Omar Haddadi

My rule for every page is that a timestamp without a zone is an incomplete value, and the confusion it causes is far larger than the characters it saves. Unix time counts seconds from a fixed origin in UTC, which makes it trivially sortable and completely unhelpful for reading. ISO 8601 has several forms and RFC 3339 profiles the subset that goes on the wire, with a zone offset or the Z designator. The distinction that catches people is that a zone offset names an instant, while a named zone names a rule set, and only the second can say what happens next March. Time zones come in three flavours. A fixed offset never changes. A region observes daylight saving and shifts by an hour, which creates a spring gap where a local time does not exist and an autumn overlap where it happens twice. And some regions have shifted by thirty minutes or by a full day, so a fixed offset that was right has stopped being right. Leap seconds are handled in ways that differ between systems. A database column counting seconds since the epoch and a runtime using a leap-aware count can disagree by a day, and the safe answer is to store what you need and not convert unnecessarily. Then there is the practical trap. A 32-bit signed integer counting seconds runs out in 2038, and plenty of deployed systems still have that column type. Every page here says which width it is using. I also separate wall clock from monotonic time. The first can jump when a clock is corrected, which breaks interval arithmetic, and the second cannot. Any code measuring how long something took wants the second.