[nottorp]

> The calendrical calculations always assume a Gregorian calendar, with no leap seconds.

That could drift a bit if you can represent a 580 million year interval (with millisecond precision), wouldn't it?

[Someone]

A library that handles leap seconds can’t do much better, as it’s as good as impossible to predict leap seconds, certainly not that far ahead.

Luckily, leap seconds are on the way out (https://en.wikipedia.org/wiki/Leap_second#International_prop...)