
libxslt compiled : (1, 1, 29)

Let me know if you need more information. :)

[scoder (scoder)]

Yes, a lovely one. Thanks for sharing. ;)

The reason is in ABCMeta.__new__():

        abstracts = {name
                     for name, value in namespace.items()
                     if getattr(value, "__isabstractmethod__", False)}

Calling getattr() on the ElementMaker will not raise an AttributeError, but return a partial(ElementMaker) that creates "__isabstractmethod__" tags. That partial object resolves to true in the code abode and ABCMeta considers that enough to assume it's an actual abstract method. :)

I'd rather not restrict the (valid) tag names that ElementMaker is allowed to create - imagine a user wants to write an XML serialiser for Python class definitions. That could actually benefit from an XML tag name "__isabstractmethod__".

In short - I'm not sure if this is worth fixing. What do you think?

[Hynek Schlawack (hyneks)]

Hm I would suggest to document it and leave it be.

Ivan Levkivskyi mentioned to me that he's rewriting ABCMeta in C so the bug should dissolve eventually: https://twitter.com/ILevkivskyi/status/889880285583724544

But ATM it's confusing af and Google brings up nothing.