TAI and leap second support: worth it? Peter McGoron 18 Nov 2026 03:02 UTC

Currently, the SRFI is written as-if leap seconds will continue to be
inserted. Now that they are a thing of the past, does new code really
need to support it?

Basing computer time off of TAI made sense when UTC encountered
discontinuous shifts regularly. However, UTC will now only be adjusted
by the hour, which will not occur for several centuries. For all
practical purposes, UTC is now a continuous timescale.

I think the SRFI should no longer require that the implementation
represent leap seconds uniquely. This lifts a huge burden off of the
backs of implementers, who would otherwise waste a bunch of time
implementing fragile and very untested code for 27 specific seconds over
the span of 45 years.

Instead:

1. Clock time seconds can no longer read “60.”
2. Moments must still refer to each moment in time unambiguously and are
monotonic, but an implementation is not required to distinguish events
on a leap second. If an implementation distinguishes leap seconds,
moments are TAI. If an implementation does not, moments can be UTC for
the next few centuries.
3. Excise all statements about leap seconds and TAI from the SRFI.
4. The sample implementation will be modified to throw out
leap-second-handling junk.

Thoughts? I will probably try to find some expert opinion on this,
although I don’t know where to look.

-- Peter McGoron