Rules and Values

Rule bodies are line-oriented instruction sequences. The current instruction families are grouped below.

Token Consumers

-> IDENT
-> KW "ret"
-> [NUMBER, STRING]

IDENT, NUMBER, and other canonical Token names consume a matching token. KW consumes a keyword with an exact string value. The bracket form stores the first matching token from its candidates.

Rule Control

-> JMP mov
-> jmpif mov, true
-> jmpif mov, #call[should_jump]
-> CALL #call[emit_value]
-> break
-> error, "invalid instruction"

JMP invokes another Rule and returns when it finishes. jmpif invokes it only when its boolean condition is true. CALL invokes a Python function. break stops the current Rule flow.

Token storage is separate from the datastack:

-> save "current"
-> load "current"

save records the current token under a label. load requires the current token to match that saved token and then consumes it.

Emission

-> put 0000 0000
-> ebit 0 1 0 1 0 1 0 1
-> emmit 83847483
-> emitfr 0000 0
00000000
0000
-> emitfrend

put emits two four-bit fields. ebit emits eight one-bit fields. emmit uses .wordsize. emitfr emits a multiline frame and accepts any VALUE in its rows.

Tables, Reservations, and Positions

-> reserve "slot" 16
-> ldres "slot" 0000000000000000
-> gtres "slot"
-> dlres "slot"
-> emmit &slot
-> emmit ::SIZE
-> emmit ::VSIZE

&slot reads a reservation as a value. reserve creates a zero-filled named region, ldres patches it, gtres copies it into the current output, and dlres removes its name without clearing its previous bits.

Immediate Fitting

-> fits 8
-> iffits 8
-> ifnfits 8
-> fits $registers.name[rax].bits

These instructions operate on the current MIR token. fits emits the low bits when the value is too wide. iffits emits only when the value fits exactly. ifnfits consumes a value only when it does not fit.

Widths can be values obtained from a table, a Python function, a reservation, or the datastack.

Value Forms

RIF values include numeric and string literals, booleans, table accesses, reservation accesses, calls, and position queries:

$registers.bits
$registers.name[rax].bits
&slot
#call[get_bits]
::SIZE
::VSIZE
::ALIGNMENT
dist "mov"
vdist "mov"
resdir "slot"
resvdir "slot"
resdist "slot"
resvdist "slot"
reswidth "slot"
vreswidth "slot"

Token references inside consumers use canonical names from gram.lexer.Token or their numeric enum value. The token_names keyword group is populated from the complete enum at frontend initialization.

Binary strings preserve their written width, including leading zeroes. This is why 0000, "0000", and a numeric value are useful in different contexts: the first form retains a source width for fixed fields, while a quoted binary string is validated as a complete bit sequence at runtime.