Accuracy of Conversion

[James W. Demmel]

You comment that converting binary floating point data to decimal
and converting back yields errors that can and should be avoided.
You are quite right, and IEEE 754 addresses this problem fairly
completely by saying exactly how well this conversion must be 
done so that the original binary floating point number is computed
on reconversion. Detailed algorithms attaining this accuracy
are described in Jerome Coonen's 1984 Math PhD thesis from
Berkeley. I believe this is also discussed in the March 1981 Computer.
    Jim