::SHE+ILA:: bytes, not nybbles

More about ::SHE+ILA:: bytes, not nybbles

So far we have represented low-level ::SHE+ILA:: code in terms of nybbles (four bit lumps), representing ten BCD digits and six opcodes/tags. This is how it was in the limited memory (64kB) of the Apple ][ and Atari 800 XL. However, with only six tags (A B C D E F) we are running out of map space to do the full polymorphism we need (e.g. combining digits and characters, and a lower-entropy representation of decimal points and negative (and quaternion) numerics), and splitting bytes into nybbles slows down the major cycle engine. So in the PC version we now map the entire 256 values of a byte as follows :
$00 to $1F ... Control characters
$20 to $7F ... ASCII characters
$80 to $E3 ... Packed decimal (BCD) digits 00 to 99
$E4 to $EF ... Tags (see below)
$F0 to $FF ... Opcodes (same as the F page)
The range $E4 (228) to $EF (239) is allocated to represent the Address tag (formerly $A), the Buf tag (formerly $B), the End tag (formerly $E), and several other flavours of those tags (such as $C when used as CONST (but not when used as CHAR, which is no longer necessary, since the full ASCII set $00 .. $7F is now available for intermingling within a value)). The "Janet and Julia Story Forum" technique of using $E4 onward as Quaternion tags is now replaced by inline use of ASCII "i", "j", "k", along with "-" as a monadic minus for negative components. This avoids over-using the newly-acquired $E4...$EF range. Similarly, the rational tags "/" and "M" are now just embedded ASCIIs within a normal BCD numeric. For example 1234-i5678j90k12/100M represents the numeric quaternion (1234 - 5678*i + 90*j + 12 * k) .div. 100 000 000. (M represents 10^6, G represents 10^9, as in the original "Janet and Julia" rational quaternions specification. PEMDAS expressions use ASCII + - * ( ) as dyadics, untokenised (as opposed to Applesoft's tokenisation of PEMDAS operators). Negative numerics use leading ASCII - rather than the IBM360-like trailing $CD tag.