Struct DictateState 

pub(crate) struct DictateState {
    pub(crate) snapshot: Mutex<DictateSnapshot>,
    pub(crate) cancelled: AtomicBool,
    pub(crate) engine: Mutex<Option<Arc<Engine>>>,
    pub(crate) active: Mutex<Vec<(DictateRole, ActiveModel)>>,
}

Preflight’s own handle on the work: the snapshot a view renders, the cancellation flag, and the active models the resolution landed.

Fields

snapshot: Mutex<DictateSnapshot>

cancelled: AtomicBool

engine: Mutex<Option<Arc<Engine>>>

The live engine, held for the whole run. Dropping it unloads the weights, which is the one thing preflight exists to avoid.

active: Mutex<Vec<(DictateRole, ActiveModel)>>

Which model each role runs: the compiled pins until the preflight resolves, then whatever the config, the runtime pick or the pin answered.

Implementations

impl DictateState

pub(crate) fn new(settings: &DictateSettings) -> Self

pub(crate) fn set_active(&self, models: &[(DictateRole, ActiveModel)])

Move the resolved models in, before the preflight fetches anything: the rows follow the files the roles will actually run, keeping the state a row already reached when its file is unchanged.

fn set_state(&self, file: &str, state: DictateModelState)

pub(crate) fn mark_all(&self, state: &DictateModelState)

Move every row to one state: preflight’s Loading and Ready points, and the same Ready after a runtime swap, where every role’s model is loaded by the engine that just came up.

pub(crate) fn fail(&self, failure: DictateFailure, file: Option<&str>)