refactor(coil): retire the local desktop autobuild - #92
Merged
Conversation
Update delivery superseded it. `auto-build-desktop.sh` polled `origin/main` every 12h, built a dmg on this Mac and copied it over `/Applications`; since #51/#55 CI builds every green merge to main, signs it with the same identity, and the app offers the restart itself. Two mechanisms racing to replace the same bundle is worse than either alone — and the local one has not run since 2026-08-07, so this is recording a fact rather than making a decision. One behavioural difference, stated rather than discovered later: releases now gate on green CI, because coil-release.yml triggers on `workflow_run` of the CI workflow with `conclusion == 'success'`. The autobuild did not care. A flaky test therefore withholds a build until CI is green — the escape hatch is `gh workflow run coil-release.yml -f sha=<sha>`. Goes with it: the opt-in `post-merge` git hook (its only job was invoking the script) and the runbook. The bundle-id test drops its autobuild arm — `build-desktop-artifact.ts`'s env hook is still set by the release workflow, so the seam and its guard both stay, unchanged at +11/-1. The ledger header is untouched: nothing upstream-owned moved. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Important Review skippedAuto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Organization UI Review profile: ASSERTIVE Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Uh oh!
There was an error while loading. Please reload this page.
radroid added a commit
that referenced
this pull request
Aug 12, 2026
app.asar 189.66 MiB / 14,765 files -> 99.02 MiB / 3,429 files (-47.8%, -76.8% files). The updater's .zip drops 20.1 MB; the electron-builder phase drops 62%. Three cuts, each measured from the packed asar rather than the install tree: - Web renderer source maps (37.4 MiB) via T3CODE_WEB_SOURCEMAP=0, an existing upstream variable. All 385 renderer chunks are byte-identical afterwards apart from the trailing sourceMappingURL comment. - Third-party source maps (18.2 MiB). First-party maps are kept, so NODE_OPTIONS=--enable-source-maps still gives readable frames for our code. - Two renderer-only dependency trees: Clerk's browser SDK (23.2 MiB, 16 packages) and @pierre/diffs' shiki tree (21.0 MiB, 50 packages). Both are bundled by a client build before staging, and the renderer runs sandboxed with nodeIntegration off, so neither is resolvable at runtime. Sync cost is one upstream file, +21/-0. The seam is an env hook on scripts/build-desktop-artifact.ts rather than an edited literal: upstream's test asserts DESKTOP_FILE_EXCLUSIONS deep-equals exactly its one entry, and the fork's list is 67 globs and grows, so an inline list would make every future size fix an edit to an upstream test assertion. An unset environment packages exactly what upstream packages. No new seam rows, no new deletions. Two new fork-owned tools: - desktop-bundle-reachability.mjs analyses a staged build. It follows literal import edges AND string-mention edges, because PlaywrightInjectedRuntime.ts resolves playwright-core through a const and an import-only scan calls 10.2 MiB of live code dead. - verify-desktop-bundle.mjs gates every release on both platforms, failing any packaged import that is not loadable. Found two bugs in itself under test: it would have failed every Windows release by ignoring app.asar.unpacked, and an earlier severity split passed a deliberately-broken build. Re-measured on 136b351, after the #71 Coil rename, #58's dependency re-resolution and #92 retiring the local autobuild. The packed asar is the same size on both sides of the rename, and all 66 non-wildcard globs still match a shipped package. Corrects issue #53 on five points, recorded in docs/coil/desktop-bundle-size.md — its size table measured `du` block allocation (core-js "15 MB" is 1.25 MiB of content), it misses source maps entirely, DESKTOP_FILE_EXCLUSIONS is not fork-owned, the mac leg is not the release critical path, and the pnpm store is already cached. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
radroid pushed a commit
that referenced
this pull request
Aug 17, 2026
…dored subtrees Closes two advisories against the desktop app the fork ships. Upstream is still on 41.5.0, so the sync does not deliver this — it is a fork-owned pin that retires itself once upstream passes 41.10.3. - GHSA #94 (high): a sandboxed iframe can bypass the `allow-popups` restriction via the OpenURL navigation path. Fixed in 41.10.3. - GHSA #92 (medium): `ProtocolResponse.url` reuses the default session cache instead of the registering session. Fixed in 41.9.1. Dependabot bumped `apps/desktop/package.json` alone, so `vp install` failed on the frozen lockfile in 40s and CI never reached typecheck. Regenerated `pnpm-lock.yaml` with `pnpm install --lockfile-only`; the diff stays inside the electron tree and nets -37 lines. Also adds `.github/dependabot.yml` (fork-owned; upstream has none) ignoring `.repos/**`, and records the seam-ledger rows the change owes: `apps/desktop/package.json` as row 36, and `pnpm-lock.yaml` grown to risk 12416, now the top row.
radroid added a commit
that referenced
this pull request
Aug 17, 2026
`SEAMS.md` cited `scripts/t3x/mac-signature.test.ts`, `scripts/t3x/verify-mac-signature.ts` and `docs/t3x/agents/` — all three moved under `coil/` in #71 and the ledger's prose did not follow. The files themselves were fine; `AGENTS.md` already links the right paths, so this was drift in the description, not a broken link. The bundle-id row also said "every build path sets it". Since #92 there is one build path — the release workflow. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
radroid added a commit
that referenced
this pull request
Aug 17, 2026
app.asar 189.66 MiB / 14,765 files -> 99.02 MiB / 3,429 files (-47.8%, -76.8% files). The updater's .zip drops 20.1 MB; the electron-builder phase drops 62%. Three cuts, each measured from the packed asar rather than the install tree: - Web renderer source maps (37.4 MiB) via T3CODE_WEB_SOURCEMAP=0, an existing upstream variable. All 385 renderer chunks are byte-identical afterwards apart from the trailing sourceMappingURL comment. - Third-party source maps (18.2 MiB). First-party maps are kept, so NODE_OPTIONS=--enable-source-maps still gives readable frames for our code. - Two renderer-only dependency trees: Clerk's browser SDK (23.2 MiB, 16 packages) and @pierre/diffs' shiki tree (21.0 MiB, 50 packages). Both are bundled by a client build before staging, and the renderer runs sandboxed with nodeIntegration off, so neither is resolvable at runtime. Sync cost is one upstream file, +21/-0. The seam is an env hook on scripts/build-desktop-artifact.ts rather than an edited literal: upstream's test asserts DESKTOP_FILE_EXCLUSIONS deep-equals exactly its one entry, and the fork's list is 67 globs and grows, so an inline list would make every future size fix an edit to an upstream test assertion. An unset environment packages exactly what upstream packages. No new seam rows, no new deletions. Two new fork-owned tools: - desktop-bundle-reachability.mjs analyses a staged build. It follows literal import edges AND string-mention edges, because PlaywrightInjectedRuntime.ts resolves playwright-core through a const and an import-only scan calls 10.2 MiB of live code dead. - verify-desktop-bundle.mjs gates every release on both platforms, failing any packaged import that is not loadable. Found two bugs in itself under test: it would have failed every Windows release by ignoring app.asar.unpacked, and an earlier severity split passed a deliberately-broken build. Re-measured on 136b351, after the #71 Coil rename, #58's dependency re-resolution and #92 retiring the local autobuild. The packed asar is the same size on both sides of the rename, and all 66 non-wildcard globs still match a shipped package. Corrects issue #53 on five points, recorded in docs/coil/desktop-bundle-size.md — its size table measured `du` block allocation (core-js "15 MB" is 1.25 MiB of content), it misses source maps entirely, DESKTOP_FILE_EXCLUSIONS is not fork-owned, the mac leg is not the release critical path, and the pnpm store is already cached. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
radroid pushed a commit
that referenced
this pull request
Aug 18, 2026
…dored subtrees Closes two advisories against the desktop app the fork ships. Upstream is still on 41.5.0, so the sync does not deliver this — it is a fork-owned pin that retires itself once upstream passes 41.10.3. - GHSA #94 (high): a sandboxed iframe can bypass the `allow-popups` restriction via the OpenURL navigation path. Fixed in 41.10.3. - GHSA #92 (medium): `ProtocolResponse.url` reuses the default session cache instead of the registering session. Fixed in 41.9.1. Dependabot bumped `apps/desktop/package.json` alone, so `vp install` failed on the frozen lockfile in 40s and CI never reached typecheck. Regenerated `pnpm-lock.yaml` with `pnpm install --lockfile-only`; the diff stays inside the electron tree and nets -37 lines. Also adds `.github/dependabot.yml` (fork-owned; upstream has none) ignoring `.repos/**`, and records the seam-ledger rows the change owes: `apps/desktop/package.json` as row 36, and `pnpm-lock.yaml` grown to risk 12416, now the top row.
radroid added a commit
that referenced
this pull request
Aug 18, 2026
`SEAMS.md` cited `scripts/t3x/mac-signature.test.ts`, `scripts/t3x/verify-mac-signature.ts` and `docs/t3x/agents/` — all three moved under `coil/` in #71 and the ledger's prose did not follow. The files themselves were fine; `AGENTS.md` already links the right paths, so this was drift in the description, not a broken link. The bundle-id row also said "every build path sets it". Since #92 there is one build path — the release workflow. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
radroid added a commit
that referenced
this pull request
Aug 18, 2026
app.asar 189.66 MiB / 14,765 files -> 99.02 MiB / 3,429 files (-47.8%, -76.8% files). The updater's .zip drops 20.1 MB; the electron-builder phase drops 62%. Three cuts, each measured from the packed asar rather than the install tree: - Web renderer source maps (37.4 MiB) via T3CODE_WEB_SOURCEMAP=0, an existing upstream variable. All 385 renderer chunks are byte-identical afterwards apart from the trailing sourceMappingURL comment. - Third-party source maps (18.2 MiB). First-party maps are kept, so NODE_OPTIONS=--enable-source-maps still gives readable frames for our code. - Two renderer-only dependency trees: Clerk's browser SDK (23.2 MiB, 16 packages) and @pierre/diffs' shiki tree (21.0 MiB, 50 packages). Both are bundled by a client build before staging, and the renderer runs sandboxed with nodeIntegration off, so neither is resolvable at runtime. Sync cost is one upstream file, +21/-0. The seam is an env hook on scripts/build-desktop-artifact.ts rather than an edited literal: upstream's test asserts DESKTOP_FILE_EXCLUSIONS deep-equals exactly its one entry, and the fork's list is 67 globs and grows, so an inline list would make every future size fix an edit to an upstream test assertion. An unset environment packages exactly what upstream packages. No new seam rows, no new deletions. Two new fork-owned tools: - desktop-bundle-reachability.mjs analyses a staged build. It follows literal import edges AND string-mention edges, because PlaywrightInjectedRuntime.ts resolves playwright-core through a const and an import-only scan calls 10.2 MiB of live code dead. - verify-desktop-bundle.mjs gates every release on both platforms, failing any packaged import that is not loadable. Found two bugs in itself under test: it would have failed every Windows release by ignoring app.asar.unpacked, and an earlier severity split passed a deliberately-broken build. Re-measured on 136b351, after the #71 Coil rename, #58's dependency re-resolution and #92 retiring the local autobuild. The packed asar is the same size on both sides of the rename, and all 66 non-wildcard globs still match a shipped package. Corrects issue #53 on five points, recorded in docs/coil/desktop-bundle-size.md — its size table measured `du` block allocation (core-js "15 MB" is 1.25 MiB of content), it misses source maps entirely, DESKTOP_FILE_EXCLUSIONS is not fork-owned, the mac leg is not the release critical path, and the pnpm store is already cached. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
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 freeto 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.
Update delivery superseded the local autobuild. Removing it.
Why
scripts/coil/auto-build-desktop.shpolledorigin/mainevery 12h, built a dmg on the maintainer's Mac and copied it over/Applications. Since #51/#55, CI builds every green merge to main, signs it with the same identity, and the app offers the restart itself — so the two mechanisms now race to replace the same bundle.It has not run since 2026-08-07, and its LaunchAgent is not loaded. This mostly records a fact.
The one behavioural difference
Releases gate on green CI:
coil-release.ymltriggers onworkflow_runofcoil fork CIwithconclusion == 'success'. The autobuild did not care. So a flaky test now withholds a build rather than merely reddening CI — main's knownReactor.test.tsflake already skipped one release today.Escape hatch, unchanged:
gh workflow run coil-release.yml -f sha=<sha>.What goes
scripts/coil/auto-build-desktop.shdocs/coil/auto-build-runbook.mdscripts/coil/hooks/post-mergePlus the stale prose it left in
mac-signing-runbook.md,SEAMS.md,setup-mac-signing.shand twoupdateDeliverycomments.What deliberately stays
T3X_DESKTOP_APP_IDseam inscripts/build-desktop-artifact.ts. The autobuild was one of two things that set it; the release workflow is the other, and still needs it. The row is unchanged at +11/-1, so the ledger header (47 files, +2234/-1067) does not move.mac-signature.test.ts, minus the arm that read the deleted script. The remaining three assertions still fail loudly if a sync reverts the seam.SEAMS.mdexplaining that the local build loop existed and why it went, because "why is there no local build loop" is a question the next reader will have.Verification
🤖 Generated with Claude Code