Struct ComputeDefinition 

pub struct ComputeDefinition {
    pub name: String,
    pub shield_ref: String,
    pub parameters: Vec<Parameter>,
    pub return_type: String,
    pub body: Option<Expr>,
    pub loc: Loc,
    pub leading_trivia: Vec<Trivia>,
    pub trailing_trivia: Vec<Trivia>,
}

Fields

name: String

shield_ref: String

parameters: Vec<Parameter>

v2.67.0 — the typed parameters. Before v2.67.0, parse_compute SKIPPED everything between the name and the brace (“Skip optional parameters/ return type”), so a compute had no inputs at all — which is one reason it could not compute anything.

return_type: String

v2.67.0 — the declared result type.

body: Option<Expr>

v2.67.0 — the body: a v2.26.0 Expr.

This is what makes compute honest. The README sells it as “Deterministic muscle — native Fast-Path execution BYPASSING the LLM” and even asserts a complexity class (“compute steps: O(n)”). A v2.26.0 expression is exactly that and nothing more: a closed, total, side-effect-free term the runtime evaluates with eval_expr — the same native evaluator let, grad and conditional already use. Linear in the term. No model in the loop.

None ⇒ a compute that cannot compute; applying it is refused (axon-T941), rather than binding the literal string "compute:Name(args)" as the pre-v2.67.0 runtime did — which a downstream step then consumed as if it were a number.

loc: Loc

leading_trivia: Vec<Trivia>

v1.5.2 — leading comment trivia attached to this declaration (comments preceding the declaration’s first token, since the previous declaration or file start). Empty by default.

trailing_trivia: Vec<Trivia>

v1.5.2 — trailing comment trivia (same line as the declaration’s last effective token). Empty by default.

Trait Implementations

impl Debug for ComputeDefinition

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.