Paul Eggert <eggert@cs.ucla.edu> writes:
Ken Murchison wrote:
Would it also be possible to have this option (or a separate one) create the first header and body (32-bit time values) with counts of zero?
That'd be a change to the format, since the intent (as I understand it) is for the 32- and 64-bit parts to agree for timestamps in 32-bit range. (Although this isn't written down; it should be.) So this should be a separate zic option.
If part of the objective is to expose code that fails to deal with the newer format options, I think you need to be able to generate files in which (a) the 32-bit data isn't there, *and* (b) the TZ string actually controls results for current timestamps in some popular zones. If the TZ string only determines post-2038 results in Outer Mongolia, it's going to take a long time for anyone to notice if their tzdb client code is broken. I don't know if Paul's proposed behavior change would make (b) true automatically or not. regards, tom lane