On 2020-06-12 21:42, Paul Eggert wrote:
But this is curious, as that story also makes sense only if clocks were set back at 00:00 "old time" - which is not what I installed yesterday in tzdb master where clocks were set back at 00:09:21 "old time" so there was no difficulty with the unfortunate hypothetical children in question.
Yes, the offset is wrong and so is the instant implied for the jump. A single sentence in a contemporary newspaper should not be taken at face value, especially if some arithmetic is involved.
This new citation suggests that it was more common for people to think of clocks as being set back at midnight old time, which means we should revert to tzdb 2020a practice (while keeping the other changes you've suggested).
Well, the common perception of people about the transition can hardly be determined from a single newspaper report either -- but common sense tells us that most people did not care a bit about whether the jump was from 00:09:21 to 00:00:00 or from 00:00:00 to 23:50:39 (or anything in between). However, we _do_ have detailed reports on the switch in French railway time (implying a jump from 00:04:21 to 00:00:00), and we know that radio signals since 1911-03-11T00:00:00 (not since 1911-03-10T23:50:39) were ordered to be transmitted with the label "heure nouvelle". This is, in my opinion, evidence that the switch in French legal time occurred when UT was 1911-03-11T00:00:00.
In other words, since the law doesn't specify how the transition occurred, if actual practice varied and some people switched at midnight old-time, then tzdb should leave the time blank (defaulting to 00:00) to hint that "0:00" would be overspecifying.
I do not know whether I have seen all legal texts about the switch. The switch in railway time seems to have been planned meticulously, and there are references to additional legal texts concerning the switch which I have not found. Michael Deckers.