25 Sep
2020
25 Sep
'20
12:17 a.m.
On Thu, Sep 24, 2020 at 1:19 PM Paul Winalski <paul.winalski(a)gmail.com>
wrote:
Or separate-sign numerical representations, for that matter, such as
the IBM 1620 had. And packed decimal on the S/360/370
and the VAX.
Negative zero in packed decimal is responsible for a lot of the
"computerized bill for $0.00" bugs that were rife in the 1960s.
IEEE floats are still sign-magnitude, for that matter, both the binary ones
and the little-used decimal ones, which is why MAX_FLOAT and MAX_DOUBLE are
the same in both positive and negative directions.
I once had to deal with some mainframe data that had been transferred to
our PDP-11 or Vax (not sure which), and I noticed right away that some of
the allegedly numeric data had a non-digit immediately following and one
fewer significant digit than the rest. A little digging in my memory (I
think this was before I got access to the Internet in 1985 or so), plus a
little experimentation, established that these were
trailing-overpunched-sign numbers that had been mechanically translated
from EBCDIC to ASCII. So on the principle of "If a problem is not
interesting, generalize it until it is", I wrote a little filter that
looked for such numbers in its input and rewrote them as
leading-separate-sign (i.e. the Usual Way), copying everything else.
Unfortunately the source is long since lost, but I wouldn't write it in C
today anyway.