
***********************************************************************

                       The Elliott 903
                       Algol simulator
                       ---------------

While converting some programs from a real 903 which happened to
use logarithms quite a lot, an error arose while evaluating ln(0.2).
It gives the result 1.163151 instead of -1.609432. It was tracked
down by Terry Froggatt and is caused when a multiply instruction
references a store location containing -1.0, ie 131072. Although 
it would seem to be a rarely occurring error, it affects about 40%
of all logarithms where the argument is less than 1, as shown by
a random test.

The new version, SIM900AL.EXE, cures this and has two other minor
changes :-

(1) The initial messages no longer mention the loading of modules
    0 and 1;

(2) The action taken when establishing the library by an initial 
    jump to 18 only served to do a checksum on the library
    material, but is now followed by translation of any program
    which follows the library. This could be important for fitting
    large programs into the available space because there is more
    chance of success with a pre-existing library than with a
    translation in library mode. See the small case in EXAM4.DAT.

Just download MAY900AL.ZIP into the directory which already contains
the 903 Algol stuff and unpack it. It contains SIM900AL.ADA, 
SIM900AL.EXE, EXAM4.DAT and this file itself (03README.TXT).


DGNH [ed CPB]  07-MAY-1998

***********************************************************************

