🚀 go-pugleaf

gt7804b@prism.gatech.EDU (Wayne Edward Sheppard) noticed:
>I have just found a bug in CoreWarsDeluxe for the PC.  It involves
>the MOV ADD and SUB commands when using predecrement in the B-field.
>Here's how I found it:
>
>        mov 2,<2
>        jmp -1
>        dat #1
>
>What happens is when CWD reads the A field, it goes ahead and grabs
>a copy of what it is going to move.  Then if the B field decrements
>that same instruction, the mov works as normal except that it uses
>the copy of the instruction that has not been decremented.

   This is not a bug.  The specification for corewar is very vague; it simply
states that the a-field is interpreted first, then the b-field.  However, it
does NOT say whether "interpreted" means that the address is calculated,
that the value is actually read in, or what.
   Both CoreWarsDeluxe and KotH use "in-register" evaluation; that is, when
the a-field is interpreted, the value is read into an internal register of
the MARS and stored.  Then the b-field is read in.  The result is calculated
and written back.
   I think I remember Mark Durham saying that the latest version of MADgic
corewar has a switch where you can select between "in-register" evaluation
(such as above) and "in-memory" evaluation (which you apparently expected to
see).
				-Bill (wms@iwarp.intel.com)