Calendar attribute on a bounds variable

[martinjuckes]

Some CMIP6 data has hit an obscure ambiguity in the CF convention: when a time axis has a bounds variable, and a calendar is specified on both, then the two calendars must be the same.

Some CMIP6 data has time:calendar = "365_day" and time_bnds:calendar = "noleap". These two strings, 365_day and noleap, do refer to the same calendar, CF Checker flags this as an error:

ERROR: (7.1): Boundary var time_bnds has inconsistent calendar to time.

Is the CF Checker being over zealous here, or do we really want to insist there is consistency not only in the calendar being used, but also on the string used to represent it?

Using the two different terms is clearly not desirable, but I'm not sure whether it should be forbidden.