Gracefully handle long time intervals

[DPeterK]

There's an odd discrepancy between UDUNITS2, which will handle months and years as valid time intervals, and netcdftime, which won't. This means you can produce a valid time unit with a time interval of years, which you then cannot convert between dates and numbers.

This PR adds a new method is_long_time_interval to the Unit class -- does this unit instance describe a time unit with a long interval (that is, months or years)? If it does and it is used in a date <--> number conversion then the unit will raise an exception, rather than the error being raised in the depths of netcdftime.

Downstream, the above discrepancy affects Iris coord printing. If the __str__ method of a coord is called and the coord is a time coord, then the __str__ will try and produce pretty time coord using num2date, which will error, causing the somewhat horrid situation of a print statement causing an exception.

[DPeterK]

Note that as part of this change I've modified how cf_units.num2date and cf_units.date2num function. They now create a cf_units.Unit instance and use that instance to run the conversion so that the new long time interval exception in cf_units is raised and not the exception in netcdftime.

[coveralls]

Coverage increased (+0.1%) to 90.653% when pulling 48bec11 on dkillick:long_time_intervals into 77eac61 on SciTools:master.

[lbdreyer]

typo: These long time intervals are supported by UDUNITS2 but are not supported by netcdftime.

[DPeterK]

Thanks @lbdreyer - I've corrected the typo.

Any more for any more?

[DPeterK]

And now the license headers tests fail...

[coveralls]

Coverage increased (+0.1%) to 90.653% when pulling 792a523 on dkillick:long_time_intervals into 77eac61 on SciTools:master.

[lbdreyer]

I'm happy with the implementation + tests. Thanks @dkillick !

[pelson]

@dkillick - sorry to drag this back up, but did you raise this with the netcdftime repository? Seems like a thing they may be interested in addressing there...

[DPeterK]

@pelson I did not - I should. We should also try and get the associated Iris PR SciTools/iris#2354 merged, a job which is a bit overdue...

Edit: note that merging the above-linked still requires a release of cf_units.

[pelson]

Edit: note that merging the above-linked still requires a release of cf_units.

cf_units 1.2.0 has the magic. 👍