Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!ico!vail!rcd
From: rcd@ico.ISC.COM (Dick Dunn)
Newsgroups: comp.lang.c
Subject: Re: need EBCDIC to ASCII function
Summary: well, it was and wasn't EBCDIC
Message-ID: <16204@vail.ICO.ISC.COM>
Date: 13 Oct 89 05:10:44 GMT
References: <1060@einstein.misemi> <1989Oct4.203729.11700@utzoo.uucp> <10946@riks.csl.sony.co.jp>
Organization: Interactive Systems Corp, Boulder, CO
Lines: 37

diamond@csl.sony.co.jp (Norman Diamond) writes:
> ...henry@utzoo.uucp (Henry Spencer) writes:
> >You will have to be more specific.  Which flavor of EBCDIC?  EBCDIC is
> >not a single well-defined character code, but a family of somewhat-similar
> >codes...
> In fact EBCDIC is just as well-defined as ASCII.  Only some IBM print
> trains did not use EBCDIC...

I think it's not quite this simple.  Haul out your trusty yellow card
(that's the successor to the green card, right?:-) and look at the "Code
Translation Table."  You will see a pair of columns labeled "EBCDIC(1)".
It is this pair of columns (at least) which give rise to Henry's comment
about "somewhat-similar codes" and Norman's comment about print trains.
However, if you read the footnote (1) referenced by the column heading, you
see:  "Two columns of EBCDIC graphics are shown.  The first gives standard
bit pattern assignments.  The second shows the T-11 and TN text printing."

In other words, there are two forms of EBCDIC here (Henry's point), but one
of them is standard (Norman's point).  Ouch!  Dumb!  Keep in mind that this
wasn't the result of some dispute among vendors; IBM didn't get it right
among themselves.

What does this mean to an implementor?  There are some interesting impli-
cations here!  On input, you need to accept whatever you get.  On output,
you want to produce the codes that print right.  If you're parsing input
text (I fell into this while working on a Pascal compiler), you can simply
accept both codes for characters which differ.  (There aren't any conflicts
which matter; there are lots of holes in the codesets.)  But for character
and string constants, you gotta generate the codes you're given, which
means that the code for left bracket may not be equal to the code for left
bracket (if they were from different flavors of EBCDIC)!  Worse, the
character that prints correctly isn't the "standard" one--so you can either
have the program listing (remember those?:-) look right, or get the right
answer!
-- 
Dick Dunn     rcd@ico.isc.com    uucp: {ncar,nbires}!ico!rcd     (303)449-2870
   ...No DOS.  UNIX.