Whatever number of stamps end up in the file, my guess is that the thing to do is put at the start of the file a count of the number of bytes worth of stamp that appear, then put the stamp information itself at the end of the file. This minimizes the encroachment on the twenty remaining bytes of "reserved for future use" space; it also would allow us to have longish, easy-for-humans-to-read stamp information (either in addition to or instead of shorter, easy-for-computers-to-process information). My sense is that one stamp might be sufficient. A client picking up time zone information from multiple sources would use the one with the largest stamp. The stamp might be automatically generated as a function of the maximum "until" time appearing on a "Zone" line for the zone in question and the maximum "from" and "to" years appearing on a "Rule" line for the zone in question. Doing this would eliminate the need for separate maintenance of information from which to generate stamps; it would also avoid adding information to the zic input files that old versions of zic would not be able to handle. The down side would be that if we learned of a correction to zone information that only affected past times, we wouldn't get an *automatic* update of the stamp for the zone. Comments? --ado
From: "Olson, Arthur David" <OLSONA@dc37a.nci.nih.gov> Date: Fri, 31 Oct 1997 08:30:14 -0500 my guess is that the thing to do is put at the start of the file a count of the number of bytes worth of stamp that appear, then put the stamp information itself at the end of the file. How would this work with the Unix `file' command? On Solaris 2.6, at least, that command looks only at the first 512 bytes. The stamp might be automatically generated as a function of the maximum "until" time appearing on a "Zone" line for the zone in question and the maximum "from" and "to" years appearing on a "Rule" line for the zone in question. I'm not sure how well this would work in practice, even for future times. E.g. suppose Asia/Tehran switched from its current system (based on the Persian calendar) to some other set of rules that could be expressed in two lines rather than two lines per year. Then the UNTIL time wouldn't change, and the FROM and TO years would shrink, so it'd look like we had an older version. And of course I'd like to have past times updated correctly, as well. Why not use a timestamp as a comment in the first line? E.g. change # @(#)asia 7.30 to something like this: # @(#)asia 7.30 1997-11-01 10:20:30 I think SCCS will do that for you.
participants (2)
-
Olson, Arthur David -
Paul Eggert