metadata string reduction

[rcomer]

🚀 Pull Request

Description

Suppose I make, for example, a coordinate, and print it

coord = iris.coords.DimCoord(range(3), long_name='foo', units='m')
print(coord)

I get info about what it contains:

DimCoord(array([0, 1, 2]), standard_name=None, units=Unit('m'), long_name='foo')

However, if I just print the metadata,

print(coord.metadata)

I see all the unset stuff too, which I'm not really interested in:

DimCoordMetadata(standard_name=None, long_name='foo', var_name=None, units=Unit('m'), attributes={}, coord_system=None, climatological=False, circular=False)

For differences, it's hard to see at a glance what is actually different:

coord2 = iris.coords.DimCoord(range(3), long_name='foo', units='m', attributes=dict(bar=1), var_name='wibble')
print(coord2.metadata.difference(coord.metadata))

DimCoordMetadata(standard_name=None, long_name=None, var_name=('wibble', None), units=None, attributes=({'bar': 1}, {}), coord_system=None, climatological=None, circular=None)

With my branch, these outputs become:

DimCoordMetadata(long_name=foo, units=m, attributes={}, climatological=False, circular=False)
DimCoordMetadata(var_name=('wibble', None), attributes=({'bar': 1}, {}))

Note that, if you do want to see all the unset metadata, you can still use the repr. We could potentially also skip empty attributes dictionaries for the str.

Does something like this need tests? I had a peek at the coord string tests, but they seem to be focussed on getting date formats right.

---

Consult Iris pull request check list

[bjlittle]

@rcomer Hey, thanks for pushing this 👍

I'm about to make a similar change to _DimensionalMetadata for having a slimmer __str__ as opposed to an unambiguous __repr__ ... btw it will also allow you to keep data lazy rather realizing it just to print it.

[bjlittle]

@rcomer Awesome, thanks! 😄

Just a couple of suggestions... also is you wanted to add some specific tests to exercise this, then that would be great 👍

[bjlittle]

Perhaps explicitly clarify, with string -> __str__ ?

[bjlittle]

@rcomer I think that we could be even more liberal here and ignore empty strings perhaps, but definitely and empty dict for the attributes member.

[rcomer]

It took me longer than I care to admit to figure out the logic for this 😆

[rcomer]

Hi @bjlittle, I've addressed those specific comments. Do you have anything particular in mind for tests? I haven't come up with any prime candidate for "this might go wrong", so was just thinking of checking the method works without crashing for one example out of each subclass. It would be good to know that's the Right Way Forward before embarking on that though.

[rcomer]

I can confirm that this method gets called indirectly by 60 other tests because, well, I broke them!

[bjlittle]

@rcomer I think this is totally fine TBH.

I may add two iris.common.metadata.BaseMetadata tests just to be explicit.

I'll push that separately, if you fancy reviewing and merged?

[bjlittle]

@rcomer See #4043 👍

[rcomer]

Sweet. Thanks @bjlittle!