fix: validated ticket sweep — tier 1 + tier 2 bugs #538

Merged

dries merged 11 commits from fix/validated-ticket-sweep into main 2026-08-20 23:12:38 +02:00

Conversation 2 Commits 11 Files changed 24 +897 -100

dries commented 2026-08-20 18:38:05 +02:00

Owner

Copy link

One commit per validated ticket, each with a failing-first (red/green) test where testable. All fixes were verified against current main before implementation (see the validation comments on each ticket).

Tier 1

• #529 — sends fired during the navigation stale window targeted the previous session. handleSend/handleRetrySend now fail closed when the session prop lags the route id. (useSessionActions.staleRoute.test.ts, red on main)
• #490 — the global SSE hub silently dropped ocman.session.idle / ocman.session.changed on a full subscriber buffer. Both are identity-keyed, so they now coalesce last-write-wins per (event, session) like queue.updated; remaining drops log at Debug. (TestBroadcastHubCoalescesSessionEventsOnFullBuffer, red on main)
• #462 — autoapprove tee buffer was unbounded and re-parsed from offset 0 per write. Now incremental parsing (cost per Write ∝ new bytes), 4 MB pending cap, resync at the next event terminator on overflow. (tee_buffer_test.go, red on main)
• #532 — MCP splits and scheduled prompts ensured against the raw worktree dir, launching a second opencode instance per worktree. Both paths now fold to the project root like the queue path. (host_ensure_fold_test.go, red on main)

Tier 2

• #456 — the ensure/restart singleflight body ran under the winning caller's ctx; a cancelled winner failed every coalesced waiter. The body now runs detached (2 min bound); callers wait on their own ctx. (TestEnsureProjectOpencode_WinnerCancelDoesNotFailWaiters, red on main)
• #458 — ChildResultBroker.wait discarded an already-received result on a cancel race. A result in hand now always wins. (TestWaitReturnsBufferedResultOverCancellation, red on main)
• #459 — composer retried a downed backend every 1s forever with the composer locked. Now 5 retries with exponential backoff, then unlock with the draft intact. (Composer.queue test, red on main)
• #460 — the session cache mirror wrote a full Map clone on every streaming token. Now debounced 500 ms with an unmount/session-switch flush. (useSession write-count test, red on main)
• #533 — remote create-session ran EnsureProjectOpencode before platform validation; an unknown platform launched a managed instance on the remote. Registry check now precedes the ensure. Local half N/A on main (PR #523 unmerged). (TestCreateSessionValidatesPlatformBeforeEnsure, red on main)
• #530 — architecture diagram now shows the production port :8228, with the dev split noted below it.

Verification

• make lint clean (guards included)
• make test: backend + frontend fully green (1855 frontend tests)
• go test -race on the touched concurrency packages (hostsvc/local)

Closes #529, closes #490, closes #462, closes #532, closes #456, closes #458, closes #459, closes #460, closes #533, closes #530.

One commit per validated ticket, each with a failing-first (red/green) test where testable. All fixes were verified against current main before implementation (see the validation comments on each ticket). ## Tier 1 - **#529** — sends fired during the navigation stale window targeted the previous session. handleSend/handleRetrySend now fail closed when the session prop lags the route id. (useSessionActions.staleRoute.test.ts, red on main) - **#490** — the global SSE hub silently dropped ocman.session.idle / ocman.session.changed on a full subscriber buffer. Both are identity-keyed, so they now coalesce last-write-wins per (event, session) like queue.updated; remaining drops log at Debug. (TestBroadcastHubCoalescesSessionEventsOnFullBuffer, red on main) - **#462** — autoapprove tee buffer was unbounded and re-parsed from offset 0 per write. Now incremental parsing (cost per Write ∝ new bytes), 4 MB pending cap, resync at the next event terminator on overflow. (tee_buffer_test.go, red on main) - **#532** — MCP splits and scheduled prompts ensured against the raw worktree dir, launching a second opencode instance per worktree. Both paths now fold to the project root like the queue path. (host_ensure_fold_test.go, red on main) ## Tier 2 - **#456** — the ensure/restart singleflight body ran under the winning caller's ctx; a cancelled winner failed every coalesced waiter. The body now runs detached (2 min bound); callers wait on their own ctx. (TestEnsureProjectOpencode_WinnerCancelDoesNotFailWaiters, red on main) - **#458** — ChildResultBroker.wait discarded an already-received result on a cancel race. A result in hand now always wins. (TestWaitReturnsBufferedResultOverCancellation, red on main) - **#459** — composer retried a downed backend every 1s forever with the composer locked. Now 5 retries with exponential backoff, then unlock with the draft intact. (Composer.queue test, red on main) - **#460** — the session cache mirror wrote a full Map clone on every streaming token. Now debounced 500 ms with an unmount/session-switch flush. (useSession write-count test, red on main) - **#533** — remote create-session ran EnsureProjectOpencode before platform validation; an unknown platform launched a managed instance on the remote. Registry check now precedes the ensure. Local half N/A on main (PR #523 unmerged). (TestCreateSessionValidatesPlatformBeforeEnsure, red on main) - **#530** — architecture diagram now shows the production port :8228, with the dev split noted below it. ## Verification - make lint clean (guards included) - make test: backend + frontend fully green (1855 frontend tests) - go test -race on the touched concurrency packages (hostsvc/local) Closes #529, closes #490, closes #462, closes #532, closes #456, closes #458, closes #459, closes #460, closes #533, closes #530.

dries force-pushed fix/validated-ticket-sweep from b2c7fac942 to b3aa0f34aa 2026-08-20 20:57:54 +02:00 Compare

dries commented 2026-08-20 20:58:21 +02:00

Author

Owner

Copy link

Rebased onto main after #523 landed (local pre-ensure + create-path worktree folding). Two notes: (1) the #533 platform-validation guard now covers BOTH paths — local and remote ensure — with an allowance for the empty-platform auto-pick; (2) the description's "local half N/A" note for #533 is stale post-rebase. #531 stays closed: main's local soft-fail Debug log is now justified by an explanatory comment in the handler, which was the ticket's alternative fix. Full suite re-run green after rebase (go test ./..., 1866 frontend tests, tsc, eslint).

Rebased onto main after #523 landed (local pre-ensure + create-path worktree folding). Two notes: (1) the #533 platform-validation guard now covers BOTH paths — local and remote ensure — with an allowance for the empty-platform auto-pick; (2) the description's "local half N/A" note for #533 is stale post-rebase. #531 stays closed: main's local soft-fail Debug log is now justified by an explanatory comment in the handler, which was the ticket's alternative fix. Full suite re-run green after rebase (go test ./..., 1866 frontend tests, tsc, eslint).

dries referenced this pull request 2026-08-20 22:57:15 +02:00

fix: data race on the bodyReadTimeout test global (TestStreamingSurvivesBodyDeadline) #539

dries added 1 commit 2026-08-20 22:57:54 +02:00

fix: validate the create-session platform via the session service ... 897aa89028

Fixes the CI backend failure introduced by the #533 commit: the handler
read s.registry directly, which is nil in the owner-routing test doubles
(they stub s.sessions over their own registry), panicking in
TestHandlersAcceptLocalOwner. Worse, s.registry and the registry behind
s.sessions are not necessarily the same object, so the pre-ensure check
could disagree with the Create it was guarding.

Move the rule to sessionsvc.KnownPlatform, which answers from the same
registry Create validates against and is nil-safe. Covered by
TestKnownPlatform.

forgejo-actions commented 2026-08-20 23:08:22 +02:00

Copy link

Coverage ratchet: ✅ pass

Suite	Baseline	This PR	Δ
frontend	71.33%	71.39%	+0.06	✅
go	81.40%	81.30%	-0.10	✅

Tolerance: -0.1%. Baseline stored on gh-pages.

<!-- coverage-ratchet --> ### Coverage ratchet: ✅ pass | Suite | Baseline | This PR | Δ | | |---|---|---|---|---| | frontend | 71.33% | 71.39% | +0.06 | ✅ | | go | 81.40% | 81.30% | -0.10 | ✅ | _Tolerance: -0.1%. Baseline stored on `gh-pages`._

dries merged commit 897aa89028 into main 2026-08-20 23:12:38 +02:00

dries deleted branch fix/validated-ticket-sweep 2026-08-20 23:12:39 +02:00

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels   

backend

Go backend code

  

bug

Something is broken or incorrect

  

chore

Maintenance, tooling, or cleanup

  

duplication

Code duplication to be addressed

  

effort:complex

Significant effort or risk

  

effort:medium

Moderate effort required

  

effort:trivial

Quick fix, low complexity

  

enhancement

New feature or capability

  

follow-up

Follow-up from a PR review or shipped change

  

frontend

TypeScript / React frontend code

  

fullstack

Touches both backend and frontend

  

priority:high

Should be addressed soon

  

ready-for-agent

Agent-grabbable ticket, fully specced

  

refactor

Code restructuring without behaviour change

  

security

Security related

No labels

backend

bug

chore

duplication

effort:complex

effort:medium

effort:trivial

enhancement

follow-up

frontend

fullstack

priority:high

ready-for-agent

refactor

security

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

2 participants

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

dries/ocman!538

No description provided.