dc and date numerology

Greg Lehey wondered about the date in 1973 is indeed post-Second-Edition. But it's not the right date; just as the permission flags were different in the early years, so was the date representation. Here are some gleanings from old manuals that tell the story. The relatively recent ls or tar or whatnot that printed the line above presumably interpreted the date as if it were in modern form: seconds since 1 Jan 1970 UTC. So the raw number stored in the i-node was probably about 105000000 decimal (30 Apr 1973 in my time zone), or about 1200 days into the epoch. But the file system described in the First Edition manual takes the date as a count of clock ticks since 1 Jan 1971. The clock ticked at 60Hz, so the date is really about 1200/60 = 20 days into the epoch; if this file came from a 1e file system, it was written on 21 Jan 1971. The trouble with keeping a 60Hz clock in a 32-bit number is that it takes just a couple of years before it overflows. A band-aid had been stuck on by the time the Third Edition manual was printed: the base date changed to 1 Jan 1972. So maybe bin/dc was written on 21 Jan 1972 instead. There's no way to tell just from the bits in the i-node. The modern time format (1-second resolution) appeared in the Fourth Edition manual. It is probably not a coincidence that the file system format changed a lot at the same time; groups appeared, permission modes changed to approximately their current form, directory entries changed, and so on. The 60Hz scheme seems to have come from the PDP-7, on which it made more sense; the -7 has 36-bit words, so a 60Hz counter lasts 16 times longer. I bet the base date changed at least once between the original PDP-7 system and the PDP-11 as well, since 1 Jan 1971 seems too recent for the PDP-7 system. See http://www.cita.utoronto.ca/~norman/old-unix/old-fs.html for many more such grotty details, collected in an insomniac night with a stack of old manuals some months ago. Norman Wilson Received: (from major@localhost) by minnie.cs.adfa.edu.au (8.9.3/8.9.3) id KAA42069 for pups-liszt; Thu, 28 Oct 1999 10:02:33 +1000 (EST) Received: from henry.cs.adfa.edu.au (henry.cs.adfa.edu.au [131.236.21.158]) by minnie.cs.adfa.edu.au (8.9.3/8.9.3) with ESMTP id KAA42060 for <pups(a)minnie.cs.adfa.edu.au>; Thu, 28 Oct 1999 10:02:26 +1000 (EST) Received: (from wkt@localhost) by henry.cs.adfa.edu.au (8.9.2/8.9.3) id KAA26898 for pups(a)minnie.cs.adfa.edu.au; Thu, 28 Oct 1999 10:02:25 +1000 (EST) From: Warren Toomey <wkt(a)cs.adfa.edu.au> Message-Id: <199910280002.KAA26898(a)henry.cs.adfa.edu.au> Subject: Re: dc and date numerology In-Reply-To: <199910272349.JAA41914(a)minnie.cs.adfa.edu.au> from "norman(a)nose.cita.utoronto.ca" at "Oct 27, 1999 7:48:50 pm" To: Date: Thu, 28 Oct 1999 10:02:25 +1000 (EST) Cc: pups(a)minnie.cs.adfa.edu.au Reply-To: wkt(a)cs.adfa.edu.au X-Mailer: ELM [version 2.4ME+ PL43 (25)] MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: owner-pups(a)minnie.cs.adfa.edu.au Precedence: bulk In article by norman(a)nose.cita.utoronto.ca: [ much omitted ] Norman details the fact that early Unixes stored time in 60ths of a second, i.e the normal clock tick, and as such, a 32-bit integer overflows in around 2.5 years. However, I think Norman is not exactly right when he said that the tar archive was reinterpreting this 1/60 sec time in units of seconds. Dennis Ritchie, with help from Keith Bostic and a DECtape drive, managed to retrieve these files from an old DECtape. These old files were stored in tap(1) archive format. Dennis wrote a program to read in the tap(1) format archives and extract their contents while trying to maintain the _correct_ permissions and timestamps. Here is his email describing this: The tapes were written in either the 'tap' or 'tp' format, which are similar in that they have a directory of up to 192 entries at the start with names and other information including the size and tape address of the files. 'tp' was the later format, and was in use by November 1973, the date of the 4th edition manual. With `tap', the times associated with the files were recorded in pre-modern units: sixtieths of a second, from an origin that changed. The first three editions of the manual had BUGS sections noting that 32 bits can represent only about 2.5 years in this unit, and this implied continuing crises as the time overflowed. I believe that the change to use seconds for Unix time took place along with the change to the C version of the operating system, which occurred about the end of the summer of 1973, and also that the change from `tap' to `tp' took place at the same time. (This is consistent with the dates of the 3rd and 4th edition manuals). Thus the dates recorded with the `tp' tapes probably correspond reliably to the modification dates of the files at the time of saving them (of course, this gives only a upper bound on their creation, since they might have been copied or trivially touched just before saving them). Recovering the proper dates for the `tap' tapes is less reliable, because there was at least one change of epoch (from 1971 to 1972) during the period they could possibly have been produced. I believe that the 1972 epoch is most likely the correct one for the tapes here. In other words, Dennis had to guess the epoch when recovering these files. He got it right with the `nsys' kernel files, because there is enough other data lying around documenting the kernel rewrite from assembly to C, and the inclusion of pipes into the kernel. However, with the s2.tar archive, I think Dennis got the epoch one year out, i.e everything should be dated a year earlier. The most obvious is that there are so many 0405 magic a.out files in the archive, and apparently this a.out format disappeared in the 2nd Edition. Cheers all, Warren