Is PrePARE OK with leap years ?

[senesis]

Using two distinct libraries (python/datetime and Xios), it seems that gregorian calendar has 127835 days between 1500-01-01 and 1850-01-01

>>> import datetime
>>> j1850=datetime.datetime(1850,1,1)
>>> j1500=datetime.datetime(1500,1,1)
>>> (j1850-j1500).days
127835

However, when PrePARE is fed with a file having the time variable dumped below, it complains in a way which implies that there is a 9 days discrepancy with its own calculation .

Which library is right ?

! Error: Your filename 
! "clt_day_CNRM-CM6-1_piControl_r1i1p1f2_gr_18500101-18501231.nc" 
! does not match the CMIP6 requirement.
! 
! Your output filename should be: 
! "clt_day_CNRM-CM6-1_piControl_r1i1p1f2_gr_18500110-18510109.nc"

double time(time) ;
                ...
		time:calendar = "gregorian" ;
		time:units = "days since 1500-01-01 00:00:00" ;
		time:time_origin = "1500-01-01 00:00:00" ;
data:
 time = 127835.5, 127836.5, 127837.5, 127838.5, 127839.5, 127840.5, 127841.5, 

[taylor13]

The problem can be addressed by paying attention to the description of calendars at the end of http://cfconventions.org/Data/cf-conventions/cf-conventions-1.7/cf-conventions.html#calendar . When a simulation spans a period crossing the date when the Gregorian Calendar was first adopted (i.e., 1582-10-15), you need to set calendar="proleptic_gregorian". Otherwise a 9-day discrepancy is found.

By the way, according to this calendar, leap years occur in years evenly divisible by 4 except years evenly divisible by 100 and not evenly divisible by 400. By my reckoning, that gives us int((1850-1500)/4+1) - 3 (to account for century adjustments) = 85 leap years for years 1500 through 1850 (inclusive). So the number of days in that period should be: 351*365 + 85 = 128200, and the first time sample for year 1850 should be 128200 - 365 + 0.5 (for the middle of the day) = 127835.5.
So, it appears that at least the first day of 1850 found in your netcdf file is consistent with a proleptic_gregorian calendar.

[taylor13]

This issue was also noted in item 17c of global attributes section of https://goo.gl/neswPr . Is there a way to make that information more prominent?

[senesis]

Thanks for this clear explanation. I had never dig that question, and didn't notice that the Python datetime library states that it is:

An idealized naive date, assuming the current Gregorian calendar always was, and always will be, in effect

and that Xios , when instructed to use a gregorian calendar, actually behaves as if a proleptic_gregorian option was used
The document quoted above is clear enough, and the myriad of specifications cannot all be made prominent !

[senesis]

And PrePARE duly decalres that the data is correct once the value of calendar_type is changed to proleptic_gregorian.
Thanks

[taylor13]

this issue can be closed.