[TUHS] Validating the s2-bits tape epoch

[TUHS] Validating the s2-bits tape epoch

[Diomidis Spinellis]

The document 
http://www.tuhs.org/Archive/PDP-11/Distributions/research/1972_stuff/Readme 
discusses the uncertainty regarding the epoch used for the file timestamps.

"The biggest problem here is to pin down the epoch for the files. In the
early version of UNIX, timestamps were in 1/60th second units. A 32-bit
counter using these units overflows in 2.5 years, so the epoch had to
be changed periodically, and I believe 1970, 1971, 1972 and 1973 were
all epochs at one stage or another."

"Given that the C compiler passes, and the library, are dated in June
of the epoch year, and that Dennis has said ``1972-73 were the truly
formative years in the development of the C language'', it's therefore
unlikely that the epoch for the s2 tape is 1971: it is more likely to
be 1972. The tape also contains several 1st Edition a.out binaries,
which also makes it unlikely to be 1973."

"Therefore, Warren's decoding of the s2-bits file, in s2-bits.tar.gz,
uses 1972 as the epoch. However, Dennis decoding in s2.tar.gz uses 1973."

"Finally, the date(1) a.out on the tape uses 1971 as its archive. How 
annoying! After a bit of discussion, Dennis and Warren have agreed that 
1972 is the most probable epoch for s2-bits."

I thought I could validate the epoch by looking at the distribution of 
weekdays for the three alternative years (1971 to 1973).  Here are the 
results.

wget 
http://www.tuhs.org/Archive/PDP-11/Distributions/research/1972_stuff/Readme
for guess in 1971 1972 1973 ; do
   echo $guess
   EPOCH=$(date +'%s' -d "$guess/01/01 00:00 UTC")
   awk '/\/core/,/\/etc\/init/ {
    if ($9) print strftime("%a", '$EPOCH' + $9 / 60)}' Readme |
    sort |
    uniq -c |
    sort -n
done

1971
       1 Sat
       6 Mon
       8 Thu
       8 Tue
      17 Fri
      21 Wed
      34 Sun
1972
       1 Sun
       6 Tue
       8 Fri
       8 Wed
      17 Sat
      21 Thu
      34 Mon
1973
       1 Tue
       6 Thu
       8 Fri
       8 Sun
      17 Mon
      21 Sat
      34 Wed

As you can see, unless weekends at the Bell Labs were highly atypical, 
1972 has the most probable distribution of work among the days of the week.

[TUHS] Validating the s2-bits tape epoch

[Warren Toomey]

On Fri, Dec 23, 2016 at 07:15:04PM +0200, Diomidis Spinellis wrote:
> "Finally, the date(1) a.out on the tape uses 1971 as its archive. How
> annoying! After a bit of discussion, Dennis and Warren have agreed that 1972
> is the most probable epoch for s2-bits."
> 
> I thought I could validate the epoch by looking at the distribution of
> weekdays for the three alternative years (1971 to 1973).  Here are the
> results.
...
> As you can see, unless weekends at the Bell Labs were highly atypical, 1972
> has the most probable distribution of work among the days of the week.

Thanks Diomidis, that helps to confirm the 1972 date for the s2-bits tape.

Cheers, Warren