issue_comments: 280856640

@spencerahill many thanks for looking into this further. I agree that simulations of tens of thousands of years are beginning to appear on the research landscape, and shouldn't necessarily be a use-case we dismiss.

I got some more concrete information c.f. on my previous comment on negative and/or 5 digit dates. The TRACE simulation outputs netCDF files uses units of thousands of years relative to 1950, and therefore doesn't use 5 integers. But it does use negative and positive floats...negative for <1950, positive for >1950.

I think it's important to note that it doesn't matter how the date information is stored in the original files; what matters is what it looks like once it's decoded. If someone analyzing this simulation would like to use a NetCDFTimeIndex, it would require that they decode these floats into actual datetimes (which would need five-digit years to express). So they would need support for five digit (possibly negative) years in partial datetime string indexing. As a side note though, this decoding is currently not possible with netCDF4.num2date (since "thousands of years" is not an accepted unit there), but I could see a way someone could do the decoding manually.

I'll need to think more about options for date parsing; ideally I would like there to be at least one common string format that one could use to specify dates for both a NetCDFTimeIndex and a DatetimeIndex (with the ISO8601 parser we have now, supporting only 4-digit, positive years, I believe that is currently the case). If we can achieve that while adding support for five-digit and negative years, that would be great, but perhaps it's best to leave that for a future PR.