Sept 200
  - fixed bug in handling of rationals (found by Marko)
  - set DEFAULT_FORCE_COUNT=1
  - added cotangent family of functions.
  - corrected math-lib/Makefile to include neg_R
  - corrected asin_R and acos_R to give correct sign for negative arguments
  - moved to gmp-3.1
  - now version 6.2
  
Jul 2000
  - fixed pi bug (found by Marko and Reinhold)
  - fixed bug in reduction code which meant that most functions failed
    for rational arguments (found by Marko and Reinhold)
  - added Marko's corrected force_R_Dec
  - cleaned up stack structure
  - added environment variables ICR_STACK_SIZE, ICR_FORCE_DEC_UPPER_BOUND and
    ICR_DEFAULT_FORCE_COUNT. See the manual for details.
  - a number of corrections to the manual.
  - added secant and cosecant family of functions.
  
Apr, May, Jun 2000
  - extensive modifications. Implemented lazy conditional (on the heap)
    and in-place reduction. All eager reduction abandoned. Most of the
    math-lib involving explicit tensors re-written.
  - added Marko's force_R_Dec.
  - added a couple of functions suggested by Reinhold.
  - made atan_R work on the whole real line.
  - extensive changes to the user manual by Reinhold.

Mar 15,16
  - abandoned automatic guards on matrices and vectors. This represents
    a fairly drastic change in the structure of the implementation and
    advances us to version 6.0
  - extended predicates to operate on matrices and vectors directly rather
    than only on streams.

Mar 14
  - adjusted many math-lib functions (eg tan_R) to work with the more
    eager reduction strategy. Also tan_R now uses stdTensorCont rather than
    custom continuations.
  - fixed the handling of strsep(3) in base/davinciInterface.c and borrowed
    a copy of strsep from the GNU glibc library as strsep is not available
    in Solaris 2.5.
  - when running daVinci, the child process (which execls daVinci) now
    sets the DAVINCI_ICONDIR environment variable. This requires that
    REALDIR be set to the root of the real library tree in Makefile.
  - rewrote the i/o routines which talk to daVinci since Linux version
    would not work reliably with Solaris 2.5 and the interface to daVinci
    was needed for debugging.

Mar 13
  - extended (and simplified) the reduction of digit streams and lfts in
    base/reduce.c. Reduction is now much more eager. This may be a mixed 
    blessing. Earlier it was the case that, in some circumstances, the arguments
    to matrices and tensors would not be reduced. Hence, eventually, computation    would be done as streams, even when the arguments are rational. For small
    rationals, this is less efficient, but for large rationals, treating them
    as streams is much more efficient. It could be there needs to be a test
    introduced in reduce.c whereby reduction of lfts is inhibited when
    the entries in the lft get too large (or some other criteria).

Mar 12 
  - fixed bug in forceInfo which caused unfair forcing of the arguments of
    a tensor. Thanks to Reinhold Heckman for pointing out the problem.

Mar 8 
  - added static to qualify local variable `doneInit' in sqrt_R
  - reverted to old version numbering sequence. Version 1.03 is now
    version 5.03.

Feb 18, 2000
  - fixed typos in sinh_R, cosh_R, tanh_R which meant they were useless
  - fixed bug in tan_Q which caused it to diverge for 0.
  - release version 1.03

Oct 22, 1999
  - A little tidying up prior to first public release
  - fixed some compatability problems between the Solaris and Linux environments
    and for different versions of gcc.
  - release version 1.02

Sept 5, 1999
  - Added the file base/digitHandling.c and adjusted real.h and base/Makefile
  - removed reference to error.h in math-lib files
  - changed B_ constants for lazy booleans to avoid clash
    with Solaris types.h file
  - release version 1.01

June 1999 initial release verison (1.00), fifth version overall.