From: matz@... (Yukihiro Matsumoto)
Date: 2003-04-18T18:33:37+09:00
Subject: Re: Roundoff problem with Float and Marshal

Hi,

In message "Re: Roundoff problem with Float and Marshal"
    on 03/04/18, Rudi Cilibrasi <cilibrar@ugcs.caltech.edu> writes:

|I have implemented the idea I gave earlier.  I had to use 2 bytes to
|pass all my test cases.  There is a 2-bit space for other floating formats
|also.  I hope this technique preserves portability as well as before,
|yet also gives numerical stability which is essential for some types of
|scientific computing, e.g. distributed and long-term simulations.
|What do you think of this solution?

Interesting.  Is this endian safe, i.e. does this work properly on
both little endian and big endian machine?

If it is endian safe, then we can put mantissa bits at the tail of
float representation, e.g. "2.3847655 \001\001", to make it work well
with older Marshal.

							matz.