Datastack

The datastack is a LIFO value stack shared by Rules during one external MIR compilation. It is independent from the token stack declared with @stack.

Push and Pop

@rule prepare:
-> dsave 64

@rule encode:
-> fits dload

dsave VALUE evaluates a value and pushes it. dload used where a VALUE is expected pops the top value. dpeek reads the top value without removing it. dpop discards the top value. ::DDEPTH returns the current stack depth.

The following flow is common:

-> dsave 64
-> fits dload

The first Rule line can push a width for a later emission. fits dload checks the width and removes it only after a successful fit. A failed iffits leaves the value available for recovery logic.

Table-Driven Loads

The instruction form of dload queries a table using the current MIR token:

-> dload ::registers.names : bits

It performs these operations atomically from the language point of view:

1. Read the current token value.
2. Find it in registers.names.
3. Push the matching row's bits value.
4. Consume the current token.

The selector column must contain unique keys. If the token is absent, the Rule fails without consuming the token or pushing a value.

State Lifetime

RIFMachine.execute starts an isolated run and clears the datastack. execute_line processes one external MIR line while preserving the datastack, reservations, and other cross-line state. This is what allows one MIR line to prepare a value for a later line.

Python functions can inspect or manipulate the live list through:

context.datastack

Target authors should keep stack ownership clear: use the datastack for temporary values and the token storage instructions for token identity.