On 12/08/2017 03:52 AM, Ian Abbott wrote:
I'm not sure that switching from positive daylight savings in summer to negative daylight savings in winter is a terribly good idea, as it will probably result in various software headaches.
Yes, it does have its problems and perhaps I was a bit hasty. On the other hand, the Irish statute seems quite clear: Irish Standard Time == UTC +01 for decades and this can't be waved away as a temporary aberration. Also, any POSIX-conforming system must already deal with the situation, e.g.: TZ='IST-1GMT0,M10.5.0,M3.5.0/1' specifies Irish time since 1996 with negative DST in winter. So, although there will undoubtedly be some software headaches, these headaches can happen on POSIX-conforming platforms even if we left tzdb alone. I checked all the timestamp-using code that I normally deal with, and found one fairly-obscure bug uncovered by this change: the "holidays" function of GNU Emacs ignores the tm_isdst flag and so reports this years' transitions as follows: Sunday, March 26, 2017: Daylight Saving Time Begins 1:00am (GMT) Sunday, October 29, 2017: Daylight Saving Time Ends 2:00am (IST) where it should output this: Sunday, March 26, 2017: Daylight Saving Time Ends 1:00am (GMT) Sunday, October 29, 2017: Daylight Saving Time Begins 2:00am (IST) That is, the timestamps and abbreviations are correct, but the words "Begins" and "Ends" are misplaced. This bug can be demonstrated without the recently proposed tzdata patch, by setting TZ as shown above. I fixed it in Emacs by installing the following: https://git.savannah.gnu.org/cgit/emacs.git/commit/?id=ff105b366c24779769487... If this is illustrative of the sorts of glitches people will encounter by the change, I think we'll be OK; it's not a big deal. If there's something more serious, though, we should probably revert it (though of course keep documentation about the situation).