Date: 31 Aug 1999 05:51:17 -0000 From: "D. J. Bernstein" <djb@cr.yp.to> Why aren't transition times labelled with the starting offset? Or, more simply, always specified in terms of standard time? Part of the motivation for the seemingly unnecessary complexity of the rule lines is a desire to match the natural-language statutes as closely as possible. When translating statutes or regulations into tz format, if the original text specifies standard time, that's what I put into the Rule line, and likewise for wall-clock or universal time. Also, standard time is not always the simplest. The current EU rules are specified in terms of universal time, and translating this to standard time would complicate them. I see that Tunisia one year had two transitions in the same month. Why are these part of the rule rather than the zone? Unfortunately, when I entered the Tunisian rules, I didn't have access to the primary data -- just a series of transition times -- so I couldn't easily tease out which changes corresponded to DST transitions and which changes corresponded to rule changes (due to the Allies taking territory from the Axis or vice versa). In cases like these I generally just tried to minimize the length of the tz text rather than guess exactly what the original data were. Are there other examples of two transitions in the same month? Certainly there have been others in real life. They may not have made it into the tz tables. The Tunisian example is the shortest that I know of in the current tables. Hawaii had DST for only three weeks in 1933; is that close enough? Are there any DST systems that don't consist of two jumps in each applicable year, one jump early in the year, one jump late in the year? Yes. For example, occasionally Britain has had four jumps per year, with double daylight saving in the middle. A more obscure example: from October 1918 to February 1943 British Honduras used DST in the winter, not the summer -- though obviously this can be modeled using the same techniques that model ordinary southern-hemisphere DST. Are there any rule lines that cover multiple years without being part of such a pair? I'd guess that you could find an example of a rule line that (for some year) has no paired line, because the rules changed in the middle of the year. Can't think of one offhand though.

Paul Eggert writes:

Part of the motivation for the seemingly unnecessary complexity of the rule lines is a desire to match the natural-language statutes as closely as possible.

Okay; but why not include the previous ``save'' value, so that the times have unambiguous meanings?

Hawaii had DST for only three weeks in 1933; is that close enough?

What I'm asking for is a clear definition of the order of rule lines. Is it enough to sort them by month, and by day within month when the day is constant? Why are the rule lines allowed to be in any order? Aren't they always in the same order that they occur in each applicable year? ---Dan