I have been informed of a bug in the posted version of decimal.g
(the sign of some negative numbers was not printed). I have uploaded
to the directory /pub/incoming in Aachen a new version of the file which
fixes the bug, and as well contain a tremendous speed improvement for
evalf() on cyclotomic numbers which I get by caching already computed
primitive roots of unity. Because of this cache I have changed the interface
for changing the precision. This is now done by the function
SetDecimalPrecision which takes care of chores such as flushing the
cache. Also I removed the limitation on the numbers of decimal digits by
replacing the wired-in value of Pi by a routine which computes its digits.
The complete documentation still resides in the file header.