Dear Jim, Chris, Ben
I haven't copied the whole of the email thread this time because it's getting
so long. See
http://mailman.cgd.ucar.edu/pipermail/cf-metadata/2015/058150.html
if you want to read it all.
I agree with Chris, and I am puzzled by Jim's view on his point 3. I think that
the calculator is used to encode and decode a date-time as an elapsed time
relative to an epoch (reference date-time). The calculation uses the epoch in
the units attribute of the time coordinate and the calendar attribute of the
time coordinate; the calendar attribute indicates which algorithm to use. There
is no good reason why a different calendar would be used to encode and decode,
because it's stated which one should be used in the calendar attribute (or it's
defaulted, in the present convention). That problem should not arise (if it
does, it is a mistake by the user of the data). Hence we solve 3 by clarifying
the calendar (whether it has leap seconds or not). Point 2 seems to me not a
problem because the epoch is a timestamp; it doesn't indicate or depend on the
calendar.
I considered gregorian_noleapseconds but didn't suggest it because it's longer,
although it's OK with me if that's the majority preference. What would you say
to gregorian_nls? That is the same length as the alternative gregorian_utc. But
because the difference between these two varieties of gregorian is small, as
Chris and Jim agree, I tend to think we don't need to enforce the pain of
making data-writers change gregorian to gregorian_nls, and break existing
data-reading software thereby. As Ben says, we can blame it on careless
writers (or readers) if we simply redefine gregorian at CF1.7. That wouldn't be
ambiguous nor invalidate existing data.
There is little difference in practice between parsing "X Y" (blank-separated)
and "X_Y", but the former is slightly more awkward because there might be any
number of blanks, and whitespace is sometimes accidentally not really spaces.
The only place we use modifiers is the standard_names, and it's not been a
great success there (but that's a semantic problem rather than a syntactic
one). The format "X_Y" looks more CF-like to me and consistency of style makes
the convention more memorable and user-friendly, I think.
Yes, different countries changed to the Gregorian calendar on different dates
but I think we are safe enough if we state which one we are using viz 1582,
when it was invented in Italy. That's what udunits and CF presently use.
Best wishes
Jonathan
Received on Tue May 12 2015 - 11:12:18 BST
