Consider faster implementations of accessor functions

[hadley]

I think many of the access functions (e.g. year(), yday()) could be reimplemented much more efficiently by taking advantage of the fact that it's easy to determine year offsets with a lookup table (e.g. https://github.com/hadley/readr/blob/master/src/DateTime.h). My intuition is that plus some modular arithmetic should be orders of magnitude faster than using as.POSIXlt

[vspinu]

My intuition is that plus some modular arithmetic should be orders of magnitude faster than using as.POSIXlt

It will probably work for all units in UTC. I guess it could straightforwardly work for year, month and day irrespectively of the timezone. For smaller units it's tricky.

Our real problem is with the other way around -as.POSIXct - which is crucial for all updates. It's 5x slower than as.POIXlt.

[hadley]

No longer think that this is that important (or that easy)