| 1 | //! Active-session Work Graph authority and legacy tool adapters. |
| 2 | |
| 3 | use std::path::Path; |
| 4 | use std::sync::{Arc, Mutex, MutexGuard}; |
| 5 | |
| 6 | use crate::fleet::ledger::FleetLedger; |
| 7 | use crate::tools::plan::{PlanSnapshot, PlanState, SharedPlanState, StepStatus}; |
| 8 | use crate::tools::todo::{SharedTodoList, TodoList, TodoListSnapshot, TodoStatus}; |
| 9 | use codewhale_lane::LaneRegistry; |
| 10 | |
| 11 | use super::{ |
| 12 | BindingId, ChangeCtx, CompatPlanMetadata, CompatProjectionState, CompatTodoBinding, EdgeKind, |
| 13 | IdempotencyKey, NodeKind, NodeState, OperationBinding, OperationIntent, OperationObservation, |
| 14 | OperationOwnerSnapshot, Provenance, ReasoningEffortTier, WorkActivityEvent, WorkEdge, |
| 15 | WorkEdgeId, WorkGraph, WorkGraphChange, WorkGraphSnapshot, WorkNode, WorkNodeId, WorkNodePatch, |
| 16 | external_identity_is_well_formed, fleet_task_owner_snapshot, import_legacy, |
| 17 | lane_owner_snapshot, project_plan, project_todos, validate, |
| 18 | }; |
| 19 | |
| 20 | pub(crate) const ACTIVE_OPERATION_SUMMARY_START: &str = |
| 21 | "<!-- codewhale:active-work-operations:start -->"; |
| 22 | pub(crate) const ACTIVE_OPERATION_SUMMARY_END: &str = |
| 23 | "<!-- codewhale:active-work-operations:end -->"; |
| 24 | |
| 25 | #[derive(Debug, Clone, PartialEq)] |
| 26 | pub struct WorkRuntimeSnapshot { |
| 27 | pub graph: WorkGraphSnapshot, |
| 28 | pub todos: TodoListSnapshot, |
| 29 | pub plan: PlanSnapshot, |
| 30 | } |
| 31 | |
| 32 | #[derive(Debug, Default)] |
| 33 | struct ActiveGraph { |
| 34 | session_id: Option<String>, |
| 35 | snapshot: Option<WorkGraphSnapshot>, |
| 36 | pending_publish: bool, |
| 37 | } |
| 38 | |
| 39 | /// One active session graph plus the read-only legacy views it publishes. |
| 40 | pub struct WorkRuntime { |
| 41 | todos: SharedTodoList, |
| 42 | plan: SharedPlanState, |
| 43 | graph: Mutex<ActiveGraph>, |
| 44 | } |
| 45 | |
| 46 | pub type SharedWorkRuntime = Arc<WorkRuntime>; |
| 47 | |
| 48 | impl std::fmt::Debug for WorkRuntime { |
| 49 | fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { |
| 50 | let graph = lock_unpoisoned(&self.graph); |
| 51 | f.debug_struct("WorkRuntime") |
| 52 | .field("session_id", &graph.session_id) |
| 53 | .field("has_graph", &graph.snapshot.is_some()) |
| 54 | .field("pending_publish", &graph.pending_publish) |
| 55 | .finish() |
| 56 | } |
| 57 | } |
| 58 | |
| 59 | #[must_use] |
| 60 | pub fn new_shared_work_runtime(todos: SharedTodoList, plan: SharedPlanState) -> SharedWorkRuntime { |
| 61 | Arc::new(WorkRuntime { |
| 62 | todos, |
| 63 | plan, |
| 64 | graph: Mutex::new(ActiveGraph::default()), |
| 65 | }) |
| 66 | } |
| 67 | |
| 68 | impl WorkRuntime { |
| 69 | #[must_use] |
| 70 | pub fn matches_todos(&self, todos: &SharedTodoList) -> bool { |
| 71 | Arc::ptr_eq(&self.todos, todos) |
| 72 | } |
| 73 | |
| 74 | #[must_use] |
| 75 | pub fn matches_plan(&self, plan: &SharedPlanState) -> bool { |
| 76 | Arc::ptr_eq(&self.plan, plan) |
| 77 | } |
| 78 | |
| 79 | #[must_use] |
| 80 | pub fn has_operation_binding(&self, session_id: Option<&str>, external: &str) -> bool { |
| 81 | let active = lock_unpoisoned(&self.graph); |
| 82 | if let (Some(expected), Some(actual)) = (session_id, active.session_id.as_deref()) |
| 83 | && expected != actual |
| 84 | { |
| 85 | return false; |
| 86 | } |
| 87 | active.snapshot.as_ref().is_some_and(|graph| { |
| 88 | graph.nodes.iter().any(|node| { |
| 89 | node.binding |
| 90 | .as_ref() |
| 91 | .is_some_and(|binding| binding.external == external) |
| 92 | }) |
| 93 | }) |
| 94 | } |
| 95 | |
| 96 | /// Durable owner bindings that still require restore-time confirmation. |
| 97 | /// Terminal operations are excluded because their saved owner receipt is |
| 98 | /// already sufficient and must not be reopened by a missing live handle. |
| 99 | #[must_use] |
| 100 | pub fn reconcilable_durable_bindings(&self, session_id: Option<&str>) -> Vec<String> { |
| 101 | let active = lock_unpoisoned(&self.graph); |
| 102 | if let (Some(expected), Some(actual)) = (session_id, active.session_id.as_deref()) |
| 103 | && expected != actual |
| 104 | { |
| 105 | return Vec::new(); |
| 106 | } |
| 107 | active |
| 108 | .snapshot |
| 109 | .as_ref() |
| 110 | .into_iter() |
| 111 | .flat_map(|graph| graph.nodes.iter()) |
| 112 | .filter(|node| node.state.is_live() || node.state == NodeState::Stale) |
| 113 | .filter_map(|node| node.binding.as_ref()) |
| 114 | .filter(|binding| binding.durable) |
| 115 | .map(|binding| binding.external.clone()) |
| 116 | .collect() |
| 117 | } |
| 118 | |
| 119 | /// Register an Operation before its owner starts work. The operation is |
| 120 | /// first added inert, connected to an Objective/PlanStep, and only then |
| 121 | /// advanced to `Initializing`, so every reducer intermediate satisfies |
| 122 | /// the no-orphan invariant. |
| 123 | pub fn register_operation( |
| 124 | &self, |
| 125 | session_id: &str, |
| 126 | intent: OperationIntent, |
| 127 | ) -> Result<WorkNodeId, String> { |
| 128 | if !external_identity_is_well_formed(&intent.external) { |
| 129 | return Err(format!( |
| 130 | "invalid lifecycle binding external {:?}", |
| 131 | intent.external |
| 132 | )); |
| 133 | } |
| 134 | let todos = retry_lock(&self.todos, 100) |
| 135 | .ok_or_else(|| "To-do state is busy; operation was not registered".to_string())?; |
| 136 | let plan = retry_lock(&self.plan, 100) |
| 137 | .ok_or_else(|| "Plan state is busy; operation was not registered".to_string())?; |
| 138 | let mut active = lock_unpoisoned(&self.graph); |
| 139 | let base = graph_for_update(&mut active, session_id, &plan.snapshot(), &todos.snapshot())?; |
| 140 | if let Some(existing) = base.nodes.iter().find(|node| { |
| 141 | node.binding |
| 142 | .as_ref() |
| 143 | .is_some_and(|binding| binding.external == intent.external) |
| 144 | }) { |
| 145 | let binding = existing.binding.as_ref().expect("binding matched above"); |
| 146 | if binding.durable != intent.durable { |
| 147 | return Err(format!( |
| 148 | "lifecycle binding {} changed durability", |
| 149 | intent.external |
| 150 | )); |
| 151 | } |
| 152 | return Ok(existing.id.clone()); |
| 153 | } |
| 154 | |
| 155 | let mut graph = WorkGraph::from_snapshot(base); |
| 156 | let parent = operation_parent(&mut graph, session_id, &intent.source)?; |
| 157 | let node_id = WorkNodeId::derive(session_id, &format!("operation:{}", intent.external)); |
| 158 | let now = now_ms(); |
| 159 | apply_change( |
| 160 | &mut graph, |
| 161 | session_id, |
| 162 | &intent.source, |
| 163 | WorkGraphChange::AddNode { |
| 164 | node: WorkNode { |
| 165 | id: node_id.clone(), |
| 166 | kind: NodeKind::Operation, |
| 167 | title: bounded_operation_title(&intent.title), |
| 168 | state: NodeState::Ready, |
| 169 | acceptance: intent.acceptance, |
| 170 | binding: Some(OperationBinding { |
| 171 | external: intent.external, |
| 172 | durable: intent.durable, |
| 173 | last_observation: None, |
| 174 | }), |
| 175 | evidence: None, |
| 176 | provenance: Provenance::ToolUpdate { |
| 177 | tool: intent.source.clone(), |
| 178 | call_id: intent.call_id, |
| 179 | }, |
| 180 | created_at: now, |
| 181 | updated_at: now, |
| 182 | }, |
| 183 | }, |
| 184 | )?; |
| 185 | ensure_contains(&mut graph, session_id, &intent.source, &parent, &node_id)?; |
| 186 | let title = graph |
| 187 | .snapshot() |
| 188 | .node(&node_id) |
| 189 | .map(|node| node.title.clone()) |
| 190 | .ok_or_else(|| format!("operation {node_id} disappeared during registration"))?; |
| 191 | patch_existing_node( |
| 192 | &mut graph, |
| 193 | session_id, |
| 194 | &intent.source, |
| 195 | &node_id, |
| 196 | title, |
| 197 | NodeState::Initializing, |
| 198 | )?; |
| 199 | let next = graph.into_snapshot(); |
| 200 | validate_combined(&next, &project_plan(&next), &project_todos(&next))?; |
| 201 | active.snapshot = Some(next); |
| 202 | active.pending_publish = true; |
| 203 | Ok(node_id) |
| 204 | } |
| 205 | |
| 206 | /// Retain a crossed approval boundary as provenance attached to an |
| 207 | /// operation. The completed Approval node is historical evidence only: it |
| 208 | /// does not grant capabilities or change the owner's runtime authority. |
| 209 | pub fn record_operation_approval( |
| 210 | &self, |
| 211 | session_id: &str, |
| 212 | external: &str, |
| 213 | reference: &str, |
| 214 | source: &str, |
| 215 | call_id: &str, |
| 216 | ) -> Result<(), String> { |
| 217 | let todos = retry_lock(&self.todos, 100) |
| 218 | .ok_or_else(|| "To-do state is busy; approval was not recorded".to_string())?; |
| 219 | let plan = retry_lock(&self.plan, 100) |
| 220 | .ok_or_else(|| "Plan state is busy; approval was not recorded".to_string())?; |
| 221 | let mut active = lock_unpoisoned(&self.graph); |
| 222 | let base = graph_for_update(&mut active, session_id, &plan.snapshot(), &todos.snapshot())?; |
| 223 | let operation = base |
| 224 | .nodes |
| 225 | .iter() |
| 226 | .find(|node| { |
| 227 | node.kind == NodeKind::Operation |
| 228 | && node |
| 229 | .binding |
| 230 | .as_ref() |
| 231 | .is_some_and(|binding| binding.external == external) |
| 232 | }) |
| 233 | .map(|node| node.id.clone()) |
| 234 | .ok_or_else(|| format!("operation binding {external} is not registered"))?; |
| 235 | let approval = WorkNodeId::derive( |
| 236 | session_id, |
| 237 | &format!("approval:operation:{external}:{reference}"), |
| 238 | ); |
| 239 | let mut graph = WorkGraph::from_snapshot(base); |
| 240 | if graph.snapshot().node(&approval).is_none() { |
| 241 | let now = now_ms(); |
| 242 | apply_change( |
| 243 | &mut graph, |
| 244 | session_id, |
| 245 | source, |
| 246 | WorkGraphChange::AddNode { |
| 247 | node: WorkNode { |
| 248 | id: approval.clone(), |
| 249 | kind: NodeKind::Approval, |
| 250 | title: bounded_operation_title(&format!( |
| 251 | "verification approved: {reference}" |
| 252 | )), |
| 253 | state: NodeState::Completed, |
| 254 | acceptance: Vec::new(), |
| 255 | binding: None, |
| 256 | evidence: None, |
| 257 | provenance: Provenance::ToolUpdate { |
| 258 | tool: source.to_string(), |
| 259 | call_id: call_id.to_string(), |
| 260 | }, |
| 261 | created_at: now, |
| 262 | updated_at: now, |
| 263 | }, |
| 264 | }, |
| 265 | )?; |
| 266 | } |
| 267 | let edge = WorkEdgeId::derive( |
| 268 | session_id, |
| 269 | &format!( |
| 270 | "requires-approval:{}:{}", |
| 271 | operation.as_str(), |
| 272 | approval.as_str() |
| 273 | ), |
| 274 | ); |
| 275 | if graph.snapshot().edge(&edge).is_none() { |
| 276 | apply_change( |
| 277 | &mut graph, |
| 278 | session_id, |
| 279 | source, |
| 280 | WorkGraphChange::AddEdge { |
| 281 | edge: WorkEdge { |
| 282 | id: edge, |
| 283 | kind: EdgeKind::RequiresApproval, |
| 284 | from: operation, |
| 285 | to: approval, |
| 286 | }, |
| 287 | }, |
| 288 | )?; |
| 289 | } |
| 290 | let next = graph.into_snapshot(); |
| 291 | validate_combined(&next, &project_plan(&next), &project_todos(&next))?; |
| 292 | active.snapshot = Some(next); |
| 293 | active.pending_publish = true; |
| 294 | Ok(()) |
| 295 | } |
| 296 | |
| 297 | /// Apply one owner observation through the reducer. Unknown bindings are |
| 298 | /// rejected rather than materialized after the fact: spawn intent must be |
| 299 | /// registered before work begins. |
| 300 | pub fn reconcile_operation( |
| 301 | &self, |
| 302 | session_id: &str, |
| 303 | snapshot: OperationOwnerSnapshot, |
| 304 | ) -> Result<bool, String> { |
| 305 | let external = snapshot.external.clone(); |
| 306 | self.reconcile_observation(session_id, &external, snapshot.into_observation()) |
| 307 | } |
| 308 | |
| 309 | pub fn reconcile_observation( |
| 310 | &self, |
| 311 | session_id: &str, |
| 312 | external: &str, |
| 313 | observation: OperationObservation, |
| 314 | ) -> Result<bool, String> { |
| 315 | let mut active = lock_unpoisoned(&self.graph); |
| 316 | if active |
| 317 | .session_id |
| 318 | .as_deref() |
| 319 | .is_some_and(|id| id != session_id) |
| 320 | { |
| 321 | return Err(format!( |
| 322 | "lifecycle observation for session {session_id} does not match active session" |
| 323 | )); |
| 324 | } |
| 325 | let base = active |
| 326 | .snapshot |
| 327 | .clone() |
| 328 | .ok_or_else(|| "lifecycle observation arrived before graph registration".to_string())?; |
| 329 | let Some(next) = apply_observation_to_snapshot(&base, session_id, external, observation)? |
| 330 | else { |
| 331 | return Ok(false); |
| 332 | }; |
| 333 | active.snapshot = Some(next); |
| 334 | active.pending_publish = true; |
| 335 | Ok(true) |
| 336 | } |
| 337 | |
| 338 | /// Compact graph-owned continuity injected after context compaction. |
| 339 | /// It contains identities and state only; no raw output or reasoning. |
| 340 | #[must_use] |
| 341 | pub fn active_operation_summary(&self, session_id: Option<&str>) -> Option<String> { |
| 342 | let active = lock_unpoisoned(&self.graph); |
| 343 | if let (Some(expected), Some(actual)) = (session_id, active.session_id.as_deref()) |
| 344 | && expected != actual |
| 345 | { |
| 346 | return None; |
| 347 | } |
| 348 | let graph = active.snapshot.as_ref()?; |
| 349 | let operations = graph |
| 350 | .nodes |
| 351 | .iter() |
| 352 | .filter(|node| { |
| 353 | node.kind == NodeKind::Operation |
| 354 | && (node.state.is_live() || node.state == NodeState::Stale) |
| 355 | }) |
| 356 | .take(24) |
| 357 | .collect::<Vec<_>>(); |
| 358 | if operations.is_empty() { |
| 359 | return None; |
| 360 | } |
| 361 | let mut out = format!( |
| 362 | "{ACTIVE_OPERATION_SUMMARY_START}\n## Active Work Graph Operations\n\nOwner records remain authoritative; reconcile before acting on a restored operation.\n" |
| 363 | ); |
| 364 | for node in operations { |
| 365 | let external = node |
| 366 | .binding |
| 367 | .as_ref() |
| 368 | .map_or("unbound", |binding| binding.external.as_str()); |
| 369 | out.push_str(&format!( |
| 370 | "- `{}` - {} - {}\n", |
| 371 | external, |
| 372 | operation_state_label(node.state), |
| 373 | prompt_safe_title(&node.title) |
| 374 | )); |
| 375 | } |
| 376 | out.push_str(ACTIVE_OPERATION_SUMMARY_END); |
| 377 | Some(out) |
| 378 | } |
| 379 | |
| 380 | /// Record one reasoning-effort configuration change as bounded graph |
| 381 | /// activity. The event contains typed tiers and a non-secret route |
| 382 | /// identity only; there is no field capable of carrying reasoning text. |
| 383 | /// When live operations exist, the most recently updated one receives the |
| 384 | /// historical link. Later terminalization does not invalidate that link. |
| 385 | #[allow(clippy::too_many_arguments)] |
| 386 | pub fn record_reasoning_effort_change( |
| 387 | &self, |
| 388 | session_id: Option<&str>, |
| 389 | requested: ReasoningEffortTier, |
| 390 | effective: ReasoningEffortTier, |
| 391 | provider_kind: crate::config::ApiProvider, |
| 392 | provider: &str, |
| 393 | endpoint_identity: Option<&str>, |
| 394 | model: Option<&str>, |
| 395 | ) -> Result<Option<WorkNodeId>, String> { |
| 396 | let todos = retry_lock(&self.todos, 100) |
| 397 | .ok_or_else(|| "To-do state is busy; effort change was not recorded".to_string())?; |
| 398 | let plan = retry_lock(&self.plan, 100) |
| 399 | .ok_or_else(|| "Plan state is busy; effort change was not recorded".to_string())?; |
| 400 | let mut active = lock_unpoisoned(&self.graph); |
| 401 | let session_id = resolved_session_id(&active, session_id); |
| 402 | let base = graph_for_update( |
| 403 | &mut active, |
| 404 | &session_id, |
| 405 | &plan.snapshot(), |
| 406 | &todos.snapshot(), |
| 407 | )?; |
| 408 | let operation = base |
| 409 | .nodes |
| 410 | .iter() |
| 411 | .filter(|node| node.kind == NodeKind::Operation && node.state.is_live()) |
| 412 | .max_by(|left, right| { |
| 413 | left.updated_at |
| 414 | .cmp(&right.updated_at) |
| 415 | .then_with(|| left.id.as_str().cmp(right.id.as_str())) |
| 416 | }) |
| 417 | .map(|node| node.id.clone()); |
| 418 | let now = now_ms(); |
| 419 | let mut graph = WorkGraph::from_snapshot(base); |
| 420 | graph |
| 421 | .apply( |
| 422 | WorkGraphChange::RecordActivity { |
| 423 | event: WorkActivityEvent::ReasoningEffortChanged { |
| 424 | requested, |
| 425 | effective, |
| 426 | provider_kind: Some(provider_kind), |
| 427 | provider: provider.to_string(), |
| 428 | endpoint_identity: endpoint_identity.map(str::to_string), |
| 429 | model: model.map(str::to_string), |
| 430 | ts: now, |
| 431 | operation: operation.clone(), |
| 432 | }, |
| 433 | }, |
| 434 | ChangeCtx { |
| 435 | session_id, |
| 436 | now, |
| 437 | idempotency_key: None, |
| 438 | }, |
| 439 | ) |
| 440 | .map_err(|err| format!("reasoning effort: {err}"))?; |
| 441 | let next = graph.into_snapshot(); |
| 442 | validate_combined(&next, &project_plan(&next), &project_todos(&next))?; |
| 443 | active.snapshot = Some(next); |
| 444 | active.pending_publish = true; |
| 445 | Ok(operation) |
| 446 | } |
| 447 | |
| 448 | /// Apply an `update_plan` payload through the graph and publish both |
| 449 | /// legacy projections only after the candidate graph validates. |
| 450 | pub async fn apply_plan_update( |
| 451 | &self, |
| 452 | session_id: &str, |
| 453 | tool: &str, |
| 454 | plan: &PlanSnapshot, |
| 455 | ) -> Result<PlanSnapshot, String> { |
| 456 | let todos_guard = self.todos.lock().await; |
| 457 | let plan_guard = self.plan.lock().await; |
| 458 | let mut active = lock_unpoisoned(&self.graph); |
| 459 | let base = graph_for_update( |
| 460 | &mut active, |
| 461 | session_id, |
| 462 | &plan_guard.snapshot(), |
| 463 | &todos_guard.snapshot(), |
| 464 | )?; |
| 465 | let next = update_plan_graph(base, session_id, tool, plan)?; |
| 466 | let derived_plan = project_plan(&next); |
| 467 | let derived_todos = project_todos(&next); |
| 468 | let next_plan = PlanState::from_snapshot(&derived_plan); |
| 469 | let next_todos = TodoList::from_snapshot(&derived_todos)?; |
| 470 | validate_combined(&next, &next_plan.snapshot(), &next_todos.snapshot())?; |
| 471 | active.snapshot = Some(next); |
| 472 | active.pending_publish = true; |
| 473 | Ok(derived_plan) |
| 474 | } |
| 475 | |
| 476 | /// Apply a legacy To-do/checklist payload through the graph and publish |
| 477 | /// both projections from the committed candidate. |
| 478 | pub async fn apply_todo_update( |
| 479 | &self, |
| 480 | session_id: &str, |
| 481 | tool: &str, |
| 482 | todos: &TodoListSnapshot, |
| 483 | ) -> Result<TodoListSnapshot, String> { |
| 484 | let todos_guard = self.todos.lock().await; |
| 485 | let plan_guard = self.plan.lock().await; |
| 486 | let mut active = lock_unpoisoned(&self.graph); |
| 487 | let base = graph_for_update( |
| 488 | &mut active, |
| 489 | session_id, |
| 490 | &plan_guard.snapshot(), |
| 491 | &todos_guard.snapshot(), |
| 492 | )?; |
| 493 | let next = update_todo_graph(base, session_id, tool, todos)?; |
| 494 | let derived_plan = project_plan(&next); |
| 495 | let derived_todos = project_todos(&next); |
| 496 | let next_plan = PlanState::from_snapshot(&derived_plan); |
| 497 | let next_todos = TodoList::from_snapshot(&derived_todos)?; |
| 498 | validate_combined(&next, &next_plan.snapshot(), &next_todos.snapshot())?; |
| 499 | active.snapshot = Some(next); |
| 500 | active.pending_publish = true; |
| 501 | Ok(derived_todos) |
| 502 | } |
| 503 | |
| 504 | /// Publish the latest validated legacy views after the caller has queued |
| 505 | /// their graph-backed session/checkpoint write. |
| 506 | pub async fn publish_pending(&self) -> Result<bool, String> { |
| 507 | let mut todos = self.todos.lock().await; |
| 508 | let mut plan = self.plan.lock().await; |
| 509 | let mut active = lock_unpoisoned(&self.graph); |
| 510 | if !active.pending_publish { |
| 511 | return Ok(false); |
| 512 | } |
| 513 | let graph = active |
| 514 | .snapshot |
| 515 | .as_ref() |
| 516 | .ok_or_else(|| "pending Work projection has no graph".to_string())?; |
| 517 | let derived_plan = project_plan(graph); |
| 518 | let derived_todos = project_todos(graph); |
| 519 | let next_plan = PlanState::from_snapshot(&derived_plan); |
| 520 | let next_todos = TodoList::from_snapshot(&derived_todos)?; |
| 521 | validate_combined(graph, &next_plan.snapshot(), &next_todos.snapshot())?; |
| 522 | *plan = next_plan; |
| 523 | *todos = next_todos; |
| 524 | active.pending_publish = false; |
| 525 | Ok(true) |
| 526 | } |
| 527 | |
| 528 | /// Synchronous counterpart for explicit save/rename/fork commands that |
| 529 | /// have already completed their atomic disk write. |
| 530 | pub fn publish_pending_sync(&self) -> Result<bool, String> { |
| 531 | let mut todos = retry_lock(&self.todos, 100).ok_or_else(|| { |
| 532 | "To-do state is busy; saved Work views were not published".to_string() |
| 533 | })?; |
| 534 | let mut plan = retry_lock(&self.plan, 100) |
| 535 | .ok_or_else(|| "Plan state is busy; saved Work views were not published".to_string())?; |
| 536 | let mut active = lock_unpoisoned(&self.graph); |
| 537 | if !active.pending_publish { |
| 538 | return Ok(false); |
| 539 | } |
| 540 | let graph = active |
| 541 | .snapshot |
| 542 | .as_ref() |
| 543 | .ok_or_else(|| "pending Work projection has no graph".to_string())?; |
| 544 | let derived_plan = project_plan(graph); |
| 545 | let derived_todos = project_todos(graph); |
| 546 | let next_plan = PlanState::from_snapshot(&derived_plan); |
| 547 | let next_todos = TodoList::from_snapshot(&derived_todos)?; |
| 548 | validate_combined(graph, &next_plan.snapshot(), &next_todos.snapshot())?; |
| 549 | *plan = next_plan; |
| 550 | *todos = next_todos; |
| 551 | active.pending_publish = false; |
| 552 | Ok(true) |
| 553 | } |
| 554 | |
| 555 | #[must_use] |
| 556 | pub fn has_pending_publish(&self) -> bool { |
| 557 | lock_unpoisoned(&self.graph).pending_publish |
| 558 | } |
| 559 | |
| 560 | /// Latest graph-derived To-do view, including an unpublished transaction. |
| 561 | pub async fn current_todos(&self) -> Result<TodoListSnapshot, String> { |
| 562 | let projected = { |
| 563 | let active = lock_unpoisoned(&self.graph); |
| 564 | active.snapshot.as_ref().map(project_todos) |
| 565 | }; |
| 566 | if let Some(projected) = projected { |
| 567 | return Ok(projected); |
| 568 | } |
| 569 | Ok(self.todos.lock().await.snapshot()) |
| 570 | } |
| 571 | |
| 572 | /// Capture a persistence-ready graph plus fully populated old views. |
| 573 | /// Legacy-only in-memory state is imported once and normalized in place. |
| 574 | pub fn capture(&self, session_id: Option<&str>) -> Result<Option<WorkRuntimeSnapshot>, String> { |
| 575 | self.capture_with_retries(session_id, 100) |
| 576 | } |
| 577 | |
| 578 | /// Non-blocking capture for the render/event loop. |
| 579 | pub fn try_capture( |
| 580 | &self, |
| 581 | session_id: Option<&str>, |
| 582 | ) -> Result<Option<WorkRuntimeSnapshot>, String> { |
| 583 | self.capture_with_retries(session_id, 1) |
| 584 | } |
| 585 | |
| 586 | fn capture_with_retries( |
| 587 | &self, |
| 588 | session_id: Option<&str>, |
| 589 | retries: u32, |
| 590 | ) -> Result<Option<WorkRuntimeSnapshot>, String> { |
| 591 | let todos = retry_lock(&self.todos, retries) |
| 592 | .ok_or_else(|| "To-do state is busy; try saving again".to_string())?; |
| 593 | let plan = retry_lock(&self.plan, retries) |
| 594 | .ok_or_else(|| "Plan state is busy; try saving again".to_string())?; |
| 595 | let mut active = lock_unpoisoned(&self.graph); |
| 596 | let todos_snapshot = todos.snapshot(); |
| 597 | let plan_snapshot = plan.snapshot(); |
| 598 | if todos_snapshot.is_empty() |
| 599 | && plan_snapshot.is_empty() |
| 600 | && active |
| 601 | .snapshot |
| 602 | .as_ref() |
| 603 | .is_none_or(WorkGraphSnapshot::is_empty) |
| 604 | { |
| 605 | return Ok(None); |
| 606 | } |
| 607 | let had_graph = active.snapshot.is_some(); |
| 608 | let had_pending_publish = active.pending_publish; |
| 609 | let session_id = resolved_session_id(&active, session_id); |
| 610 | let graph = graph_for_update(&mut active, &session_id, &plan_snapshot, &todos_snapshot)?; |
| 611 | let derived_plan = project_plan(&graph); |
| 612 | let derived_todos = project_todos(&graph); |
| 613 | validate_combined(&graph, &derived_plan, &derived_todos)?; |
| 614 | if had_graph |
| 615 | && !had_pending_publish |
| 616 | && (derived_plan != plan_snapshot || derived_todos != todos_snapshot) |
| 617 | { |
| 618 | return Err("live Work Graph and legacy views disagree".to_string()); |
| 619 | } |
| 620 | active.snapshot = Some(graph.clone()); |
| 621 | if !had_graph { |
| 622 | active.pending_publish = true; |
| 623 | } |
| 624 | Ok(Some(WorkRuntimeSnapshot { |
| 625 | graph, |
| 626 | todos: derived_todos, |
| 627 | plan: derived_plan, |
| 628 | })) |
| 629 | } |
| 630 | |
| 631 | /// Validate and atomically activate persisted state. Sessions without a |
| 632 | /// graph are deterministically imported from their complete old views. |
| 633 | pub fn restore( |
| 634 | &self, |
| 635 | session_id: &str, |
| 636 | graph: Option<&WorkGraphSnapshot>, |
| 637 | todos: &TodoListSnapshot, |
| 638 | plan: &PlanSnapshot, |
| 639 | ) -> Result<Option<WorkRuntimeSnapshot>, String> { |
| 640 | self.restore_internal(session_id, graph, todos, plan, None) |
| 641 | } |
| 642 | |
| 643 | /// Restore a saved graph and reconcile workspace-scoped durable owners as |
| 644 | /// one candidate transaction. No live state changes until the restored |
| 645 | /// graph, owner observations, and both legacy projections all validate. |
| 646 | pub fn restore_with_workspace_owner_bindings( |
| 647 | &self, |
| 648 | session_id: &str, |
| 649 | workspace: &Path, |
| 650 | graph: Option<&WorkGraphSnapshot>, |
| 651 | todos: &TodoListSnapshot, |
| 652 | plan: &PlanSnapshot, |
| 653 | ) -> Result<Option<WorkRuntimeSnapshot>, String> { |
| 654 | self.restore_internal(session_id, graph, todos, plan, Some(workspace)) |
| 655 | } |
| 656 | |
| 657 | fn restore_internal( |
| 658 | &self, |
| 659 | session_id: &str, |
| 660 | graph: Option<&WorkGraphSnapshot>, |
| 661 | todos: &TodoListSnapshot, |
| 662 | plan: &PlanSnapshot, |
| 663 | workspace: Option<&Path>, |
| 664 | ) -> Result<Option<WorkRuntimeSnapshot>, String> { |
| 665 | let had_graph = graph.is_some(); |
| 666 | let graph = match graph { |
| 667 | Some(graph) => { |
| 668 | validate(graph).map_err(|err| err.to_string())?; |
| 669 | graph.clone() |
| 670 | } |
| 671 | None if todos.is_empty() && plan.is_empty() => WorkGraphSnapshot::new(), |
| 672 | None => import_legacy(session_id, plan, todos)?, |
| 673 | }; |
| 674 | let (mut graph, reconciled_ephemeral) = |
| 675 | mark_restored_ephemeral_operations_stale(graph, session_id)?; |
| 676 | let reconciled_workspace = if let Some(workspace) = workspace { |
| 677 | let (reconciled, changed) = reconcile_workspace_snapshot(graph, session_id, workspace)?; |
| 678 | graph = reconciled; |
| 679 | changed > 0 |
| 680 | } else { |
| 681 | false |
| 682 | }; |
| 683 | let derived_plan = project_plan(&graph); |
| 684 | let derived_todos = project_todos(&graph); |
| 685 | if graph.is_empty() { |
| 686 | if !todos.is_empty() || !plan.is_empty() { |
| 687 | return Err("empty Work Graph cannot carry non-empty legacy views".to_string()); |
| 688 | } |
| 689 | } else if graph.import_digest.is_some() && graph.compat.is_empty() { |
| 690 | return Err("imported Work Graph is missing compatibility projections".to_string()); |
| 691 | } |
| 692 | validate_combined(&graph, &derived_plan, &derived_todos)?; |
| 693 | if had_graph && (&derived_plan != plan || &derived_todos != todos) { |
| 694 | return Err("persisted Work Graph and legacy views disagree".to_string()); |
| 695 | } |
| 696 | let next_plan = PlanState::from_snapshot(&derived_plan); |
| 697 | let next_todos = TodoList::from_snapshot(&derived_todos)?; |
| 698 | let mut todos_guard = retry_lock(&self.todos, 100) |
| 699 | .ok_or_else(|| "To-do state is busy; session was not restored".to_string())?; |
| 700 | let mut plan_guard = retry_lock(&self.plan, 100) |
| 701 | .ok_or_else(|| "Plan state is busy; session was not restored".to_string())?; |
| 702 | let mut active = lock_unpoisoned(&self.graph); |
| 703 | *todos_guard = next_todos; |
| 704 | *plan_guard = next_plan; |
| 705 | active.session_id = Some(session_id.to_string()); |
| 706 | active.snapshot = Some(graph.clone()); |
| 707 | // A legacy load has already restored its complete old views, but its |
| 708 | // newly imported graph still needs one acknowledged graph-bearing |
| 709 | // write (and pre-import archive) before the migration is settled. |
| 710 | active.pending_publish = |
| 711 | reconciled_ephemeral || reconciled_workspace || (!had_graph && !graph.is_empty()); |
| 712 | if graph.is_empty() { |
| 713 | Ok(None) |
| 714 | } else { |
| 715 | Ok(Some(WorkRuntimeSnapshot { |
| 716 | graph, |
| 717 | todos: derived_todos, |
| 718 | plan: derived_plan, |
| 719 | })) |
| 720 | } |
| 721 | } |
| 722 | |
| 723 | /// Reconcile restored Fleet and Lane bindings from their durable owners. |
| 724 | /// Missing records fail toward `Stale`; process probes never override the |
| 725 | /// replayed ledger/registry state. |
| 726 | pub fn reconcile_workspace_owner_bindings( |
| 727 | &self, |
| 728 | session_id: &str, |
| 729 | workspace: &Path, |
| 730 | ) -> Result<usize, String> { |
| 731 | let (base, revision) = { |
| 732 | let active = lock_unpoisoned(&self.graph); |
| 733 | if active.session_id.as_deref() != Some(session_id) { |
| 734 | return Err(format!( |
| 735 | "workspace owner reconciliation does not match active session {session_id}" |
| 736 | )); |
| 737 | } |
| 738 | let base = active |
| 739 | .snapshot |
| 740 | .clone() |
| 741 | .ok_or_else(|| "workspace owner reconciliation has no active graph".to_string())?; |
| 742 | let revision = base.revision; |
| 743 | (base, revision) |
| 744 | }; |
| 745 | let (next, changed) = reconcile_workspace_snapshot(base, session_id, workspace)?; |
| 746 | if changed == 0 { |
| 747 | return Ok(0); |
| 748 | } |
| 749 | let mut active = lock_unpoisoned(&self.graph); |
| 750 | if active.session_id.as_deref() != Some(session_id) |
| 751 | || active.snapshot.as_ref().map(|graph| graph.revision) != Some(revision) |
| 752 | { |
| 753 | return Err("Work Graph changed during workspace owner reconciliation".to_string()); |
| 754 | } |
| 755 | active.snapshot = Some(next); |
| 756 | active.pending_publish = true; |
| 757 | Ok(changed) |
| 758 | } |
| 759 | |
| 760 | pub fn clear(&self, session_id: Option<&str>) -> bool { |
| 761 | let Some(mut todos) = retry_lock(&self.todos, 100) else { |
| 762 | return false; |
| 763 | }; |
| 764 | let Some(mut plan) = retry_lock(&self.plan, 100) else { |
| 765 | return false; |
| 766 | }; |
| 767 | let mut active = lock_unpoisoned(&self.graph); |
| 768 | todos.clear(); |
| 769 | *plan = PlanState::default(); |
| 770 | active.session_id = Some(resolved_session_id(&active, session_id)); |
| 771 | active.snapshot = Some(WorkGraphSnapshot::new()); |
| 772 | active.pending_publish = false; |
| 773 | true |
| 774 | } |
| 775 | } |
| 776 | |
| 777 | fn apply_observation_to_snapshot( |
| 778 | base: &WorkGraphSnapshot, |
| 779 | session_id: &str, |
| 780 | external: &str, |
| 781 | observation: OperationObservation, |
| 782 | ) -> Result<Option<WorkGraphSnapshot>, String> { |
| 783 | let node = base |
| 784 | .nodes |
| 785 | .iter() |
| 786 | .find(|node| { |
| 787 | node.binding |
| 788 | .as_ref() |
| 789 | .is_some_and(|binding| binding.external == external) |
| 790 | }) |
| 791 | .ok_or_else(|| format!("operation binding {external} is not registered"))?; |
| 792 | let stale_owner_recovery = if let OperationObservation::OwnerReported { |
| 793 | state, seq, output, .. |
| 794 | } = &observation |
| 795 | && let Some(previous) = node |
| 796 | .binding |
| 797 | .as_ref() |
| 798 | .and_then(|binding| binding.last_observation.as_ref()) |
| 799 | { |
| 800 | if *seq < previous.seq { |
| 801 | return Err(format!( |
| 802 | "operation owner {external} sequence regressed from {} to {seq}", |
| 803 | previous.seq |
| 804 | )); |
| 805 | } |
| 806 | if *seq == previous.seq { |
| 807 | if node.state != NodeState::Stale { |
| 808 | return Ok(None); |
| 809 | } |
| 810 | if *state != previous.owner_state || *output != previous.output { |
| 811 | return Err(format!( |
| 812 | "operation owner {external} changed observation at sequence {seq}" |
| 813 | )); |
| 814 | } |
| 815 | true |
| 816 | } else { |
| 817 | false |
| 818 | } |
| 819 | } else { |
| 820 | false |
| 821 | }; |
| 822 | if matches!(&observation, OperationObservation::OwnerMissing { .. }) |
| 823 | && node.state == NodeState::Stale |
| 824 | { |
| 825 | return Ok(None); |
| 826 | } |
| 827 | let idempotency_key = match &observation { |
| 828 | OperationObservation::OwnerReported { .. } if stale_owner_recovery => None, |
| 829 | OperationObservation::OwnerReported { seq, .. } => Some(IdempotencyKey { |
| 830 | binding: BindingId::derive(session_id, &format!("binding:{external}")), |
| 831 | seq: *seq, |
| 832 | }), |
| 833 | OperationObservation::OwnerMissing { .. } | OperationObservation::CancelUpdate { .. } => { |
| 834 | None |
| 835 | } |
| 836 | }; |
| 837 | let (next, receipt) = super::reducer::apply( |
| 838 | base, |
| 839 | WorkGraphChange::ReconcileOperation { |
| 840 | node: node.id.clone(), |
| 841 | obs: observation, |
| 842 | }, |
| 843 | ChangeCtx { |
| 844 | session_id: session_id.to_string(), |
| 845 | now: now_ms(), |
| 846 | idempotency_key, |
| 847 | }, |
| 848 | ) |
| 849 | .map_err(|err| format!("runtime reconcile: {err}"))?; |
| 850 | Ok((!receipt.no_op).then_some(next)) |
| 851 | } |
| 852 | |
| 853 | fn reconcile_workspace_snapshot( |
| 854 | mut graph: WorkGraphSnapshot, |
| 855 | session_id: &str, |
| 856 | workspace: &Path, |
| 857 | ) -> Result<(WorkGraphSnapshot, usize), String> { |
| 858 | let candidates = graph |
| 859 | .nodes |
| 860 | .iter() |
| 861 | .filter(|node| node.state.is_live() || node.state == NodeState::Stale) |
| 862 | .filter_map(|node| node.binding.as_ref()) |
| 863 | .filter(|binding| binding.durable) |
| 864 | .map(|binding| binding.external.clone()) |
| 865 | .filter(|external| external.starts_with("fleet:") || external.starts_with("lane:")) |
| 866 | .collect::<Vec<_>>(); |
| 867 | if candidates.is_empty() { |
| 868 | return Ok((graph, 0)); |
| 869 | } |
| 870 | |
| 871 | let fleet = candidates |
| 872 | .iter() |
| 873 | .any(|external| external.starts_with("fleet:")) |
| 874 | .then(|| FleetLedger::open(workspace).and_then(|ledger| ledger.rebuild_state())); |
| 875 | if let Some(Err(err)) = fleet.as_ref() { |
| 876 | tracing::warn!( |
| 877 | workspace = %workspace.display(), |
| 878 | error = %err, |
| 879 | "Fleet owner store could not be replayed; restored bindings will be stale" |
| 880 | ); |
| 881 | } |
| 882 | let lanes = candidates |
| 883 | .iter() |
| 884 | .any(|external| external.starts_with("lane:")) |
| 885 | .then(LaneRegistry::open_default); |
| 886 | if let Some(Err(err)) = lanes.as_ref() { |
| 887 | tracing::warn!( |
| 888 | error = %err, |
| 889 | "Lane owner registry could not be opened; restored bindings will be stale" |
| 890 | ); |
| 891 | } |
| 892 | let observed_at = now_ms(); |
| 893 | let mut changed = 0usize; |
| 894 | for external in candidates { |
| 895 | let observation = if let Some(rest) = external.strip_prefix("fleet:") { |
| 896 | rest.split_once('/') |
| 897 | .and_then(|(run_id, task_id)| { |
| 898 | fleet |
| 899 | .as_ref() |
| 900 | .and_then(|state| state.as_ref().ok()) |
| 901 | .and_then(|state| state.tasks.get(&format!("{run_id}:{task_id}"))) |
| 902 | }) |
| 903 | .map(|record| fleet_task_owner_snapshot(record, observed_at).into_observation()) |
| 904 | .unwrap_or(OperationObservation::OwnerMissing { |
| 905 | checked_at: observed_at, |
| 906 | }) |
| 907 | } else if let Some(lane_id) = external.strip_prefix("lane:") { |
| 908 | lanes |
| 909 | .as_ref() |
| 910 | .and_then(|registry| registry.as_ref().ok()) |
| 911 | .and_then(|registry| registry.load(lane_id).ok()) |
| 912 | .map(|record| lane_owner_snapshot(&record, observed_at).into_observation()) |
| 913 | .unwrap_or(OperationObservation::OwnerMissing { |
| 914 | checked_at: observed_at, |
| 915 | }) |
| 916 | } else { |
| 917 | continue; |
| 918 | }; |
| 919 | if let Some(next) = |
| 920 | apply_observation_to_snapshot(&graph, session_id, &external, observation)? |
| 921 | { |
| 922 | graph = next; |
| 923 | changed = changed.saturating_add(1); |
| 924 | } |
| 925 | } |
| 926 | Ok((graph, changed)) |
| 927 | } |
| 928 | |
| 929 | fn operation_parent( |
| 930 | graph: &mut WorkGraph, |
| 931 | session_id: &str, |
| 932 | source: &str, |
| 933 | ) -> Result<WorkNodeId, String> { |
| 934 | if let Some(parent) = graph |
| 935 | .snapshot() |
| 936 | .nodes |
| 937 | .iter() |
| 938 | .find(|node| node.kind == NodeKind::PlanStep && node.state == NodeState::Active) |
| 939 | .or_else(|| { |
| 940 | graph |
| 941 | .snapshot() |
| 942 | .nodes |
| 943 | .iter() |
| 944 | .find(|node| node.kind == NodeKind::PlanStep && node.state == NodeState::Ready) |
| 945 | }) |
| 946 | .or_else(|| { |
| 947 | graph |
| 948 | .snapshot() |
| 949 | .nodes |
| 950 | .iter() |
| 951 | .find(|node| node.kind == NodeKind::Objective) |
| 952 | }) |
| 953 | { |
| 954 | return Ok(parent.id.clone()); |
| 955 | } |
| 956 | |
| 957 | let now = now_ms(); |
| 958 | let id = WorkNodeId::derive(session_id, "objective:runtime-operations"); |
| 959 | apply_change( |
| 960 | graph, |
| 961 | session_id, |
| 962 | source, |
| 963 | WorkGraphChange::AddNode { |
| 964 | node: WorkNode { |
| 965 | id: id.clone(), |
| 966 | kind: NodeKind::Objective, |
| 967 | title: "Runtime operations".to_string(), |
| 968 | state: NodeState::Ready, |
| 969 | acceptance: Vec::new(), |
| 970 | binding: None, |
| 971 | evidence: None, |
| 972 | provenance: Provenance::RuntimeReconcile { |
| 973 | source: source.to_string(), |
| 974 | observed_at: now, |
| 975 | }, |
| 976 | created_at: now, |
| 977 | updated_at: now, |
| 978 | }, |
| 979 | }, |
| 980 | )?; |
| 981 | Ok(id) |
| 982 | } |
| 983 | |
| 984 | fn mark_restored_ephemeral_operations_stale( |
| 985 | graph: WorkGraphSnapshot, |
| 986 | session_id: &str, |
| 987 | ) -> Result<(WorkGraphSnapshot, bool), String> { |
| 988 | let candidates = graph |
| 989 | .nodes |
| 990 | .iter() |
| 991 | .filter_map(|node| { |
| 992 | let binding = node.binding.as_ref()?; |
| 993 | (!binding.durable && node.state.is_live()).then(|| node.id.clone()) |
| 994 | }) |
| 995 | .collect::<Vec<_>>(); |
| 996 | if candidates.is_empty() { |
| 997 | return Ok((graph, false)); |
| 998 | } |
| 999 | let mut graph = WorkGraph::from_snapshot(graph); |
| 1000 | for node in candidates { |
| 1001 | graph |
| 1002 | .apply( |
| 1003 | WorkGraphChange::ReconcileOperation { |
| 1004 | node, |
| 1005 | obs: OperationObservation::OwnerMissing { |
| 1006 | checked_at: now_ms(), |
| 1007 | }, |
| 1008 | }, |
| 1009 | ChangeCtx { |
| 1010 | session_id: session_id.to_string(), |
| 1011 | now: now_ms(), |
| 1012 | idempotency_key: None, |
| 1013 | }, |
| 1014 | ) |
| 1015 | .map_err(|err| format!("restart reconcile: {err}"))?; |
| 1016 | } |
| 1017 | Ok((graph.into_snapshot(), true)) |
| 1018 | } |
| 1019 | |
| 1020 | fn bounded_operation_title(title: &str) -> String { |
| 1021 | let normalized = title.split_whitespace().collect::<Vec<_>>().join(" "); |
| 1022 | let mut chars = normalized.chars(); |
| 1023 | let bounded = chars.by_ref().take(180).collect::<String>(); |
| 1024 | if chars.next().is_some() { |
| 1025 | format!("{bounded}...") |
| 1026 | } else if bounded.is_empty() { |
| 1027 | "Runtime operation".to_string() |
| 1028 | } else { |
| 1029 | bounded |
| 1030 | } |
| 1031 | } |
| 1032 | |
| 1033 | fn prompt_safe_title(title: &str) -> String { |
| 1034 | bounded_operation_title(&title.replace('`', "'")) |
| 1035 | } |
| 1036 | |
| 1037 | const fn operation_state_label(state: NodeState) -> &'static str { |
| 1038 | match state { |
| 1039 | NodeState::Ready => "ready", |
| 1040 | NodeState::Initializing => "initializing", |
| 1041 | NodeState::Active => "running", |
| 1042 | NodeState::Waiting => "waiting", |
| 1043 | NodeState::Blocked => "blocked", |
| 1044 | NodeState::Completed => "completed", |
| 1045 | NodeState::Verified => "verified", |
| 1046 | NodeState::Stale => "stale", |
| 1047 | NodeState::Superseded => "superseded", |
| 1048 | NodeState::Cancelled => "cancelled", |
| 1049 | NodeState::Failed => "failed", |
| 1050 | } |
| 1051 | } |
| 1052 | |
| 1053 | fn graph_for_update( |
| 1054 | active: &mut ActiveGraph, |
| 1055 | session_id: &str, |
| 1056 | plan: &PlanSnapshot, |
| 1057 | todos: &TodoListSnapshot, |
| 1058 | ) -> Result<WorkGraphSnapshot, String> { |
| 1059 | match active.session_id.as_deref() { |
| 1060 | // App session transitions are already blocked while runtime work is |
| 1061 | // active. Rebind the authority namespace without re-keying graph IDs |
| 1062 | // so save-as/fork/new-session flows keep one coherent snapshot. |
| 1063 | Some(active_id) if active_id != session_id => { |
| 1064 | active.session_id = Some(session_id.to_string()); |
| 1065 | } |
| 1066 | None => active.session_id = Some(session_id.to_string()), |
| 1067 | Some(_) => {} |
| 1068 | } |
| 1069 | if let Some(snapshot) = active.snapshot.as_ref() { |
| 1070 | validate(snapshot).map_err(|err| err.to_string())?; |
| 1071 | return Ok(snapshot.clone()); |
| 1072 | } |
| 1073 | let graph = if plan.is_empty() && todos.is_empty() { |
| 1074 | WorkGraphSnapshot::new() |
| 1075 | } else { |
| 1076 | import_legacy(session_id, plan, todos)? |
| 1077 | }; |
| 1078 | active.snapshot = Some(graph.clone()); |
| 1079 | Ok(graph) |
| 1080 | } |
| 1081 | |
| 1082 | fn update_plan_graph( |
| 1083 | base: WorkGraphSnapshot, |
| 1084 | session_id: &str, |
| 1085 | tool: &str, |
| 1086 | plan: &PlanSnapshot, |
| 1087 | ) -> Result<WorkGraphSnapshot, String> { |
| 1088 | let mut graph = WorkGraph::from_snapshot(base); |
| 1089 | let objective = ensure_objective(&mut graph, session_id, tool, plan)?; |
| 1090 | let desired_active_alias = plan.items.iter().enumerate().find_map(|(index, item)| { |
| 1091 | (item.status == StepStatus::InProgress) |
| 1092 | .then(|| graph.snapshot().compat.plan_order.get(index).cloned()) |
| 1093 | .flatten() |
| 1094 | .filter(|node| { |
| 1095 | graph |
| 1096 | .snapshot() |
| 1097 | .compat |
| 1098 | .todos |
| 1099 | .iter() |
| 1100 | .any(|binding| &binding.node == node) |
| 1101 | }) |
| 1102 | }); |
| 1103 | if desired_active_alias.is_some() { |
| 1104 | deactivate_projected_todos(&mut graph, session_id, tool)?; |
| 1105 | } |
| 1106 | |
| 1107 | let mut order = Vec::with_capacity(plan.items.len()); |
| 1108 | for (index, item) in plan.items.iter().enumerate() { |
| 1109 | let id = graph |
| 1110 | .snapshot() |
| 1111 | .compat |
| 1112 | .plan_order |
| 1113 | .get(index) |
| 1114 | .cloned() |
| 1115 | .unwrap_or_else(|| WorkNodeId::derive(session_id, &format!("plan:{index}"))); |
| 1116 | let provenance = tool_provenance(graph.snapshot(), tool); |
| 1117 | upsert_node( |
| 1118 | &mut graph, |
| 1119 | session_id, |
| 1120 | tool, |
| 1121 | WorkNode { |
| 1122 | id: id.clone(), |
| 1123 | kind: NodeKind::PlanStep, |
| 1124 | title: item.step.trim().to_string(), |
| 1125 | state: plan_node_state(&item.status), |
| 1126 | acceptance: Vec::new(), |
| 1127 | binding: None, |
| 1128 | evidence: None, |
| 1129 | provenance, |
| 1130 | created_at: now_ms(), |
| 1131 | updated_at: now_ms(), |
| 1132 | }, |
| 1133 | )?; |
| 1134 | ensure_contains(&mut graph, session_id, tool, &objective, &id)?; |
| 1135 | order.push(id); |
| 1136 | } |
| 1137 | let mut compat = graph.snapshot().compat.clone(); |
| 1138 | compat.plan = CompatPlanMetadata::from_plan_snapshot(plan); |
| 1139 | compat.plan_order = order; |
| 1140 | compat.todos.retain(|binding| { |
| 1141 | binding.plan_index.is_none_or(|index| { |
| 1142 | usize::try_from(index) |
| 1143 | .ok() |
| 1144 | .is_some_and(|i| i < plan.items.len()) |
| 1145 | }) |
| 1146 | }); |
| 1147 | for binding in &mut compat.todos { |
| 1148 | if let Some(index) = binding.plan_index |
| 1149 | && let Some(node) = compat |
| 1150 | .plan_order |
| 1151 | .get(usize::try_from(index).unwrap_or(usize::MAX)) |
| 1152 | { |
| 1153 | binding.node.clone_from(node); |
| 1154 | } |
| 1155 | } |
| 1156 | apply_change( |
| 1157 | &mut graph, |
| 1158 | session_id, |
| 1159 | tool, |
| 1160 | WorkGraphChange::ReplaceCompatProjection { compat }, |
| 1161 | )?; |
| 1162 | Ok(graph.into_snapshot()) |
| 1163 | } |
| 1164 | |
| 1165 | fn update_todo_graph( |
| 1166 | base: WorkGraphSnapshot, |
| 1167 | session_id: &str, |
| 1168 | tool: &str, |
| 1169 | todos: &TodoListSnapshot, |
| 1170 | ) -> Result<WorkGraphSnapshot, String> { |
| 1171 | let mut graph = WorkGraph::from_snapshot(base); |
| 1172 | deactivate_projected_todos(&mut graph, session_id, tool)?; |
| 1173 | let current_plan = project_plan(graph.snapshot()); |
| 1174 | let objective = ensure_objective(&mut graph, session_id, tool, ¤t_plan)?; |
| 1175 | let plan_order = graph.snapshot().compat.plan_order.clone(); |
| 1176 | let mut bindings = Vec::with_capacity(todos.items.len()); |
| 1177 | for item in &todos.items { |
| 1178 | let title = item.content.trim().to_string(); |
| 1179 | let alias = graph |
| 1180 | .snapshot() |
| 1181 | .compat |
| 1182 | .todos |
| 1183 | .iter() |
| 1184 | .find(|binding| binding.legacy_id == item.id) |
| 1185 | .and_then(|binding| { |
| 1186 | binding |
| 1187 | .plan_index |
| 1188 | .map(|index| (index, binding.node.clone())) |
| 1189 | }) |
| 1190 | .filter(|(index, node)| { |
| 1191 | plan_order.get(usize::try_from(*index).unwrap_or(usize::MAX)) == Some(node) |
| 1192 | }); |
| 1193 | let (node, plan_index) = if let Some((index, node)) = alias { |
| 1194 | patch_existing_node( |
| 1195 | &mut graph, |
| 1196 | session_id, |
| 1197 | tool, |
| 1198 | &node, |
| 1199 | title, |
| 1200 | todo_node_state(item.status), |
| 1201 | )?; |
| 1202 | (node, Some(index)) |
| 1203 | } else { |
| 1204 | let node = graph |
| 1205 | .snapshot() |
| 1206 | .compat |
| 1207 | .todos |
| 1208 | .iter() |
| 1209 | .find(|binding| binding.legacy_id == item.id && binding.plan_index.is_none()) |
| 1210 | .map(|binding| binding.node.clone()) |
| 1211 | .unwrap_or_else(|| WorkNodeId::derive(session_id, &format!("todo:{}", item.id))); |
| 1212 | let desired = todo_node_state(item.status); |
| 1213 | let provenance = tool_provenance(graph.snapshot(), tool); |
| 1214 | upsert_node( |
| 1215 | &mut graph, |
| 1216 | session_id, |
| 1217 | tool, |
| 1218 | WorkNode { |
| 1219 | id: node.clone(), |
| 1220 | kind: NodeKind::PlanStep, |
| 1221 | title, |
| 1222 | state: if desired == NodeState::Active { |
| 1223 | NodeState::Ready |
| 1224 | } else { |
| 1225 | desired |
| 1226 | }, |
| 1227 | acceptance: Vec::new(), |
| 1228 | binding: None, |
| 1229 | evidence: None, |
| 1230 | provenance, |
| 1231 | created_at: now_ms(), |
| 1232 | updated_at: now_ms(), |
| 1233 | }, |
| 1234 | )?; |
| 1235 | ensure_contains(&mut graph, session_id, tool, &objective, &node)?; |
| 1236 | if desired == NodeState::Active { |
| 1237 | let clean_title = graph |
| 1238 | .snapshot() |
| 1239 | .node(&node) |
| 1240 | .map(|node| node.title.clone()) |
| 1241 | .ok_or_else(|| format!("node {node} not found after insert"))?; |
| 1242 | patch_existing_node(&mut graph, session_id, tool, &node, clean_title, desired)?; |
| 1243 | } |
| 1244 | (node, None) |
| 1245 | }; |
| 1246 | bindings.push(CompatTodoBinding { |
| 1247 | legacy_id: item.id, |
| 1248 | node, |
| 1249 | plan_index, |
| 1250 | }); |
| 1251 | } |
| 1252 | let mut compat = graph.snapshot().compat.clone(); |
| 1253 | compat.todos = bindings; |
| 1254 | apply_change( |
| 1255 | &mut graph, |
| 1256 | session_id, |
| 1257 | tool, |
| 1258 | WorkGraphChange::ReplaceCompatProjection { compat }, |
| 1259 | )?; |
| 1260 | Ok(graph.into_snapshot()) |
| 1261 | } |
| 1262 | |
| 1263 | fn ensure_objective( |
| 1264 | graph: &mut WorkGraph, |
| 1265 | session_id: &str, |
| 1266 | tool: &str, |
| 1267 | plan: &PlanSnapshot, |
| 1268 | ) -> Result<WorkNodeId, String> { |
| 1269 | let id = graph |
| 1270 | .snapshot() |
| 1271 | .nodes |
| 1272 | .iter() |
| 1273 | .find(|node| node.kind == NodeKind::Objective) |
| 1274 | .map(|node| node.id.clone()) |
| 1275 | .unwrap_or_else(|| WorkNodeId::derive(session_id, "objective")); |
| 1276 | let title = plan |
| 1277 | .objective |
| 1278 | .as_deref() |
| 1279 | .or(plan.title.as_deref()) |
| 1280 | .unwrap_or("Session work") |
| 1281 | .to_string(); |
| 1282 | upsert_node( |
| 1283 | graph, |
| 1284 | session_id, |
| 1285 | tool, |
| 1286 | WorkNode { |
| 1287 | id: id.clone(), |
| 1288 | kind: NodeKind::Objective, |
| 1289 | title, |
| 1290 | state: NodeState::Ready, |
| 1291 | acceptance: Vec::new(), |
| 1292 | binding: None, |
| 1293 | evidence: None, |
| 1294 | provenance: tool_provenance(graph.snapshot(), tool), |
| 1295 | created_at: now_ms(), |
| 1296 | updated_at: now_ms(), |
| 1297 | }, |
| 1298 | )?; |
| 1299 | Ok(id) |
| 1300 | } |
| 1301 | |
| 1302 | fn upsert_node( |
| 1303 | graph: &mut WorkGraph, |
| 1304 | session_id: &str, |
| 1305 | tool: &str, |
| 1306 | node: WorkNode, |
| 1307 | ) -> Result<(), String> { |
| 1308 | if let Some(existing) = graph.snapshot().node(&node.id) { |
| 1309 | if existing.kind != node.kind { |
| 1310 | return Err(format!("node {} changed kind", node.id)); |
| 1311 | } |
| 1312 | patch_existing_node(graph, session_id, tool, &node.id, node.title, node.state) |
| 1313 | } else { |
| 1314 | apply_change(graph, session_id, tool, WorkGraphChange::AddNode { node }) |
| 1315 | } |
| 1316 | } |
| 1317 | |
| 1318 | fn patch_existing_node( |
| 1319 | graph: &mut WorkGraph, |
| 1320 | session_id: &str, |
| 1321 | tool: &str, |
| 1322 | id: &WorkNodeId, |
| 1323 | title: String, |
| 1324 | state: NodeState, |
| 1325 | ) -> Result<(), String> { |
| 1326 | let current = graph |
| 1327 | .snapshot() |
| 1328 | .node(id) |
| 1329 | .ok_or_else(|| format!("node {id} not found"))?; |
| 1330 | if current.title == title && current.state == state { |
| 1331 | // Semantic no-ops must stay no-ops. In particular, terminal nodes are |
| 1332 | // immutable, so refreshing another To-do item must not try to rewrite |
| 1333 | // a settled sibling merely to stamp newer tool provenance. |
| 1334 | return Ok(()); |
| 1335 | } |
| 1336 | let provenance = tool_provenance(graph.snapshot(), tool); |
| 1337 | apply_change( |
| 1338 | graph, |
| 1339 | session_id, |
| 1340 | tool, |
| 1341 | WorkGraphChange::UpdateNode { |
| 1342 | id: id.clone(), |
| 1343 | patch: WorkNodePatch { |
| 1344 | title: Some(title), |
| 1345 | state: Some(state), |
| 1346 | provenance: Some(provenance), |
| 1347 | ..WorkNodePatch::default() |
| 1348 | }, |
| 1349 | }, |
| 1350 | ) |
| 1351 | } |
| 1352 | |
| 1353 | fn ensure_contains( |
| 1354 | graph: &mut WorkGraph, |
| 1355 | session_id: &str, |
| 1356 | tool: &str, |
| 1357 | parent: &WorkNodeId, |
| 1358 | child: &WorkNodeId, |
| 1359 | ) -> Result<(), String> { |
| 1360 | let id = WorkEdgeId::derive( |
| 1361 | session_id, |
| 1362 | &format!("contains:{}:{}", parent.as_str(), child.as_str()), |
| 1363 | ); |
| 1364 | if graph.snapshot().edge(&id).is_some() { |
| 1365 | return Ok(()); |
| 1366 | } |
| 1367 | apply_change( |
| 1368 | graph, |
| 1369 | session_id, |
| 1370 | tool, |
| 1371 | WorkGraphChange::AddEdge { |
| 1372 | edge: WorkEdge { |
| 1373 | id, |
| 1374 | kind: EdgeKind::Contains, |
| 1375 | from: parent.clone(), |
| 1376 | to: child.clone(), |
| 1377 | }, |
| 1378 | }, |
| 1379 | ) |
| 1380 | } |
| 1381 | |
| 1382 | fn deactivate_projected_todos( |
| 1383 | graph: &mut WorkGraph, |
| 1384 | session_id: &str, |
| 1385 | tool: &str, |
| 1386 | ) -> Result<(), String> { |
| 1387 | let active = graph |
| 1388 | .snapshot() |
| 1389 | .compat |
| 1390 | .todos |
| 1391 | .iter() |
| 1392 | .filter_map(|binding| { |
| 1393 | graph |
| 1394 | .snapshot() |
| 1395 | .node(&binding.node) |
| 1396 | .filter(|node| node.state == NodeState::Active) |
| 1397 | .map(|node| (node.id.clone(), node.title.clone())) |
| 1398 | }) |
| 1399 | .collect::<Vec<_>>(); |
| 1400 | for (id, title) in active { |
| 1401 | patch_existing_node(graph, session_id, tool, &id, title, NodeState::Ready)?; |
| 1402 | } |
| 1403 | Ok(()) |
| 1404 | } |
| 1405 | |
| 1406 | fn apply_change( |
| 1407 | graph: &mut WorkGraph, |
| 1408 | session_id: &str, |
| 1409 | tool: &str, |
| 1410 | change: WorkGraphChange, |
| 1411 | ) -> Result<(), String> { |
| 1412 | graph |
| 1413 | .apply( |
| 1414 | change, |
| 1415 | ChangeCtx { |
| 1416 | session_id: session_id.to_string(), |
| 1417 | now: now_ms(), |
| 1418 | idempotency_key: None, |
| 1419 | }, |
| 1420 | ) |
| 1421 | .map(|_| ()) |
| 1422 | .map_err(|err| format!("{tool}: {err}")) |
| 1423 | } |
| 1424 | |
| 1425 | fn validate_combined( |
| 1426 | graph: &WorkGraphSnapshot, |
| 1427 | plan: &PlanSnapshot, |
| 1428 | todos: &TodoListSnapshot, |
| 1429 | ) -> Result<(), String> { |
| 1430 | validate(graph).map_err(|err| err.to_string())?; |
| 1431 | if &project_plan(graph) != plan { |
| 1432 | return Err("Work Graph Plan projection is inconsistent".to_string()); |
| 1433 | } |
| 1434 | if &project_todos(graph) != todos { |
| 1435 | return Err("Work Graph To-do projection is inconsistent".to_string()); |
| 1436 | } |
| 1437 | TodoList::from_snapshot(todos)?; |
| 1438 | Ok(()) |
| 1439 | } |
| 1440 | |
| 1441 | fn tool_provenance(snapshot: &WorkGraphSnapshot, tool: &str) -> Provenance { |
| 1442 | Provenance::ToolUpdate { |
| 1443 | tool: tool.to_string(), |
| 1444 | call_id: format!("{tool}:{}", snapshot.revision.saturating_add(1)), |
| 1445 | } |
| 1446 | } |
| 1447 | |
| 1448 | fn plan_node_state(status: &StepStatus) -> NodeState { |
| 1449 | match status { |
| 1450 | StepStatus::Pending => NodeState::Ready, |
| 1451 | StepStatus::InProgress => NodeState::Active, |
| 1452 | StepStatus::Completed => NodeState::Completed, |
| 1453 | } |
| 1454 | } |
| 1455 | |
| 1456 | fn todo_node_state(status: TodoStatus) -> NodeState { |
| 1457 | match status { |
| 1458 | TodoStatus::Pending => NodeState::Ready, |
| 1459 | TodoStatus::InProgress => NodeState::Active, |
| 1460 | TodoStatus::Completed => NodeState::Completed, |
| 1461 | TodoStatus::Cancelled => NodeState::Cancelled, |
| 1462 | } |
| 1463 | } |
| 1464 | |
| 1465 | fn resolved_session_id(active: &ActiveGraph, requested: Option<&str>) -> String { |
| 1466 | requested |
| 1467 | .map(str::to_string) |
| 1468 | .or_else(|| active.session_id.clone()) |
| 1469 | .unwrap_or_else(|| "unsaved-work".to_string()) |
| 1470 | } |
| 1471 | |
| 1472 | fn now_ms() -> i64 { |
| 1473 | chrono::Utc::now().timestamp_millis() |
| 1474 | } |
| 1475 | |
| 1476 | fn lock_unpoisoned<T>(mutex: &Mutex<T>) -> MutexGuard<'_, T> { |
| 1477 | mutex |
| 1478 | .lock() |
| 1479 | .unwrap_or_else(std::sync::PoisonError::into_inner) |
| 1480 | } |
| 1481 | |
| 1482 | fn retry_lock<T>( |
| 1483 | mutex: &tokio::sync::Mutex<T>, |
| 1484 | retries: u32, |
| 1485 | ) -> Option<tokio::sync::MutexGuard<'_, T>> { |
| 1486 | for _ in 0..retries { |
| 1487 | if let Ok(guard) = mutex.try_lock() { |
| 1488 | return Some(guard); |
| 1489 | } |
| 1490 | std::thread::sleep(std::time::Duration::from_millis(1)); |
| 1491 | } |
| 1492 | None |
| 1493 | } |
| 1494 | |
| 1495 | #[cfg(test)] |
| 1496 | mod tests { |
| 1497 | use super::*; |
| 1498 | use crate::tools::plan::PlanItemArg; |
| 1499 | use crate::work_graph::{EvidenceKind, EvidenceRef, OwnerState}; |
| 1500 | |
| 1501 | #[test] |
| 1502 | fn operation_lifecycle_is_registered_idempotent_and_receipt_only() { |
| 1503 | let runtime = new_shared_work_runtime( |
| 1504 | crate::tools::todo::new_shared_todo_list(), |
| 1505 | crate::tools::plan::new_shared_plan_state(), |
| 1506 | ); |
| 1507 | let intent = OperationIntent::new( |
| 1508 | "shell:shell_test", |
| 1509 | "silent `owner` command", |
| 1510 | false, |
| 1511 | "exec_shell", |
| 1512 | "shell_test", |
| 1513 | ); |
| 1514 | let node_id = runtime |
| 1515 | .register_operation("session", intent.clone()) |
| 1516 | .expect("register before spawn"); |
| 1517 | assert_eq!( |
| 1518 | runtime.register_operation("session", intent), |
| 1519 | Ok(node_id.clone()), |
| 1520 | "repeat spawn intent must not duplicate the operation" |
| 1521 | ); |
| 1522 | let initialized = runtime |
| 1523 | .capture(Some("session")) |
| 1524 | .expect("capture") |
| 1525 | .expect("graph"); |
| 1526 | let node = initialized.graph.node(&node_id).expect("operation node"); |
| 1527 | assert_eq!(node.state, NodeState::Initializing); |
| 1528 | assert!( |
| 1529 | initialized |
| 1530 | .graph |
| 1531 | .edges |
| 1532 | .iter() |
| 1533 | .any(|edge| { edge.kind == EdgeKind::Contains && edge.to == node_id }) |
| 1534 | ); |
| 1535 | |
| 1536 | let output = EvidenceRef::new( |
| 1537 | EvidenceKind::Receipt { |
| 1538 | owner: "shell".to_string(), |
| 1539 | }, |
| 1540 | "shell:shell_test:output", |
| 1541 | Some(4_096), |
| 1542 | true, |
| 1543 | ) |
| 1544 | .expect("safe logical receipt"); |
| 1545 | assert_eq!( |
| 1546 | runtime.reconcile_operation( |
| 1547 | "session", |
| 1548 | OperationOwnerSnapshot::new("shell:shell_test", OwnerState::Running, 7, 10,) |
| 1549 | .with_output(output), |
| 1550 | ), |
| 1551 | Ok(true) |
| 1552 | ); |
| 1553 | assert_eq!( |
| 1554 | runtime.reconcile_operation( |
| 1555 | "session", |
| 1556 | OperationOwnerSnapshot::new("shell:shell_test", OwnerState::Completed, 7, 11,), |
| 1557 | ), |
| 1558 | Ok(false), |
| 1559 | "the same binding sequence is an idempotent no-op" |
| 1560 | ); |
| 1561 | assert!( |
| 1562 | runtime |
| 1563 | .reconcile_operation( |
| 1564 | "session", |
| 1565 | OperationOwnerSnapshot::new("shell:unknown", OwnerState::Running, 1, 12,), |
| 1566 | ) |
| 1567 | .expect_err("unknown owner must not materialize after spawn") |
| 1568 | .contains("not registered") |
| 1569 | ); |
| 1570 | let running = runtime |
| 1571 | .capture(Some("session")) |
| 1572 | .expect("capture running") |
| 1573 | .expect("graph"); |
| 1574 | let binding = running |
| 1575 | .graph |
| 1576 | .node(&node_id) |
| 1577 | .and_then(|node| node.binding.as_ref()) |
| 1578 | .expect("binding"); |
| 1579 | assert_eq!( |
| 1580 | running.graph.node(&node_id).map(|node| node.state), |
| 1581 | Some(NodeState::Active) |
| 1582 | ); |
| 1583 | assert_eq!( |
| 1584 | binding |
| 1585 | .last_observation |
| 1586 | .as_ref() |
| 1587 | .and_then(|obs| obs.output.as_ref()) |
| 1588 | .and_then(EvidenceRef::raw_bytes), |
| 1589 | Some(4_096) |
| 1590 | ); |
| 1591 | let summary = runtime |
| 1592 | .active_operation_summary(Some("session")) |
| 1593 | .expect("compaction re-anchor"); |
| 1594 | assert!(summary.contains("shell:shell_test"), "{summary}"); |
| 1595 | assert!(summary.contains("silent 'owner' command"), "{summary}"); |
| 1596 | assert!(!summary.contains("4,096"), "{summary}"); |
| 1597 | |
| 1598 | assert_eq!( |
| 1599 | runtime.record_reasoning_effort_change( |
| 1600 | Some("session"), |
| 1601 | ReasoningEffortTier::Low, |
| 1602 | ReasoningEffortTier::High, |
| 1603 | crate::config::ApiProvider::Moonshot, |
| 1604 | "moonshot", |
| 1605 | Some(crate::config::DEFAULT_MOONSHOT_BASE_URL), |
| 1606 | Some("kimi-k2.5"), |
| 1607 | ), |
| 1608 | Ok(Some(node_id.clone())) |
| 1609 | ); |
| 1610 | let activity = runtime |
| 1611 | .capture(Some("session")) |
| 1612 | .expect("capture effort activity") |
| 1613 | .expect("graph") |
| 1614 | .graph |
| 1615 | .activities |
| 1616 | .last() |
| 1617 | .cloned() |
| 1618 | .expect("effort activity"); |
| 1619 | let ts = match &activity { |
| 1620 | WorkActivityEvent::ReasoningEffortChanged { ts, .. } => *ts, |
| 1621 | }; |
| 1622 | assert_eq!( |
| 1623 | activity, |
| 1624 | WorkActivityEvent::ReasoningEffortChanged { |
| 1625 | requested: ReasoningEffortTier::Low, |
| 1626 | effective: ReasoningEffortTier::High, |
| 1627 | provider_kind: Some(crate::config::ApiProvider::Moonshot), |
| 1628 | provider: "moonshot".to_string(), |
| 1629 | endpoint_identity: Some(crate::config::DEFAULT_MOONSHOT_BASE_URL.to_string()), |
| 1630 | model: Some("kimi-k2.5".to_string()), |
| 1631 | ts, |
| 1632 | operation: Some(node_id.clone()), |
| 1633 | } |
| 1634 | ); |
| 1635 | |
| 1636 | runtime |
| 1637 | .record_operation_approval( |
| 1638 | "session", |
| 1639 | "shell:shell_test", |
| 1640 | "operate-verification:shell_test", |
| 1641 | "exec_shell", |
| 1642 | "approval_test", |
| 1643 | ) |
| 1644 | .expect("approval provenance"); |
| 1645 | let approved = runtime |
| 1646 | .capture(Some("session")) |
| 1647 | .expect("capture approval") |
| 1648 | .expect("graph"); |
| 1649 | assert!( |
| 1650 | approved |
| 1651 | .graph |
| 1652 | .nodes |
| 1653 | .iter() |
| 1654 | .any(|node| node.kind == NodeKind::Approval) |
| 1655 | ); |
| 1656 | assert!( |
| 1657 | approved |
| 1658 | .graph |
| 1659 | .edges |
| 1660 | .iter() |
| 1661 | .any(|edge| edge.kind == EdgeKind::RequiresApproval) |
| 1662 | ); |
| 1663 | |
| 1664 | runtime |
| 1665 | .reconcile_operation( |
| 1666 | "session", |
| 1667 | OperationOwnerSnapshot::new("shell:shell_test", OwnerState::Completed, 8, 13), |
| 1668 | ) |
| 1669 | .expect("terminal owner report"); |
| 1670 | runtime |
| 1671 | .reconcile_observation( |
| 1672 | "session", |
| 1673 | "shell:shell_test", |
| 1674 | OperationObservation::CancelUpdate { |
| 1675 | outcome: super::super::CancelOutcome::AlreadyFinished, |
| 1676 | at: 14, |
| 1677 | }, |
| 1678 | ) |
| 1679 | .expect("already-finished cancellation receipt"); |
| 1680 | assert_eq!( |
| 1681 | runtime |
| 1682 | .capture(Some("session")) |
| 1683 | .expect("capture terminal") |
| 1684 | .expect("graph") |
| 1685 | .graph |
| 1686 | .node(&node_id) |
| 1687 | .map(|node| node.state), |
| 1688 | Some(NodeState::Completed), |
| 1689 | "already-finished cancellation must not rewrite owner state" |
| 1690 | ); |
| 1691 | } |
| 1692 | |
| 1693 | #[test] |
| 1694 | fn restore_stales_only_live_ephemeral_operations() { |
| 1695 | let runtime = new_shared_work_runtime( |
| 1696 | crate::tools::todo::new_shared_todo_list(), |
| 1697 | crate::tools::plan::new_shared_plan_state(), |
| 1698 | ); |
| 1699 | for id in ["shell:live", "shell:done"] { |
| 1700 | runtime |
| 1701 | .register_operation( |
| 1702 | "session", |
| 1703 | OperationIntent::new(id, id, false, "exec_shell", id), |
| 1704 | ) |
| 1705 | .expect("register shell"); |
| 1706 | } |
| 1707 | runtime |
| 1708 | .reconcile_operation( |
| 1709 | "session", |
| 1710 | OperationOwnerSnapshot::new("shell:live", OwnerState::Running, 1, 1), |
| 1711 | ) |
| 1712 | .expect("live shell"); |
| 1713 | runtime |
| 1714 | .reconcile_operation( |
| 1715 | "session", |
| 1716 | OperationOwnerSnapshot::new("shell:done", OwnerState::Completed, 1, 1), |
| 1717 | ) |
| 1718 | .expect("completed shell"); |
| 1719 | let saved = runtime |
| 1720 | .capture(Some("session")) |
| 1721 | .expect("capture") |
| 1722 | .expect("saved graph"); |
| 1723 | |
| 1724 | let restored = new_shared_work_runtime( |
| 1725 | crate::tools::todo::new_shared_todo_list(), |
| 1726 | crate::tools::plan::new_shared_plan_state(), |
| 1727 | ); |
| 1728 | restored |
| 1729 | .restore("session", Some(&saved.graph), &saved.todos, &saved.plan) |
| 1730 | .expect("restore graph"); |
| 1731 | let graph = restored |
| 1732 | .capture(Some("session")) |
| 1733 | .expect("capture restored") |
| 1734 | .expect("restored graph") |
| 1735 | .graph; |
| 1736 | let state_for = |external: &str| { |
| 1737 | graph |
| 1738 | .nodes |
| 1739 | .iter() |
| 1740 | .find(|node| { |
| 1741 | node.binding |
| 1742 | .as_ref() |
| 1743 | .is_some_and(|binding| binding.external == external) |
| 1744 | }) |
| 1745 | .map(|node| node.state) |
| 1746 | }; |
| 1747 | assert_eq!(state_for("shell:live"), Some(NodeState::Stale)); |
| 1748 | assert_eq!(state_for("shell:done"), Some(NodeState::Completed)); |
| 1749 | assert!(restored.has_pending_publish()); |
| 1750 | } |
| 1751 | |
| 1752 | #[test] |
| 1753 | fn durable_owner_same_sequence_recovers_stale_once_and_rejects_regression() { |
| 1754 | let runtime = new_shared_work_runtime( |
| 1755 | crate::tools::todo::new_shared_todo_list(), |
| 1756 | crate::tools::plan::new_shared_plan_state(), |
| 1757 | ); |
| 1758 | runtime |
| 1759 | .register_operation( |
| 1760 | "session", |
| 1761 | OperationIntent::new( |
| 1762 | "task:task_restore", |
| 1763 | "restored task", |
| 1764 | true, |
| 1765 | "task_create", |
| 1766 | "task_restore", |
| 1767 | ), |
| 1768 | ) |
| 1769 | .expect("register durable owner"); |
| 1770 | let running = OperationOwnerSnapshot::new("task:task_restore", OwnerState::Running, 7, 10); |
| 1771 | assert_eq!( |
| 1772 | runtime.reconcile_operation("session", running.clone()), |
| 1773 | Ok(true) |
| 1774 | ); |
| 1775 | assert_eq!( |
| 1776 | runtime.reconcile_observation( |
| 1777 | "session", |
| 1778 | "task:task_restore", |
| 1779 | OperationObservation::OwnerMissing { checked_at: 11 }, |
| 1780 | ), |
| 1781 | Ok(true) |
| 1782 | ); |
| 1783 | assert_eq!( |
| 1784 | runtime |
| 1785 | .capture(Some("session")) |
| 1786 | .expect("capture stale") |
| 1787 | .expect("graph") |
| 1788 | .graph |
| 1789 | .nodes |
| 1790 | .iter() |
| 1791 | .find(|node| { |
| 1792 | node.binding |
| 1793 | .as_ref() |
| 1794 | .is_some_and(|binding| binding.external == "task:task_restore") |
| 1795 | }) |
| 1796 | .map(|node| node.state), |
| 1797 | Some(NodeState::Stale) |
| 1798 | ); |
| 1799 | |
| 1800 | let replay = OperationOwnerSnapshot::new("task:task_restore", OwnerState::Running, 7, 12); |
| 1801 | assert_eq!( |
| 1802 | runtime.reconcile_operation("session", replay.clone()), |
| 1803 | Ok(true) |
| 1804 | ); |
| 1805 | assert_eq!( |
| 1806 | runtime.reconcile_operation("session", replay), |
| 1807 | Ok(false), |
| 1808 | "same-sequence recovery must happen only while the node is stale" |
| 1809 | ); |
| 1810 | assert_eq!( |
| 1811 | runtime |
| 1812 | .capture(Some("session")) |
| 1813 | .expect("capture recovered") |
| 1814 | .expect("graph") |
| 1815 | .graph |
| 1816 | .nodes |
| 1817 | .iter() |
| 1818 | .find(|node| { |
| 1819 | node.binding |
| 1820 | .as_ref() |
| 1821 | .is_some_and(|binding| binding.external == "task:task_restore") |
| 1822 | }) |
| 1823 | .map(|node| node.state), |
| 1824 | Some(NodeState::Active) |
| 1825 | ); |
| 1826 | assert_eq!( |
| 1827 | runtime.reconcile_operation( |
| 1828 | "session", |
| 1829 | OperationOwnerSnapshot::new("task:task_restore", OwnerState::Waiting, 7, 13,), |
| 1830 | ), |
| 1831 | Ok(false), |
| 1832 | "ordinary same-key duplicates retain the reducer no-op contract" |
| 1833 | ); |
| 1834 | runtime |
| 1835 | .reconcile_observation( |
| 1836 | "session", |
| 1837 | "task:task_restore", |
| 1838 | OperationObservation::OwnerMissing { checked_at: 14 }, |
| 1839 | ) |
| 1840 | .expect("mark owner missing again"); |
| 1841 | assert!( |
| 1842 | runtime |
| 1843 | .reconcile_operation( |
| 1844 | "session", |
| 1845 | OperationOwnerSnapshot::new("task:task_restore", OwnerState::Waiting, 7, 15,), |
| 1846 | ) |
| 1847 | .expect_err("inconsistent replay cannot revive a stale node") |
| 1848 | .contains("changed observation") |
| 1849 | ); |
| 1850 | assert!( |
| 1851 | runtime |
| 1852 | .reconcile_operation( |
| 1853 | "session", |
| 1854 | OperationOwnerSnapshot::new("task:task_restore", OwnerState::Running, 6, 16,), |
| 1855 | ) |
| 1856 | .expect_err("owner sequence cannot regress") |
| 1857 | .contains("sequence regressed") |
| 1858 | ); |
| 1859 | } |
| 1860 | |
| 1861 | #[test] |
| 1862 | fn legacy_restore_stays_pending_until_first_graph_bearing_write() { |
| 1863 | let todos = crate::tools::todo::new_shared_todo_list(); |
| 1864 | let plan = crate::tools::plan::new_shared_plan_state(); |
| 1865 | let runtime = new_shared_work_runtime(todos.clone(), plan.clone()); |
| 1866 | let legacy_plan = PlanSnapshot { |
| 1867 | items: vec![PlanItemArg { |
| 1868 | step: "Migrate once".to_string(), |
| 1869 | status: StepStatus::InProgress, |
| 1870 | }], |
| 1871 | ..PlanSnapshot::default() |
| 1872 | }; |
| 1873 | |
| 1874 | runtime |
| 1875 | .restore( |
| 1876 | "legacy-session", |
| 1877 | None, |
| 1878 | &TodoListSnapshot::default(), |
| 1879 | &legacy_plan, |
| 1880 | ) |
| 1881 | .expect("restore legacy state"); |
| 1882 | assert!(runtime.has_pending_publish()); |
| 1883 | let captured = runtime |
| 1884 | .capture(Some("legacy-session")) |
| 1885 | .expect("capture imported graph") |
| 1886 | .expect("state"); |
| 1887 | assert!(captured.graph.import_digest.is_some()); |
| 1888 | assert_eq!(plan.blocking_lock().snapshot(), legacy_plan); |
| 1889 | assert_eq!(runtime.publish_pending_sync(), Ok(true)); |
| 1890 | assert!(!runtime.has_pending_publish()); |
| 1891 | assert!(todos.blocking_lock().snapshot().is_empty()); |
| 1892 | } |
| 1893 | } |
| 1894 |