> From: "Todd C. Miller" <mill...@openbsd.org> > Date: Sat, 26 Oct 2019 06:55:02 -0600 > > On Sat, 26 Oct 2019 12:15:33 +0100, Stuart Henderson wrote: > > > The way these files are supposed to work is that you set the system > > clock to the time with leap-seconds included (UTC+leap, or TAI-10) and > > copy the entire "right" set of files to the main zoneinfo directory > > (upstream provides them as parallel directories to encourage this). > > > > And everyone else sets the system clock to UTC and uses the "posix" files. > > > > https://data.iana.org/time-zones/theory.html#leapsec > > > > We don't have much support for a non-UTC system clock (e.g. openntpd only > > seems to copy the flag from the server and doesn't use it to adjust the > > clock), and the files definitely cause some confusion. Should we follow > > FreeBSD and Solaris and not install the leap-second files at all?
NTP leapsecond support isn't really related to the use of these files though. In fact, it mostly exists to support the POSIX interpretation of time_t. The fundamental problem with the "right" files is that the time_t values end up being different from their POSIX values for the same UTC time. So whenever these are stored and compared between systems (or environments that set the TZ environment variable) things get weird. > I think so. Unless there are programs that use these files directly > I don't see a real use for them. Agreed. Software that really cares probably has its own leap-second table and will actually rely on the POSIX definition of time_t to convert times into human readable form. That's at least what the software I've seen and written does ;).
