lsm / HyperNeo / 32429798099

80%

push

github

refactor(daemon): extract pure message-ownership gates (#2622)

* refactor(daemon): extract pure message-ownership gates

Superpipe pilot 4 (ADR 0004 functional-core), additive only. Extract the
message-ownership and delivery-admission decisions behind the duplicate-
delivery bug into pure functions with full decision-table tests:

- resolveMessageOwnership: single typed answer for who owns a message
  (job_queue wins over memory_queue; unowned means a flush may deliver)
- planFlushDelivery: flush planning as noop/batch/each with skip reasons,
  mirroring deliverFlushUnderV2/deliverEachUnderV2 post-#1101 behavior
- resolveDeliveryRole: role precedence from deliverMessage
  (reuse > explicit > turn > steer-on-unique)
- decideDeferAdmission: injectMessageIntoSession defer conjuncts

No production call sites changed; wiring happens in the apply PR.

* refactor(daemon): deliver unflattenable user messages in each plans

Review fix (P2): deliverFlushUnderV2 only uses flattenDeliveryText null
to disable batching — deliverEachUnderV2 still durably enqueues every
pending user message, including multimodal ones (pinned by the
mixed-flush test in query-mode-handler.test.ts). Dropping them from
planFlushDelivery's deliver list would strand image-bearing messages
as enqueued once wired. User messages now always enter the per-message
deliver list; a null flattenedText only forces the each action. The
not_flattenable skip reason is removed with them.

* refactor(daemon): exclude memory-owned messages from flush batchability

Review fix (P2): sendEnqueuedMessagesOnTurnEnd filters
hasPendingOrInFlight messages before batchability is computed, so a
memory-owned slash or multimodal message cannot keep the remaining
free messages from batching. planFlushDelivery now excludes
memory-owned entries from the allBatchable scan the same way, while
job-queue-owned entries keep counting as in deliverFlushUnderV2.

* refactor(daemon): make job-queue-owned flush members force per... (continued)

44388 of 59800 branches covered (74.23%)

Branch coverage included in aggregate %.

34 of 34 new or added lines in 1 file covered. (100.0%)

60305 of 72071 relevant lines covered (83.67%)

3973.13 hits per line