Implement matplotlib converters?

[aulemahal]

Since the 1.5.0 release, plotting cftimes with matplotlib is broken because nc-time-axis uses deprecated parts of cftime. A PR was open to fix this issue (SciTools/nc-time-axis#59), but it hasn't seen any activity yet. In fact, that package is not very active in general (which is understandable as it is quite small).

I was wondering if it would make sense to implement its functionality directly here, in cftime? Matplotlib would not be added as a dependency, but when present the converters could be registered upon import? The code seems low maintenance in general and with a compatible licence, but moving it here would reduce the number situations like right now.

I am willing to contribute to this.

What do @jbusecke, @trexfeathers and @jswhit think?

[jbusecke]

I honestly do not understand these inner workings enough and as such would rely on the opinion of others here. Thanks for the heads up though.

[jswhit]

I know nothing about how nc-time-axis works either. If it's a simple matter of converting to a matplotlib standard calendar, we could provide that here - but providing tickers and formatters should be left in a separate package I think. Seems like the fundamental problem is that nc-time-axis is not very actively maintained.

[trexfeathers]

Seems like the fundamental problem is that nc-time-axis is not very actively maintained.

I agree. We create many good ideas in the SciTools organisation without knowing if we have the resource/priority to maintain them. I will continue to keep my colleagues up to date on these problems, so we can try to find time to sort them out.

[aulemahal]

@jswhit Sadly, I don't think converting to a standard calendar would work as expected, all those non-standard dates would get in the way. The nc-time-axis package is pretty much 3 classes : a Ticker, a Locator and a Converter. The latter converts to integers (rather than to standard datetimes).

I believe it is not actively maintained in part because it is relatively low maintenance : the matplotlib interface seems quite stable.

[jswhit]

@aulemahal so you are proposing moving the Converter from nc-time-axis here (without adding a matplotlib dependency)? From a cursory look at the code it seems that it won't be that simple (the Converter uses the Locator and the Formatter).

[aulemahal]

Oh no, sorry. previous comment was a bit ambiguous.
I am suggesting moving the whole thing to cftime. Without adding a matplotlib dependency indeed. Idea: put this in a "mpl_utils.py" and, in __init__.py do:

try: 
    import matplotlib
except ImportError:
    pass
else:
    import mpl_utils

[jswhit]

I would only want to do this at the request of the nc-time-axis team - it seems to me that there are a lot of good reasons to keep it separate. @trexfeathers what do you think?

[trexfeathers]

@trexfeathers what do you think?

I'm going to spend some time this week trying to sort out nc-time-axis, cf-units, and any other dependent packages we have that are struggling to keep pace.

[bjlittle]

Ping @aulemahal @jbusecke

Fancy joining the discussion over on SciTools/nc-time-axis ? 😄

[jswhit]

sounds like nc-time-axis is getting jumpstarted - which means we shouldn't throw a monkey wrench in that effort by merging nc-time-axis functionality into cftime. Going to close this now.