monorepo follow-ups: glance dashboards and poller, gitq undo, extension bundle (test first) - #519
Merged
Merged
Conversation
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…ll tick A cold tick that threw left the poller warm, so the retry replayed the first page as invalidations; a tick that threw on page 2 kept page 1's new etag, so the next tick 304'd past the unseen pages. Both are now staged in locals and committed once every page has been fetched. Fixes GLANCE-36 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
createSingleDashboard built a status listener it never handed to watchMR, so its connection stayed 'connecting' forever. RealtimeWatcherOptions now carries onStatusChange (watchMR already spreads its options into the watcher), and the watcher reports 'disconnected' after a connected cable drops instead of falling back to 'connecting'. Fixes GLANCE-37 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…hers The group dashboard called watchMR once per MR only to learn about pushes, which ran a fetch and poll loop per MR and dropped every event and connection change the shared cable delivered. GitProvider gains an optional push-only subscribePullRequestEvents; GitLabProvider implements it over the same shared cable as watchMR (the registration is now one private method both use), and the group subscribes once for the whole group, so a cable event triggers one batched refetch and connect or disconnect moves the group's connection state. Behaviour change: on GitHub, where the method is absent, group dashboards no longer call watchMR (which throws there); they report connected and poll, as the branch dashboard already does. Fixes GLANCE-38 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The two poller throw tests reuse scriptedFetch (now scripted Errors throw), the single dashboard test pins reconnect back to connected, and a watcher test pins that the params onStatusChange wins over the options one. Refs GLANCE-36, GLANCE-37 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
integration.live.ts picked its mutation targets from the token user's own PR and MR lists, so a hand run with real credentials wrote notes, approvals and merges onto real, unrelated work. Every mutating step now takes its target from GLANCE_HARNESS_GITHUB_SANDBOX (owner/repo) or GLANCE_HARNESS_GITLAB_SANDBOX (group/project!iid) and refuses, sending no request, while it is unset. Read-only probes keep the own-list fallback. GITLAB_PROJECT_PATH no longer steers mutations. Target selection lives in tests/live/sandboxTarget.ts with unit tests; the live run against the sandbox is still to do. Refs GLANCE-39 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
undo checked out each snapshotted branch and reset --hard it in the launch tree: uncommitted edits carried across the checkout were discarded, trunk was checked out and reset too (throwing part way when another worktree held it, with the store update skipped), and commits made since the operation were dropped. Now undo refuses a dirty launch tree, skips the stack root, and checks every branch before moving any: one that moved since the operation refuses the whole undo, naming each. The operation log records resultHeads (every branch's head once the operation finished) for that check; older entries fall back to the head read at undo time. Each ref then moves through finalizeBranchRef from its result head to its snapshot sha, so undo works from any worktree. A move that still fails part way stops, names the branch, and reports what was already restored; the stack record is updated only when every branch restored. Fixes GITQ-27 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A group fetch that returns an MR its push subscription does not cover now re-attaches (new before old, so the shared cable never tears down in between, and without a spurious reconnect refetch). A watcher attached while the shared cable is connecting or down is told it is connected only on the next welcome, not immediately. Shared-cable log lines drop the watchMR prefix since groups use the cable too. Tests add a burst of events from different MRs giving one refetch, and watchMR's subscribe, refetch, reconnect and dispose against a fake socket. Fixes GLANCE-38 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GitHub's sandbox now parses as exactly owner/repo (no extra segments, no !iid) and says so in its refusal. Path segments must be [\w.-]+ and never . or ..; an iid must be 1 to 10 digits with no leading zero. A sandbox MR that is missing fails that step instead of killing the run, one that is not open is refused, and any refusal makes the run exit non-zero so an all-refused run cannot read as success. Refs GLANCE-39 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A branch already at its snapshot now counts as restored, so rerunning undo after one that stopped part way finishes it instead of refusing those branches as moved. The precondition pass also applies finalizeBranchRef's own refusals (holding worktree dirty, mid-rebase, or off the branch head) and refuses with nothing moved, naming each branch and its slot; finalizeBranchRef keeps them as the race guard. A failed undo's JSON no longer carries the snapshot tree, since the stack record was not touched. The partial path is tested for real: a reference-transaction hook makes git reject one branch's update, then a retry without it completes. Fixes GITQ-27 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GitLab keys cable channels by MR, so re-subscribing the whole group and then disposing the old subscription unsubscribed every MR already covered: after a late MR arrived, only that MR kept push. The group now subscribes just the uncovered MRs and keeps every subscription until restart or dispose; the added ones forward events only, so connection changes are reported once. Tested against the real GitLabProvider over a fake socket. Fixes GLANCE-38 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…s it
Two load failures killed every shipped VSIX. esbuild picked jsonc-parser's
UMD entry, whose runtime require('./impl/format') the bundle cannot
satisfy; the build now prefers ESM entries (--main-fields=module,main).
And rt-client's bare require("bun:sqlite") made Bun's bundler emit a
module-scope createRequire(import.meta.url), which a CJS bundle executes
at load with an empty import.meta. repos.ts now builds that require from
the module URL inside loadBunSqliteDatabase and returns null when there is
none, since a CJS bundle runs outside Bun. ESM behaviour is unchanged.
scripts/load-smoke.mjs loads the bundle in plain node with a stub vscode
module, run by a bun test in the extension and by the release job right
after bun run package, so a VSIX that cannot load fails the release.
Fixes RT-327
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Deck neither registers nor serves gitq, so the manifest's dev.deploy (deck restart gitq) could never run; the manifest stays for its bundle recipe. hasStagedDiff no longer destructures a stdout it never reads. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The partial-undo tests wrote their reference-transaction hook into the common dir's hooks/, which a global core.hooksPath silently bypasses. They now share rejectRefUpdates, which pins a repo-local core.hooksPath, so the hook always fires and never lands in a global hooks directory. Refs GITQ-27 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Refs RT-327 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ings A failed run exited inside the try, skipping the finally that removes the stub vscode dir; it now sets exitCode. The bun test asserts the exit status and the ok line instead of an empty stderr, so a future node deprecation warning cannot fail it. Refs RT-327 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Also notes on rt-client's dist-cjs-safe test that its regex matches Bun's current emit shape, with the rt-context bundle-load test as the backstop. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Warning Review limit reachedNext included review available in 32 minutes. View limit detailsLimit details: You’ve used all 2 included reviews currently available. Your 71 included PR review attempts over the past 7 days set your current allowance at 2 reviews per hour. This review is too large to run within your organization's remaining usage spending cap. Raise or remove your spending cap in the billing tab, then retry. Review configuration: ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Advanced Run ID: 📒 Files selected for processing (38)
Comment |
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.
Closes the review follow-ups left over from the monorepo fold-in, each fixed test-first: four glance bugs, the gitq undo data-loss bug, and the VS Code extension bundle that never loaded. Plan:
docs/superpowers/plans/2026-09-27-monorepo-followups.md.What changed
glance (
packages/glance)hasTickedand its etag only after a full tick, so a failed first tick retries cold and a failed later page never skips unseen events (GLANCE-36)disconnectedafter a drop once it has connected (GLANCE-37)GitProvider.subscribePullRequestEvents: one fetch and poll loop per group, every cable event reaches the group once, late MRs attach incrementally, and attaching during a cable outage waits for the welcome (GLANCE-38). GitHub groups no longer call GitHub's throwingwatchMR; they pollGLANCE_HARNESS_GITHUB_SANDBOXandGLANCE_HARNESS_GITLAB_SANDBOX, refuses otherwise, and exits non-zero on any refusal (GLANCE-39)gitq (
apps/gitq)gitq undomoves refs by compare-and-swap from the heads the operation recorded to the snapshot, never checks out, never moves trunk, refuses a dirty tree or a dirty, rebasing or moved holding worktree up front, and can resume a partial undo left by a race (GITQ-27)dev.deployfrom the manifest and an unused destructureVS Code extension (
extensions/vscode/rt-context,packages/rt-client,release.yml)bun:sqliterequire inside the function instead of at module scopebun run packagein the release job, and the extension has atestscript sobun run checkruns its testsFollow-up
packages/glance/harness_credentials.json.vsixinstalls and activates in an editor) is Matt's to checkVerification
bun run checkgreen (46/46 package tasks, purity ok); rt unit suite 11453 pass with the six known rotating flakes green alone; e2e 150 pass (one failure from this machine's mise node shim in the plugin scaffold test); pty 8/8; glance 620/620, gitq 527 unit and 431 integration, rt-client 946/946, extension 22/22. Every task had an Opus review and scoped re-reviews of its fixes, then an Opus whole-branch review. Main merged in at 8ab1df5.🤖 Generated with Claude Code