simonjwright / analytical-engine Goto Github PK

check_for_prime.ae doesn’t check for even numbers; and because the first divisor it tests is 3, it finds that 10 is prime (because 5 is). It also finds 3 non-prime (because it’s divisible by 3!) which is a different problem.

As pointed out by Rainer Glaschick (section 3.5, Memory peculiarities), it would have been necessary to ensure that a column that was to be written to was zero beforehand; the operation of storing a value involved racks transferring the value on the source to the destination as the source digits were rotated back to zeros.

This means that (for example) operation 2 in the Bernouilli program would fail if written

-
L004
L001
S004

and should instead be written

-
Z004
L001
S004

Because this error would have resulted in a very difficult debugging exercise in the real machine, there should be a switch to allow detection of the problem (maybe this would have been a reason for raising the ERROR flag!)

It seems the same considerations would apply to loading a register from a number card.

Exceptions are caught during card reading, but not during execution, so that the user is left looking at a stack trace.

The Print card gets the value to print from the Egress axis of the Mill. Walker’s code gets it from the last transfer to the Mill or the result of the last operation (the remainder in the case of division).
Not clear to me what Babbage intended.