Reducing the Swelling of the Phone Bill with DIF and SSED
November 17, 1981
Chuck Forsberg
Computer Development Inc
Beaverton OR
Lately (if not sooner) it has become obvious that there must be
a better and cheaper way to distribute software updates to
changing programs than to transmit all of the new files in
their totality, even though only a few lines in each have been
changed.
For some years the Unix differential file print program diff(1)
(the (1) refers to the section of the Unix Programmers Manual
in which it is described) has had a -e flag which provides a
set of ed commands suitable for transforming the first file to
the second.
With these tools, only an update file need be transmitted,
provided, of course, that both the sender and the receiver had
copies of the same antecdent file.
I have written a "new" diff called dif.c which manages to
operate in the primitive CP/M environment. The editing commands
output in response to the -e option refrence sequential lines
in the source files, so they (the commands) can be executed by
a stream editor. (The Unix diff(1) creates difference files
with non-forward-sequential commands.)
To generate a difference file, the command is
dif -e oldfile newfile >file.dif
The >file.dif redirects the standard output to the file. A +
may be susbtituted for > if simultaneous console output is
desired.
The receiver then invokes:
ssed oldfile <file.dif >newfile
Which will result in newfile being created identical to the
oroginal newfile. Well, not precisely identical, but identical
up to and including the EOF (^Z) character. The dribble after
that may change, so CRCK may say they are different. To check,
compare the two files with dif.
Unix folks with 14 character file names and modification times
stored by the filesystem have little trouble keeping the files
synchronized. (If the antecedent files are different, there's
no telling what the output file will look like!) For us poor
CP/M folks (verrry) patiently awaiting something like Unix to
appear magically on out desktops, I propose that the revision
or revision date of the antecedent file be placed in the new
file adjacent to the new revision or date, preferably on the
same line. This way the user may easily verify that he has the
correct antecedent.
Dif Versions 1.10 and later place hash indices of the RETAINED
lines of the antecedent file in the difference output. This
allows ssed 1.10 or later to verify correctness of the
antecedent file. the new .dif files are compatible with the old
ssed, but, alas, not with Unix ed or sed.
The array sizes in dif.c may have to be shrunk somewhat to run
on a 48k system.
It ought to work if you said
dif -e filea fileb |ssed filea |dif fileb
and it does, with version 2.0.
Version 2.0 of dif.c adds a -u flag which will unsqueeze filea
before comparing it to fileb.
Thus you can say
sq filea
dif -eu filea.qqq fileb |ssed filea |dif fileb
Or you can say
dif -eu filea.qqq fileb |ssed -u filea.qqq |dif fileb
to test dif and ssed. (Be sure dif and ssed are exactly where
you say they are, or else pipes will be broken.)
Restriction: Since the BDS Standard I/O library and the
Directed I/O package are somewhat confused about translation
between CP/M's cr/lf terminated lines and **nixs' \n terminated
lines, dif was written to strip cr's from the input in order
that only one cr appear on the output. As a result, lines
terminated by cr/lf, lf, and lf/cr all come out the same! This
would munge files where lf/cr has a special meaning (MBASIC
continuation lines) or where embedded cr's are used (RTTY art).
Unix is a trademark of WECO, CP/M of Digital Research.
