Struct SolverState 

pub struct SolverState<E: BodyEvaluator> { /* private fields */ }

Live state of the solver.

Implementations

impl<E: BodyEvaluator> SolverState<E>

pub fn new(graph: ModuleGraph, evaluator: E) -> Self

Construct a fresh solver for graph. Builds the reverse- dependency index up-front so per-iteration scheduling is O(1) per dirty path.

pub fn env(&self) -> &EnvSnapshot

Borrow the current env snapshot.

pub fn graph(&self) -> &ModuleGraph

Borrow the compiled graph back out.

pub fn setter_count(&self) -> usize

Total setter count across all modules.

pub fn schedule_for_dirty( &self, dirty_paths: &[ConfigPath], ) -> BTreeSet<GlobalSetterId>

Find every CONSUMER setter that needs to fire because at least one of dirty_paths matches a prefix of its slice.

“Matches a prefix” means: if a setter’s slice is [["services", "atticd"]] and dirty_paths contains ["services", "atticd", "enable"], the setter is scheduled — a slice over a subtree fires on any descendant change.

Writers are deliberately not included here. A writer’s own re-firing happens only when its OWN slice changes (covered by the readers lookup that finds it as a consumer of its upstream’s outputs). Multi-writer conflict resolution happens at the evaluator layer via ConfigSetter::priority — firing each writer once per topo pass is enough.

pub fn topological_order(&self) -> Result<Vec<GlobalSetterId>, SolverError>

Topological order of all setters by writes→reads dependency. Kahn’s algorithm.

Errors

SolverError::Cycle if the writes→reads graph has a cycle (which is a module-author bug — cppnix would also fail here).

pub fn fire(&mut self, gid: GlobalSetterId) -> Result<ConfigPath, SolverError>

Fire one setter, write its result into the env, mark it fired. Returns the setter’s own assigns_path so callers can register it as newly-dirty.

Errors

SolverError::BodyEval when the body evaluator rejects.

pub fn run( &mut self, initial_dirty: &[ConfigPath], ) -> Result<Vec<GlobalSetterId>, SolverError>

Run the solver to quiescence starting from initial_dirty. Returns the firing order in temporal sequence. Caller can inspect Self::env afterward for the final config attrset.

On a cold rebuild (no env, no prior dirty), pass an empty initial_dirty and the solver runs every setter in topological order. On a warm rebuild where only services.atticd.enable changed, pass [vec!["services", "atticd", "enable"]] and only the setters depending on that slice fire.

Errors

Cycles or body-eval failures surface as SolverError.