Skip to content

Upstream sync 2026-10-22: 274 upstream commits #102

Closed

Description

@tannerpolley

Daily upstream sync: merge upstream/main at 4eae8ae (274 commits) into main at c942f0d.

Conflicts

A trial git merge-tree gives 136 conflicted files and 225 conflict blocks.

Measurement details are on #24.

Done when

  • upstream is merged, with upstream winning by default and fork code kept additive and modular;
  • the lost-fork-code audit (scripts/fork-sync/lost-fork-code.py) has been run, and every flagged hunk judged;
  • per-package tsc-rs and the touched and added tests pass;
  • the PR has been reviewed, main fast-forwarded, and 2.12.45 built and staged.

Out of scope

Mobile; providers other than Codex and Claude.

Activity

  1. tannerpolley commented on Oct 22, 2026

    @tannerpolley
    OwnerAuthor

    T3 Code status: Unavailable

    Issue: #102 Upstream sync 2026-10-22: 274 upstream commits

    Thread: thread:delegated-task:command%3Amcp%3Ad46f5624-981f-4b85-8e16-9d7bae6511f6%3Adelegate-task%3Async-2026-10-22-fix-1

    Model: gpt-6.1-sol

    Branch: issue-102-upstream-sync-2026-10-22-274-upstream-3

    Updated: 2026-11-13T04:58:54.527Z

  2. tannerpolley commented on Oct 22, 2026

    @tannerpolley
    OwnerAuthor

    T3 Code closeout: failed

    Issue: #102 Upstream sync 2026-10-22: 274 upstream commits
    Thread: thread:delegated-task:command%3Amcp%3Aa0786a13-d78b-45b6-a885-2b08c1be52ef%3Adelegate-task%3Async-2026-10-22-merge
    Model: gpt-6.1-sol
    Branch: issue-102-upstream-sync-2026-10-22-274-upstream
    Completed: 2026-10-22T18:50:50.461Z

    Codex request 'thread/start' (107) timed out after 60000ms; closing the transport.

  3. tannerpolley commented on Oct 22, 2026

    @tannerpolley
    OwnerAuthor

    T3 Code closeout: done

    Issue: #102 Upstream sync 2026-10-22: 274 upstream commits
    Thread: thread:delegated-task:command%3Amcp%3Aa0786a13-d78b-45b6-a885-2b08c1be52ef%3Adelegate-task%3Async-2026-10-22-merge-2
    Model: gpt-6.1-sol
    Branch: issue-102-upstream-sync-2026-10-22-274-upstream-2
    Completed: 2026-10-22T19:38:04.133Z

    Committed effd415abe5593766ef98d5b309c9dc57103c6a0 on issue-102-upstream-sync-2026-10-22-274-upstream.

    The commit has exactly the requested parents, c942f0d6ed and 4eae8ae011, includes Refs #102 and the GPT-6.1 Sol co-author trailer, and leaves the worktree clean. Nothing was pushed; no PR was opened and main was not modified.

    Resolved 136 conflicted files: 97 tests and 39 other files, grouped by area:

    • Server orchestration: retained upstream’s module splits while preserving fork request routing, issue effects, projections, provider behavior, scheduled work, and workspace handling.
    • MCP and contracts: reconciled toolkit handlers, capabilities, approvals, settings, notification references, and shared client operations.
    • Web: ported fork behavior into extracted chat, composer, markdown, sidebar, settings, usage, notification, and pull-request components.
    • Desktop: preserved backend lifecycle handling, fork update behavior, zoom controls, and packaging assertions.
    • Tests and dependencies: kept upstream’s test pruning, moved fork assertions into fork-owned suites, and accommodated the compiler-package replacement.

    The exact committed-HEAD audit was rerun:

    scripts/fork-sync/lost-fork-code.py c942f0d6ed 4eae8ae011 HEAD

    Its output matches the audited candidate byte for byte. The final 484 findings are 420 restyled, 11 replaced by upstream, and 53 restored. The original worker’s 521 findings are also fully judged: 447 restyled, 9 replaced, and 65 restored, plus the formerly orphaned Codex harness, which is now imported. These are finding counts, so several findings can describe one restored behavior. There are zero deleted fork-owned files and zero unimported fork modules.

    The complete audit includes every judgment, source references, and the assertion-move ledger covering 88 upstream test sources and 295 original named cases. The raw final audit is retained separately.

    The restored items are:

    • Context-sensitive pull-request Back labels.
    • Pairing/auth test-directory cleanup.
    • Selected-thread and empty shell-snapshot filtering in SQL and memory.
    • Secondary-environment thread-menu permission assertions.
    • Backend bootstrap and telemetry file-descriptor assertions.
    • Bounded backend teardown after the quit grace period.
    • Main-window zoom and preview zoom reapplication.
    • Isolated-state SQLite backup safeguards and shared-home refusal.
    • Terminal acknowledgement retries with fresh command IDs.
    • Claude monitor metadata after many newer monitors finish.
    • Claude sibling-thread roster cleanup when processes change.
    • Codex fallback to a nonblank grant root.
    • Codex session MCP configuration preserving launch-argument servers.
    • Codex subagent results surviving an empty final response and resume.
    • Stored provider usage surviving terminal updates.
    • Latest failed-root usage-limit projection into SQL and memory shells.
    • Cleanup of interrupted provider opens whose scope closure hangs.
    • Persisted session release when scope closure hangs.
    • Alias-based provider startup using the physical workspace path.
    • Inline-shell block records injected into provider prompts.
    • Issue references rendered without review-comment labels.
    • Removed token-usage helper assertions consolidated into the real projection test.
    • Server-generated worktree naming.
    • Durable attachment bytes after a lost launch result.
    • Title-regeneration marker arming and clearing.
    • Regenerated titles derived from the conversation digest.
    • Existing titles retained when generation reproduces them.
    • Regeneration markers cleared after generation failure.
    • Linux listener discovery through /proc when lsof is missing.
    • Idle listener scans stopping and resuming with candidates.
    • Claude plugin enablement and native registry ordering.
    • Pull-request review against its own base, with checkout cleanup.
    • Fork elapsed-duration formatting.
    • Restoring a setting when it is the only changed setting.
    • Asset operations targeting the correct environment.
    • Instructions distinguishing delegated children from top-level threads.
    • Usage-reset boundary formatting.
    • Platform-specific desktop packaging targets and exclusions.

    The fork fragments dropped as covered or made unnecessary by upstream are documented individually in the audit:

    • Four generic scheduler-test fixture adaptations: upstream’s effect-worker daemon remains the owner; fork lane-isolation tests remain separate.
    • Disabled issue-surface props in two pruned tab fixtures: the current RightPanelTabs implementation replaces those fixtures.
    • An undefined repository-identity fixture prop: retained the upstream ThreadDetailsPrRows split.
    • Environment/new-thread mocks in the pruned PR panel test: retained upstream’s PullRequestDetailPanel, with the fork Back label restored separately.
    • A back-button mock in the pruned refresh test: retained the current UsagePage.
    • Fork missing-question wording: followed upstream’s thread handler error semantics.
    • The narrower live-only cleanup predicate: retained upstream’s broader pending-native-question cleanup.
    • The duplicate native-question cancellation test: upstream’s ingestor tests already cover cleanup and the answer-commit race.

    Verification accounts for all 1,166 selected files, using c942f0d6ed for the test-file selection:

    • 13,818 tests passed; zero remaining assertion failures.
    • 69 additional script tests passed: 41 Node tests and 28 desktop script/extension tests.
    • All 31 non-mobile type-check scopes passed, including desktop.
    • Scoped lint passed without errors.

    The long test command ended with SIGTERM before producing its final JSON. Its completed per-file output was preserved, and the remaining 142 files passed in a clean continuation run. Newer focused runs verified the final fixes. The verification report explains the reconciliation; per-file results, long-run log, and continuation log provide the evidence.

    One suite remains unverified: ServerBrowserPage.test.ts cannot launch because Playwright’s Chromium headless-shell revision 1223 is missing. That blocked 15 cases; five other selected tests were skipped normally. The cause and absent executable are recorded in browser-blocker.log.

    For your separate integration merge, the comparison against origin/main at b22d5a8400 found these overlapping commits:

    • a37ba7f84a — issue inbox polling: docs/user/source-control.md.
    • a07c881b66 — sidebar model cells and effort meters: ThreadRelationshipsControl.tsx, threadExecution.ts, and its test.
    • 0187b27be9 — legacy import caching: LegacyV1ThreadImporter.test.ts.
    • be097335d8 — merge commit carrying overlapping main changes.
    • b22d5a8400 — merge commit carrying overlapping main changes.

    The integration-overlap report lists full SHAs and every overlapping path. Those commits were not ported into this sync.

    Explicit browser-test files and mobile were excluded as requested. Browser verification and the separately reviewed newer-main integration remain open gates.

  4. tannerpolley commented on Oct 22, 2026

    @tannerpolley
    OwnerAuthor

    T3 Code closeout: done

    Issue: #102 Upstream sync 2026-10-22: 274 upstream commits
    Thread: thread:delegated-task:command%3Amcp%3Ad46f5624-981f-4b85-8e16-9d7bae6511f6%3Adelegate-task%3Async-2026-10-22-fix-1
    Model: gpt-6.1-sol
    Branch: issue-102-upstream-sync-2026-10-22-274-upstream-3
    Completed: 2026-10-22T20:04:54.636Z

    Pushed f24ffaccaade31bd5086ab6e1ced156143ac8845 to issue-102-upstream-sync-2026-10-22-274-upstream with plain git push. PR #113 is linked, and the worktree is clean.

    The failures came from:

    • Codex and Claude Stop assertions: stale expectations exposed by the sync. Upstream commit 849b3f8948 intentionally sends { threadId: parent, subagentId }; the shared client command handles the parent task and routes Stop to the child. Both cases pass at c942f0d6ed and fail at effd415abe and 2de1eb5463.
    • Supervisor heading: a pre-existing stale expectation. Fork commit 849097092a already counted pending and waiting agents as active. This failure reproduces at all three commits.
    • Sidebar first-render hook: a transient setup timeout, not reproduced. It passes alone, in the scratch checkout at 2de1eb5463, and in the full requested set. The test and SidebarLayoutSync are identical across all three commits. Load is plausible, but the precise trigger remains unproven.

    The #105 merge changed only effort-label formatting in this component; it did not cause these failures. Historical evidence is in history-evidence.txt, with test runs at c942f0d6ed, effd415abe, and 2de1eb5463.

    Changed three assertions in ThreadRelationshipsControl.agents.fork.browser.test.tsx: both Stop expectations now use the parent/subagent input, and the supervisor heading includes all active statuses. Product code needed no change. The commit message records the causes.

    Validation passed:

    • Four focused files: 32/32 tests — log.
    • Full requested sidebar, issues, and ThreadRelationshipsControl* set: 311/311 tests across 39 files — log.
    • Web tsc-rs --noEmit: exit 0, with advisory suggestions — log.
    • Targeted lint and format check: passed.

    The scratch worktree was removed. No repo-wide checks or app servers were run.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions