Reconcile chat pacing and queued follow-ups - #254
Closed
0xSero wants to merge 10 commits into
Closed
Conversation
… wire) Cherry-pick of 025f7f8 from codex/local-studio-fidelity-performance. Resolved conflict in json_line_wire.rs: kept the current async-fn next_message signature and ported the legacy lifecycle normalization (parse to serde_json::Value, normalize, from_value) into it. The candidate's impl-Future form was a pre-async-fn rewrite no longer needed on this base. Wave 0 disposition: accept. Captured in work/chat-performance/evidence/task-00-disposition.md.
Cherry-pick of 96e77de from codex/local-studio-fidelity-performance. Fixes a latent compile defect in the candidate's coalesce_streaming_window: merge_app_update returns Err(Box<AppStoreUpdateRecord>), so the buffered push_front must unbox with *next, matching coalesce_ready_updates. Wave 0 disposition: accept (with the one-line unbox fix). Captured in work/chat-performance/evidence/task-00-disposition.md.
Cherry-pick of f1946f4 from codex/local-studio-fidelity-performance. Applied cleanly onto the reconciled base (025f7f8 + 96e77de). Adds turn-start reservation, queued follow-up drain in store_listener, reducer/snapshot queue state, and four queue concurrency/restore-order tests under mobile_client::tests. Wave 0 disposition: accept. Captured in work/chat-performance/evidence/task-00-disposition.md.
Task 00 evidence: accept 025f7f8 (rewrite, async-fn wire conflict resolved), accept 96e77de (rewrite, one-line unbox compile fix), accept f1946f4 (clean cherry-pick), reject a4cbf2d (release-only metadata, out of performance scope). Documents mobile-parity analysis, build/test economy, and acceptance mapping.
…covery) Review 04-05 follow-up to 0c95716 (96e77de reconciled). - should_flush_streaming_now rearms next_flush_at = now + WINDOW when a same-identity delta arrives at/after the passed deadline, so an immediately-following delta coalesces instead of flushing again. - Add pending_full_resync flag on AppStoreSubscriptionState. On lag while an accumulated streaming update exists, coalesce_streaming_window returns the accumulated exact text and defers FullResync; receive_next_update consumes the flag, clears pacing, and returns FullResync on the next receive. Close still surfaces accumulated text; the following receive reports closed normally. - Remove the scheduler-sensitive elapsed < WINDOW wall-clock assertion from the isolated-first-delta test; the deterministic helper tests are the regression proof. - Extend pacing_flushes_first_after_idle_immediately_and_arms_deadline: after a passed-deadline (old+5ms) flush, assert rearmed == flush_time + WINDOW, a 2ms-later same-identity delta returns false, and rearmed != old_deadline + WINDOW. - Add app_store_subscription_surfaces_accumulated_text_then_full_resync_after_lag: deterministic, no sleeps/reducer — capacity-2 broadcast channel forced laggy by three sends, current delta passed directly into coalesce_streaming_window, asserts exact text first then FullResync next.
Review 06-08 follow-up to f07ce92 (f1946f4 reconciled). Reducer invariant: an old or duplicate terminal event is a no-op for a newer in-flight turn. - ThreadStatusChanged(Idle|SystemError): upsert_or_merge_thread's pre-merge status guard now preserves Idle for any active_turn_id (pre-existing behavior) and preserves SystemError only for a local-queued-turn reservation. The event closure additionally preserves active_turn_id, plan progress, and agent status for a local reservation. SystemError semantics for a normal authoritative active turn are unchanged. - TurnCompleted: terminal cleanup runs only when the completion is authoritative (matches active_turn_id, or no active turn for tolerant legacy completion). A nonmatching completion (including one for the old turn while a local reservation is in flight) is a true no-op: closure returns false and the caller checks .unwrap_or(false), so no active state mutates and no metadata-changed signal fires. Controlled race test (multi-thread tokio): stale_terminal_event_cannot_clear_newer_local_queued_reservation queues two drafts, parks the first turn/start after the reservation installs, applies an old TurnCompleted plus duplicate Idle and SystemError while parked, attempts a second drain (sends nothing), asserts the reservation stays authoritative and status stays Active with one draft queued, then releases the first request and asserts it resolves to the returned turn id with no duplicate send.
Review 09 follow-up to cc8efcd (025f7f8 reconciled). Wave rules require captured evidence for compatibility normalization at a wire boundary. The only committed tests for the legacy lifecycle normalizer (missing startedAtMs/completedAtMs, empty-string commandExecution.cwd) were synthetic inline JSON authored alongside the normalizer; they prove the normalizer does what it says, not that real Pi/Local Studio bridges emit the legacy shape. Search of the repository and Task 00 mission artifacts found no real scrubbed Pi/Local Studio item/started or item/completed JSON-RPC frame. .pi-missions/task-00/*.jsonl matches are agent tool-call records from running git show on the candidate, not captured server wire frames. Revert json_line_wire.rs to the frozen-base strict decode path (serde_json::from_str::<JSONRPCMessage> with no Value parse/normalize/ from_value indirection). git diff 314271a -- json_line_wire.rs is empty. The independently valid iOS required-activity preference migration from cc8efcd is retained. Classification: 025f7f8 is partial/superseded pending capture; the concrete capture task is recorded in work/chat-performance/evidence/task-00-disposition.md.
Reflect the final commits and remove overclaims: - pacing: first delta immediate, sustained cadence-batched, deadline rearm/lag/close behavior; app_store::tests 16 passed. - queued follow-ups: controlled parked-request race + failure restoration; store_listener::tests 6 passed; explain restore/reapply rebuild cost and the patch-once/restore-once wave policy. - lifecycle candidate: partial/superseded pending capture; record the search performed (no real scrubbed frame found) and the concrete capture task; remove the stale claim that synthetic tests alone prove production acceptance. - current commit hashes, commands actually run, surfaces not proven, remaining risks.
Review 10 follow-up: documentation-only. - Remove stale claimed origin/main SHA e42ce...; moving-tip reconciliation is coordinator-owned, represented by branch history not a prose SHA. - Fix 'What the candidate did,two parts):' to grammatical prose. - Replace final-lifecycle-commit placeholders with actual commits 15e094a (wire reversal) and 400f130 (evidence update). - Split the clean/pushed/reproducible acceptance criterion honestly: clean/reproducible/no-release-metadata/submodule-clean are proven here; push and latest-origin/main reconciliation remain coordinator gates. - Label version-field facts as observed during Task 00 review, not a durable invariant.
This was referenced Aug 14, 2026
Owner
Author
|
Closing as superseded by #319 (merged as |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Purpose
Complete Wave 0 Mission 00 by reconciling the existing Local Studio fidelity/performance candidate stack onto the supervised chat-performance plan.
Key changes
Delegated mission
019ff2c5-ca5b-7659-9c70-c62941da9a94homelab/glm-5.2ACCEPTEDafter ten review/repair turns using the same persistent sessionVerification
cargo test -p codex-mobile-client: 778 passed, 7 ignored; integration and doc tests passedcargo test -p codex-slingshot: 14 passed; doc tests passedapp_store::tests: 16 passed after pacing repairsmobile_client::store_listener::tests: 6 passed after queue repairsgit diff --checkpassedFull crate suites were intentionally not repeated after focused repairs. Complete command output is retained in the ignored mission log directory; the evidence document records the exact surfaces and remaining risks.
Stacked on #240 so the implementation diff contains only Mission 00.