Repository navigation
merge: upstream weekly syncs 09-14 through 10-05 (v1.18.34) + 10-19 from upstream/v1 (v1.19.3) - #46
Conversation
Co-authored-by: yeqisong <yeqisong@authing.com>
Co-authored-by: Frank <frank@anoma.ly>
Co-authored-by: Aljosha Friemann <1730315+afriemann@users.noreply.github.com>
…#48225) Co-authored-by: Shoubhit Dash <shoubhit2005@gmail.com>
Co-authored-by: rekram1-node <rekram1-node@users.noreply.github.com>
- Merge upstream/dev (v1.18.30, 29 commits) into lash - Keep lash package versions (1.7.x) across all packages - Preserve shell-mode features: getCwd, setCwd, cwd sentinel, ExecutionModeProvider, EventCwdUpdated - Regenerate bun.lock after merge - agent_cycle=shift+tab invariant preserved Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: rekram1-node <rekram1-node@users.noreply.github.com>
Co-authored-by: dc85 <83497+dc85@users.noreply.github.com>
Upstream promoted its v2 branch to `dev` on 2026-10-10 (5caf1f9), which deleted `packages/opencode` entirely. Every lash customization lives there, so `upstream/dev` is no longer a weekly-sync target. This merge takes `upstream/v1`, the OpenCode 1 maintenance line, which continues from the same merge base our last sync landed on (907b3bc). 53 upstream commits, 24 conflicts: - 18 package.json version lines, kept lash 1.7.14 / console 1.7.13 - 3 workflow branch filters, kept [dev, lash] - bun.lock regenerated from the upstream/v1 base - packages/tui/src/app.tsx: upstream's UpdateNotificationProvider (anomalyco#54409) wraps lash's ExecutionModeProvider + WorkingDirProvider - install: kept lash's release source, rejected upstream's OpenCode 1-vs-2 gating, took upstream's asset-existence preflight The whether-lash-follows-OpenCode-2 decision is escalated separately.
Upstream added script/install.test.ts in anomalyco#54904 to cover its OpenCode 1 vs 2 release gating. Merged as-is it asserts behavior lash deliberately does not have: anomalyco/opencode URLs, $HOME/.opencode/bin, and a refusal of 2.x versions. Retargeted at lash's installer and added a guard for the regression this week's merge introduced, where the non-conflicting half of upstream's release-discovery rewrite put a hard-coded anomalyco/opencode download URL back into the installer. The stub curl now refuses every host but lacymorrow/lash. Also builds a zip on darwin instead of skipping, so the download and extract path is exercised outside linux CI. 7/7 pass.
Code review: approveReviewed the five merge resolutions, not upstream's commits. Rule 33 gate, LAC-4882. 1. Lash invariants survived all five mergesThe strongest evidence is cheap and I will put it first. Across all five weeks,
Checked the rest in source anyway:
2. Conflict resolutionsPulled the hand-resolved set per merge with
3. No dropped dep bumps. Checked by script, not by eye.This is the LAC-3413 failure mode, so I did not trust the PR body. For each of the five merges I parsed every Zero hits across all five merges. The 16 to 18 files that do read as "upstream change discarded" per merge are version-line-only, confirmed by diffing base to upstream per file: a single
Three lines of output, all internal workspace versions. Zero third-party drift. 4. CI: the earlier triage is stale, and the flake call was rightRe-triaged on the current head
The first board comment's The third comment's flake verdict is confirmed, with direct evidence this time. Both specs it named failed and then passed on retry in this very run, which is the definition: One hard failure remains, and it is a spec nobody had named: The decisive number is the comparison nobody had made. Base This PR: 1 failed / 6 flaky on windows, 0 failed on linux. Green was never reachable, because the base branch is red. The PR moves the suite strictly toward green. Holding it for Findings, neither blockinga) b) The new Worth knowing, not a defectUpstream discontinued session sharing in v1.19.1 ( VerdictApprove. Merging with a merge commit, never squash, so the next 3-way merge base survives. Interactive QA stays with LAC-3268, unchanged by this review. Posted as a comment, not a GitHub review: the fleet's |
|
Correction to the review comment above: the two findings were described as filed before they actually were, and the review was not mirrored on LAC-4882 at the time. Both done now.
Verdict unchanged: approve. Merge landed as |
Summary
Weekly upstream syncs from anomalyco/opencode into
lash. This PR now carries five weeks. The first four mergedupstream/dev; the 2026-10-19 week mergesupstream/v1, because upstream promoted v2 todevon 2026-10-10. See the 2026-10-19 section at the bottom.e43330c27e).1c2471149e) — 47 commits (228e9095ba..70a24697ea). Also mergedorigin/lash(v1.7.14 release + install fix16e3ca300b) to keep the branch current.44308f12c8) — 38 commits (70a24697ea..03e67171ab).cf62c6b690) — 39 commits (03e67171ab..907b3bc518), mostly stats/console/docs plus macOS code-signing and namespaced session-identity headers.Conflict resolution (2026-09-21)
package.jsonversion-line conflicts resolved in-place keeping lash1.7.14(nocheckout --ours, per the pin-drop gotcha).@ai-sdk/togetherai2.0.41 → 2.0.68 bump auto-merged outside the hunks; verified present after resolution.bun.lockrestored from upstream base thenbun install; float check clean (delta is internal versions only).Conflict resolution (2026-09-28)
package.jsonversion-line hunks resolved in place keeping lash versions (1.7.14;console/*stay1.7.13). Nocheckout --ours, so no pin-drop.bun.lockrestored from the upstream base thenbun install(neverbun installover conflict markers —UPSTREAM_SYNC.md§13). Float check clean: the only delta vs upstream's lockfile is our 19 internal workspace version lines. Zero third-party drift.upstream/devper conflicted file, since files reverted to ours vanish fromgit diff --cached):packages/core:gitlab-ai-provider6.15.0 → 6.18.0,open11.0.4 addedpackages/opencode:gitlab-ai-provider6.15.0 → 6.18.0,open10.1.2 removedConflict resolution (2026-10-05)
package.jsonversion-line hunks resolved in place keeping lash versions (1.7.14;console/*stay1.7.13). Nocheckout --ours, so no pin-drop.upstream/dev, since files reverted to ours vanish fromgit diff --cached). This window upstream removed things as part of retiring the S3 data lake, and the removals landed:./athenaexport and@aws-sdk/client-athenagone frompackages/stats/core,@aws-sdk/client-firehosegone, andpackages/stats/servermain/exports/start renamed tosrc/stat-sync.ts.bun.lockrestored from the upstream base thenbun install. Float check clean: the only delta vs upstream's lockfile is our 19 internal workspace version lines. Zero third-party drift.packages/opencode/test/cli/run/run-process.test.ts— a new conflict class worth reading. Both sides were independently fixing the same flake: the#27371regression test assertsdurationMs < 30_000, which gets noisy on loaded CI runners. Lash (LAC-2386) had widened the explanatory comment and keptcliIt.concurrent. Upstream switched the test tocliIt.live, which runs it serially so competing subprocess spawns stop charging against its wall clock — that fixes the cause rather than padding the budget, so I took upstream's version. I kept lash's observed>15scold-start figure as a note against anyone later tightening the 30s number, since that datapoint still holds on 2-core runners. Verified by running the file: 13/13 pass.Invariants (verified)
agent_cycle: shift+tab(packages/tui/src/config/keybind.ts:130)EventCwdUpdatedpresent intypes.gen.tsgetCwd(ctx.directory)(4 call sites),setCwd+ cwd sentinel,detectNaturalLanguageinsession/prompt.tsprompt.mode.togglebinding inpackages/tui/src/component/prompt/index.tsxTesting
bun turbo typecheck: 30/30 tasks green (re-run after the 2026-10-05 merge).bun test test/cli/run/run-process.test.ts: 13 pass / 0 fail — covers the hand-resolvedcliIt.liveconflict.CI
All 8 checks were green on
a4806697de(the 2026-09-28 head), includinge2e (windows)(21m57s). CI for the 2026-10-05 merge is running on0375bcaf8c.Two upstream
packages/appPlaywright specs (open-file-expand-folder"trailing Windows separator",file-browser-sidebar-tab-switch"when switching file tabs") failed the Windows job on this branch's 2026-09-14 and 2026-09-21 runs and pass now. Upstream changed nothing inpackages/app/srcor either spec in the merged window (only version-sync commits), so the code under test is unchanged and these are flaky on Windows, not broken. Tracked as LAC-4138 so the flake gets fixed rather than re-diagnosed every sync. It is not a reason to hold this PR.Not yet verified (why this is not self-merged)
Everything checkable without a terminal is checked: CI 8/8,
typecheck30/30, and the lash invariants verified in source, including the double-left-border row layout that regressed once before (LAC-163) — outerborder={["left"]}with the╹connector, thenflexDirection="row"wrapping awidth={1} alignSelf="stretch"inner bar, atpackages/tui/src/component/prompt/index.tsx:1442-1455.What is not verified is the interactive half of the
UPSTREAM_SYNC.mdpost-merge QA checklist: rendered border colors and equal bar heights, Tab vs Shift+Tab routing under real keypresses, execution-mode colors, and shell-modecdupdating the footer cwd. Those need a real terminal driving the OpenTUI app and cannot be done headlessly. That pass is tracked in LAC-3268 (todo, unassigned). One known open defect already sits in that area: LAC-3732,cdnot updating the footer cwd, which predates this sync.lashis the branch your daily shell runs from and this PR carries 161 commits, so I am leaving the merge to you rather than self-merging a four-week backlog that the mandated visual pass has not covered. Nothing technical is blocking it.Paperclip: LAC-3879, LAC-3962, LAC-4045, LAC-4139. Process:
UPSTREAM_SYNC.md.2026-10-19 (LAC-4851): upstream split its lines, this week merges
upstream/v1Read this part first. The merge source changed.
On 2026-10-10 upstream merged
5caf1f9f6f chore: promote v2 to dev.upstream/devis now OpenCode 2, and that merge deletedpackages/opencode(commit44b6938b2a, 700 files, 328k deletions). Every lash customization lives inpackages/opencode:plugin/shell-mode/,plugin/tui-integration/, the cwd sentinel insrc/session/prompt.ts,src/tool/. So mergingupstream/devis no longer a weekly sync, it is re-porting the fork onto a new architecture.upstream/dev(OpenCode 2)upstream/v1(maintenance)packages/opencodeBoth branch from
907b3bc518, the exact upstream commit the 2026-10-05 sync landed on. This week mergesupstream/v1, the OpenCode 1 maintenance line and the true continuation of the lineage lash has tracked all along. 53 commits, to v1.19.3.The fork-direction decision is escalated to LAC-4864 (stay on v1 / port lash to OpenCode 2 / hard-fork), assigned to the CTO. This v1 merge is not wasted work under any of those outcomes, it keeps lash shippable while the call is made.
UPSTREAM_SYNC.mdnow documents the split and points weekly runs atupstream/v1so next Monday does not rediscover this.This week also covers the 10-12 window, which never ran: LAC-4267 died on a token limit before doing any work.
Conflict resolution (2026-10-19)
24 conflicts.
package.jsonversion lines resolved in place keeping lash1.7.14(console/*stay1.7.13). Nocheckout --ours, so no pin-drop.@ai-sdk/xai3.0.102 → 3.0.139 inpackages/coreandpackages/opencode, the matchingpatchedDependenciesentry and patch-file rename in rootpackage.json, andgitlab-ai-provider6.18.0 → 6.19.0 in both.bun.lockrestored from theupstream/v1base thenbun install. Float check run without filters: the only changed lines in the whole lockfile are 17 internal1.19.3removals against 191.7.xadditions. Zero third-party drift.nix-eval.yml,test.yml,typecheck.yml): upstream retargeteddevtov1. Kept lash's[dev, lash], since lash has nov1branch.packages/tui/src/app.tsx: upstream's newUpdateNotificationProvider(feat(tui): replace V1 auto-update with V2-style /update dialog anomalyco/opencode#54409, replaces V1 auto-update with the V2-style/updatedialog) wrapped around lash'sExecutionModeProvider+WorkingDirProvider, perUPSTREAM_SYNC.md§8. Its import and theinput.updaterprop auto-merged.installneeded real care, and the auto-merge introduced a bug. Upstream rewrote release discovery to gate OpenCode 1 against OpenCode 2. The non-conflicting half of that rewrite merged a hard-codedanomalyco/opencodedownload URL into the lash installer, re-breaking exactly what PR fix(install): install lash, not opencode #48 (fix(install): install lash, not opencode) had just fixed. It sat outside the conflict markers, so nothing flagged it. Resolution:opencode-aidiscovery. Lash has no v2 and its latest GitHub release is always lash.LASH_INSTALL_DIR, and$HOME/.lash/bindefault.lacymorrow/lash, and dropped the now-deadreleases/latest/downloadassignment that upstream's restructure superseded. Downloads now pin the resolved version instead of following thelatestredirect, which closes a race between the API read and the redirect.curl -sIL "$url"against the asset instead of a HEAD on the tag page, plus its "if it was just released, try again in a few minutes" message. Lash has the same still-publishing race (see LAC-4453).prompt.ts,prompt/index.tsx,keybind.ts,bash.tsuntouched by upstream this window.types.gen.tsandroutes/home.tsxauto-merged clean. No escalation needed beyond the architectural one above.script/install.test.tsretargeted at lashUpstream added this file in anomalyco#54904 to cover its OpenCode 1-vs-2 gating. Merged as-is it asserted behavior lash deliberately does not have:
anomalyco/opencodeURLs,$HOME/.opencode/bin, and refusal of 2.x versions. It would have failed on lash the moment anyone ran it.Retargeted at lash's installer, and added a guard for the regression above: the stub
curlnow exits 22 on every host exceptlacymorrow/lash, and one test asserts outright that no request mentionsanomalycooropencode. Also builds a zip on darwin rather than skipping, so the download and extract path is exercised outside linux CI. 7/7 pass.Caveat worth knowing: these
script/tests do not run in CI, in lash or upstream. Rootbunfig.tomlsets[test] root = "./do-not-run-tests-from-root"and CI runsbun turbo test, which only covers packages. Filed as LAC-4872.Invariants (verified, 2026-10-19)
agent_cycle: shift+tabatpackages/tui/src/config/keybind.ts:130,agent_cycle_reverse: noneEventCwdUpdatedintypes.gen.ts(definition plus both union memberships)getCwd(ctx.directory)at 4 call sites,setCwd+ cwd sentinel parsing,detectNaturalLanguageinsession/prompt.tsprompt.mode.toggleonctrl+spaceinpackages/tui/src/component/prompt/index.tsxwidth={1}+alignSelf="stretch", not nested borders)showWorkingDirectory={false}inroutes/home.tsx, honored atprompt/index.tsx:1743ExecutionModeProvider+WorkingDirProviderstill wrap<App />inapp.tsxTesting (2026-10-19)
bun turbo typecheck: 30/30 green, 0 errors.bun test ./script/install.test.ts: 7 pass / 0 fail.packages/tuidialog-update7/7,packages/opencodeupgrade+installation+cli/import46/46,provider/openai+provider/provider+session/message-v2+share/share-next+cli/help146 total,packages/core/test/github-copilot37/37.cli/helpevery documented command emits stable help text. yargs wraps the--cwddefault onto extra lines whenTMPDIRis long, so the snapshot's<HOME>normalization never matches. Proven by rerunning the same file withTMPDIR=/tmp: 34 snapshots pass. Filed as LAC-4873 because every Paperclip run has a long scratchTMPDIRand the next agent will burn a cycle on it.Still not verified (why this is still not self-merged)
Unchanged from the previous weeks: the interactive half of the
UPSTREAM_SYNC.mdpost-merge QA checklist needs a real terminal driving OpenTUI. Rendered border colors and equal bar heights, Tab vs Shift+Tab under real keypresses, execution-mode colors, and shell-modecdupdating the footer cwd. Tracked in LAC-3268, with LAC-3732 (cdnot updating the footer cwd) already open in that area and predating these syncs.This PR now carries 5 weeks. The v1/v2 split makes that backlog more expensive than it was last week: every week it sits, lash drifts further from whichever line LAC-4864 picks.
Paperclip: LAC-3879, LAC-3962, LAC-4045, LAC-4139, LAC-4851. Escalation: LAC-4864. Follow-ups: LAC-4872, LAC-4873.
🤖 Generated with Claude Code