It is obvious that we disagree on my basic premise - that internally, sufficient bits be available to represent the date starting at a date far enough in the past to handle many old dates in the countries which would use the operating system. Some operating systems such as VMS achieved most of that goal. On OS/8, Unix and RT-11, as just a few examples, it would have required a decision to double the number of bits over what was actually used. For OS/8 and RT-11, the simple solution would have been to use half a word for the month and day (not sure what half a word is called on a PDP/8) and a full word for the year. Probably, the question of having adequate support for the date was the real decision. I have seen multi-million dollar systems which ignore the date (like a micro-wave) and only provide the time. But if there is a disk drive with files and dates to be noted when the file was created and at least modified (and perhaps referenced), then there can always be a reason to use dates on some files which precede the start of the operating system - to note when the information within the file was actually created is just one possibility. If this above basic premise is not considered adequate to justify allocating a word in memory for the year rather than 5 bits in the case of RT-11 (eventually 7 bits to manage dates from 1972 until 2099) and however many bits were used in OS/8, then there is really no point in this discussion. In respect of such vital information as the date, saving a few bits was not justified in my opinion. The fact that VMS arrived later than OS/8 and RT-11, but allowed dates much before VMS was ever released suggests that DEC had begun to realize that the date was important. The same with RSX-11. > >On 2012-07-13 19:00, "Jerome H. Fine" <jhfinedp3k at compsys.to> wrote: > >>>> >>> >[Snip] >>> >Ooo. So TECO-8 actually lie in their documentation... Even worse. >>> >A year in the range 1986-1994 would just have looked like 1970-1977. >>> >That's ugly of them. >> >> What seems even more evident to me is that DEC took >> (as most other companies did as well) the attitude that >> even the internal representation of date and time was >> not important enough to allow the same information to >> be exchanged between operating systems on a consistent >> basis. > > Sorry, but I fail to see the point. The internal representation of a > date will almost by necessity be different between different OSes and > hardware. Having a bunch of 16 bit values represent a date on a PDP-8 > would be incredibly stupid and difficult, not to mention that OS/8 > have no concept of time to more detail than a day. And even that needs > to be updated manually every day. Your assumption differs from what I would suggest - see above. That would be one solution, but other solutions are available. The most important point is to have sufficient internal bits to start with. I agree - the point I attempt to make is to stress the requirement to have sufficient internal date range to start with - see above. Which I think provides a maximum year of 2038 if 32 signed bits are used. Far better would have been to use 64 bits with a higher resolution such as milliseconds and start at least as far back as 1900. I am well aware of the nature of how complex the subject of dates and time can be. To start with, the Common Era (aka Gregorian) Calendar is not used as the primary date representation by even a majority of the Earth's population. However, with respect to using computers, the countries and societies which first started to use computers and the internet primarily use the Common Era Calendar and I would suggest that the only dates which early computers did support were Common Era (Gregorian) dates. I doubt that time zones were even considered at that early stage, let alone considered important enough to support internally, let alone externally. And even if Russia was considered important enough as a market for computer systems, I doubt that anyone would have remembered that the country adopted the Gregorian Calendar less than a hundred years prior. As for just how complex time actually is, the biggest problem that few people realize is that the length of the year can vary by many minutes from the average length of the year. I seem to remember that the variation can be about 15 minutes. Every so often, a leap second has been used to keep clocks synchronized with astronomical time (that UTC you mentioned above). The average person completely ignores that complexity, but now with the internet, operating systems must know the time very accurately. But the date is a completely different matter. At any one location, the date changes, on average, every 86,400 seconds - except when a leap second occurs - and having a few extra bits internally to handle the year is, in my opinion, essential. So the initial decision for the internal representation of the date was the base date or earliest date to be represented. Many, probably a majority of operating systems used a very restricted choice which reflected the emphasis on using as little storage as possible. In my opinion, that was a bad choice. A "universal" design would be to use Julian Days (just a count of the number of days from an agreed upon date - which has already been accepted by astronomers) instead of dates. Then, convert from Julian Days to whichever local calendar is favoured by the party using the date. But that has already been rejected by most societies and only a small percentage of the population are even aware that Julian Days exist. There are too many reasons for most societies to not use Julian Days - mostly religious in nature which requires a seven days cycle to support a holy day every seven days. The French attempted to start a new calendar, but failed - probably mostly on the seven day cycle. I noticed you are aware of the TECO discussion on dates for a number of different operating systems, so you probably already know. RT-11 will break on January 1st, 2100 in the same manner as OS/8 has already broken - there are no more bits in the date word to handle more than a range of 128 years starting in 1972. The only solution is the add additional storage to hold more bits for the year. Thus far, I know of only one other individual (who supports RUST) who is serious about adding those extra bits to the RT-11 file system, but there is no agreement as yet about how to do that. If no one else will even discuss the question of how dates for files in RT-11 can be used and extended beyond the year 2100, then eventually a unilateral decision will be made - or not as the case will be. Since it is probably not going to be very important for RT-11 to support dates after 2099 except for someone running RT-11 on an emulator (will any actual DEC hardware still be running after 2099), the matter is probably only of academic interest or only for hobby users - if any still exist in 2100. Jerome Fine