Make Coord abstract

[stephenworsley]

This addresses #3410 by making __init__ in Coord an abstractmethod.

[stephenworsley]

One thing to bear in mind about this approach is that the __doc__ of AuxCoord.__init__ will be overwritten. Though inspect.getdoc(AuxCoord.__init__) still inherits the docstring properly, so this may be enough for things like Sphinx.

[stephenworsley]

It might also make sense to make _DimensionalMetadata properly abstract. It currently derives from ABCMeta but has no abstract methods, so it can still be instantiated.

[bjlittle]

@stephenworsley I played with a couple of options for the AuxCoord doc-string.

from functools import wraps
...
class AuxCoord(Coord):
    @wraps(Coord.__init__)
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)

Works fine, but has a minor performance overhead:

timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[10]: 11.958185139999841
timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[11]: 11.816572594994796
timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[12]: 11.72259231901262
timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[13]: 11.927083795992075

This is compared with master:

timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[5]: 11.874402218993055
timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[6]: 11.801809267999488
timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[7]: 11.369123923010193
timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[8]: 11.455261625000276

An alternative is:

class AuxCoord(Coord):
    def __init__(self, *args, **kwargs):
        self.__doc__ = super().__init__.__doc__
        super().__init__(*args, **kwargs)

which has the following behaviour:

timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[7]: 12.073399219996645
timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[8]: 12.17960434599081
timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[9]: 12.184167061001062
timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[10]: 12.351267233985709

Or indeed, there is this:

class AuxCoord(Coord):
    def __init__(self, *args, **kwargs):
        self.__doc__ = Coord.__init__.__doc__
        super().__init__(*args, **kwargs)

which has the following behaviour:

timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[10]: 11.896369188994868
timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[11]: 12.091242638998665
timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[12]: 12.14877037498809
timeit('AuxCoord(1)', setup='from iris.coords import AuxCoord')
Out[13]: 12.003538096003467

Note that timeit is executing the AuxCoord(1) statement 1,000,000 times by default.

Given the above, I'd be happy to opt for the first option using @wraps

[bjlittle]

@stephenworsley Could you please make an issue to cover _DimensionalMetadata, and reference this PR in it. Thanks 👍

[stephenworsley]

It looks like using @wraps causes @abstractmethod to carry over as well.

[bjlittle]

@stephenworsley Very nice indeed 😉

For reference, see https://docs.python.org/3.7/library/functools.html#functools.update_wrapper