Files
orca/.github/workflows/pr.yml
T
f36c03e84a fix(windows): make the install-dir ACL repair rescue the launch it runs in (#18361)
* fix(windows): repair the poisoned install-dir ACL before the window, not after

The install-dir LPAC ACL poison (electron/electron#51761) still costs every
affected machine at least one crash: the probe that detects it is
setImmediate-deferred and answers 0.9-3.0s in, while createMainWindow runs
synchronously in the same frame and its renderer dies at init 48-1373ms later.

- Persist the poison verdict the moment the probe reports it, and await the
  repair (bounded at 20s) before any window is created on a launch that already
  carries the marker.
- Do not engage the GPU safe-graphics fallback while the install-dir ACL verdict
  is poisoned or still outstanding. Safe graphics does not rescue a poisoned
  tree, and --in-process-gpu removes the GPU child, erasing the sibling-death
  evidence that identifies the shape (4 field reports landed in 'misc' this way).
- Clear the safe-graphics marker once the repair lands, so a repaired machine
  stops launching software-rendered for the rest of that build.
- Give the repair marker a bounded retry budget: it was written on failure and
  matched regardless of outcome, so one transient failure pinned a machine to
  'marker-hit' for the life of that version.

* test(windows): pin the install-dir ACL repair against the real icacls binary

* fix(windows): stop the install-DACL verdict from outliving the evidence

Adversarial review round 1. Five blocking findings, all addressed.

1. gpu-lifecycle guard had only a source grep (green with the polarity
   inverted). The stated justification -- that gpu-lifecycle's import graph
   cannot be driven in-process -- was wrong: mocking `electron` plus
   `@electron-toolkit/utils` imports it fine. Replaced with
   gpu-lifecycle-install-dir-acl-guard.test.ts, which drives the real
   handleGpuChildCrash against a stub tracker. All four cases go red when the
   guard is flipped to `if (!isInstallDirAclSuspect())`.

2. A clean probe verdict retired the on-disk marker but not the in-memory
   `poison` verdict, so a machine the probe just proved healthy kept
   suppressing the GPU safe-graphics fallback and kept the dialog accusing the
   install folder -- permanently, since a `status:'failed'` probe deliberately
   keeps the marker. A positive clean reading now latches `installDirReadClean`,
   drops the verdict, and outranks a repair result that lands after it (a
   'failed' from a repair with nothing left to fix must not re-accuse).
   'repaired' is kept: it is not a contradiction and it is what tells the user
   to reload.

3. `noteWindowsInstallDirAclProbePending()` ran on every `openMainWindow` while
   the probe is once-per-process, so every tray/second-instance reopen armed a
   15s window in which `recordGpuCrash` was never called at all -- on healthy
   machines. `probeWindowsInstallDirAcl` now reports whether THIS call
   dispatched, and only a dispatch arms the grace window.

4. The pre-window ordering guarantee was defeatable and untested.
   `focusExistingMainWindow` opens a window whenever there is none and the app
   is ready -- true for the whole 20s gate, which is exactly when a user
   double-clicks the shortcut again. Added a `canOpenWindow` seam (same
   'pending' semantics as the existing `!app.isReady()` case) wired to
   `isBlockingInstallDirAclRepairInFlight()`, plus
   windows-install-dir-acl-startup-wiring.test.ts pinning the await ahead of
   both window-creation paths and both new call sites.

5. windows-install-dir-acl-repair.win32.test.ts was absent from the pr.yml
   win32 allowlist, so it ran nowhere. Added.

Also from the non-blocking list:
- The repair no longer clears a `userConfirmed: true` safe-graphics marker;
  "keep safe graphics" is a user choice, not Orca's automatic latch.
- `repairWindowsInstallDirPackageAcl` now reports its dispatch too, so a second
  entry into the gate resolves immediately instead of eating the full 20s
  budget waiting on an `onDone` that is never coming.
- The gate is wrapped in try/catch/finally, matching the contract the probe
  documents as mandatory for anything upstream of window creation.

Rebutted, not applied:
- "Gate should be conditioned on app.isPackaged." A dev launch only carries the
  poison marker if a dev launch actually probed that tree and found the
  signature, in which case the dev renderer is dying the same way and the
  repair is exactly what is needed. The adjacent `isPackaged` check guards a
  packaged-only early-window optimisation, not a correctness boundary.
- "Fold the poison marker into the repair marker's `outcome`." They answer
  different questions with different lifetimes. The repair marker is a retry
  budget (`attempts >= 3` disables the repair for that version) and is never
  cleared; the poison marker is cleared by a successful repair and by a clean
  probe. A `'pending'` outcome written before the attempt would bump `attempts`,
  so three launches killed mid-repair would permanently disable a repair that
  never once ran icacls to completion.

* fix(windows): keep counting GPU crashes while the install-DACL verdict is pending

Adversarial review round 2. Both blocking findings addressed.

1. handleGpuChildCrash early-returned on isInstallDirAclSuspect() BEFORE
   recordGpuCrash, so the crash left no trace in the 30s rolling window. The
   suspect window is armed on every win32 non-serve launch, and the field
   bundles put it at 0.8-1.7s after main_window_created on hosts whose DACL is
   clean (matchesPoisonSignature=false) -- squarely inside the 2.1-6.2s
   bad-driver bursts this repo already pinned in
   gpu-crash-fallback-field-sessions.test.ts. A healthy machine with a failing
   driver could lose an entire coalesced burst and never engage safe graphics.

   The crash is now always recorded; only the engagement consults the verdict,
   and it waits for the verdict rather than acting on the suspicion
   (waitForInstallDirAclVerdict, resolved by the probe's onDone or by the
   existing 15s grace, whichever lands first).

   Deviation from the review's suggested shape, deliberately: awaiting the
   verdict before persisting anything reintroduces the exact race
   gpu-fallback-engagement.ts documents -- Chromium aborts the whole browser
   process on the 6th GPU crash, ~1.3s after the 3rd, which is less than the
   probe takes to answer. So the unconfirmed marker is written up front and
   withdrawn if the verdict comes back poisoned. A machine killed mid-wait
   still comes back software-rendered, and its marker is unconfirmed, which is
   the state the repair's own clear already retires.

   gpu-lifecycle-install-dir-acl-guard.test.ts now drives the real
   GpuCrashFallbackTracker and the real engagement path (the restart prompt
   firing is the signal) instead of a stub tracker, and covers the case the
   previous suite could not express: a burst that lands entirely inside the
   pending window still engages once the probe reports clean. Four reverts go
   red -- restoring the pre-record guard (2 tests), dropping the wait, dropping
   the post-wait re-check, and dropping the pre-wait marker write (2 tests).

2. The round-1 evidence block quoted commits, a test name and pass counts that
   no longer exist, and its real-icacls Windows run predated the commit that
   rewrote the gate. Re-run at this commit; counts and the live-Windows result
   are restated in the handoff rather than carried forward.

Also from the non-blocking list:
- 'marker-hit' conflated "already repaired" with "retry budget spent", because
  hasMarkerFor matches outcome === 'repaired' too. The result now carries
  alreadyRepaired, and the recovery maps that to stage 'repaired' -- so a launch
  killed between a successful repair and its marker clear no longer tells the
  user the folder needs an administrator, no longer latches
  isInstallDirAclSuspect() for the session, and does retire the poison marker.

Not applied, with reasoning:
- "clearGpuFallbackMarker narrowed to userConfirmed === false leaves the target
  population software-rendered after a repair." The summary was overstated and
  is corrected, but the narrowing stands: a userConfirmed marker now requires a
  clean DACL verdict, because the restart prompt that writes it is exactly what
  the gate above withholds while the install is a suspect. The population this
  family targets can no longer reach confirmMarker while poisoned.
- "writeInstallDirAclPoisonMarker re-stamps on a budget-exhausted machine
  forever." True, but on that machine the tree really is still poisoned and the
  gate resolves immediately ('skipped', no icacls spawn, no 20s wait), so the
  marker is telling the truth. Retiring it would be wrong; only a clean probe
  reading should.

* fix(windows): register the real-icacls spec and stop its teardown racing icacls

Two ratchets were red:
- windows-lane-tree-removal-boundary: the win32 spec's afterAll used raw
  rmSync on a tree two icacls.exe children had just rewritten DACLs on, which
  is the EPERM race removeTreeSync exists for.
- win32-test-lane-registration: the spec was in the pr.yml argv but not in
  WINDOWS_PACKAGE_TESTS, so a future diff touching only test files would not
  select package_windows and the spec would self-skip on ubuntu and report
  success.

* fix(windows): re-arm the GPU fallback latch when the install-DACL verdict withholds it

recordGpuCrash reports the threshold crossing exactly once and latches `engaged`.
handleGpuChildCrash consumes that report before consulting the DACL verdict, and
installDirAclClearsGpuFallback then discards it — so nothing could ever engage
safe graphics again in that process. A machine whose tree the repair fixes and
whose driver is genuinely broken stayed hardware-accelerated through an unbounded
crash loop, with no prompt and no marker.

disengage() releases only the one-shot latch; the crash window is untouched, so a
real driver burst is still never erased. Test is RED without the re-arm.

* fix(windows): keep the safe-graphics marker while an install-DACL repair is in flight

The gate dispatches a repair without arming the probe clock, so
waitForInstallDirAclVerdict() returns immediately and the withdrawal deleted the
marker inside Chromium's FATAL window (crash 6 lands ~1.3s after crash 3, well
inside the 20s gate). The process then died mid-repair, spent no attempt, and
relaunched hardware accelerated into the same gate — spawning the same GPU
children, FATALing again, forever.

Hold the marker while poison.stage is 'pending' so that launch comes back
software rendered and the next gate runs to completion. Still not engaged this
launch, so --in-process-gpu does not erase the sibling-death evidence. A
terminal verdict has no next step to rescue, so it still withdraws. Both new
tests are RED without the retention.

* fix(windows): stop a repaired marker outranking a fresh poison verdict

The probe reads the install DACL and finds it poisoned; `startRepair` dispatches;
`markerHitFor` sees a repair marker recording `outcome: 'repaired'` for the same
installDir+appVersion and reports `alreadyRepaired`, which the recovery module maps
to stage 'repaired'. So the launch that just proved the tree poisoned runs no icacls,
deletes the poison marker that arms the next launch's pre-window gate, clears the
suspect flag so `--in-process-gpu` can engage on a tree safe graphics cannot rescue,
and tells the user "Orca repaired the permissions."

Reachable whenever the tree is re-poisoned after one successful repair of the same
version, and whenever a repair reports success without clearing the tree — the silent
icacls no-op this module exists to document.

A DACL reading taken this launch now outranks the marker: `probeConfirmedPoisoned`
stops `outcome: 'repaired'` short-circuiting the repair. The attempt budget still
bounds it, so an unrepairable tree does not re-spawn icacls forever. The pre-window
gate does not set the flag — it acts on a marker from an earlier launch, not on
evidence of its own, so a recorded repair still outranks it there.

Also drives the GPU-fallback re-arm test through a repair that actually completes
'repaired', rather than a later clean probe, which is the route the review exercised.

* fix(windows): make the pre-window ACL gate act on the poison evidence it fired on

The gate fired on a poison marker — an earlier launch's DACL reading that nothing has
retired — but withheld `probeConfirmedPoisoned` from the repair, so a repair marker
recording an older success still short-circuited it. On the three-launch shape the gate
exists for (repair succeeds; tree is re-poisoned; the next launch's probe records the
poison but dies before writing its repair marker) the gate ran no icacls, deleted the
poison marker that arms every later gate, un-suspected the tree so --in-process-gpu could
engage, and told the user "Orca repaired the permissions." `applyInstallDirAclProbeVerdict`
then swallowed that launch's own reading behind `if (poison) return`.

Both callers of `startRepair` hold outstanding poison evidence, so the flag is now
unconditional (renamed `poisonEvidenceOutstanding`) and `marker-hit` means only that the
attempt budget is spent. The probe guard is narrowed to an in-flight gate repair: a reading
taken after the gate finished re-arms the poison marker and downgrades a claimed repair.

Also: withholding safe graphics now ends with the repair budget. A machine whose attempts
are spent while the signature persists was denied safe graphics on every launch for the
life of that appVersion — and had its marker deleted each time — including the healthy
installs the probe's flag-blind ACE match over-matches, where the driver really is broken.

Non-blocking, same lane: re-read `isQuitting` after the up-to-15s verdict wait, and skip
the recovered-launch prompt when the ACL gate retired the marker read before whenReady.

* fix(windows): stop a timed-out gate repair outranking a later poison reading

The gate's 20s budget expires while icacls runs on under its own 120s cap, so
the probe can read the tree poisoned while that repair is still in flight. Its
success claim then deleted the poison marker, un-suspected the tree and told the
user their permissions were fixed. The reading is now latched and outranks it.

* fix(windows): stop a gate repair claim pre-empting this launch's probe reading

Round-7 adversarial findings, both driven against the real modules:

- isInstallDirAclSuspect returned false the moment the pre-window gate set
  stage 'repaired', short-circuiting ahead of the probe-pending grace check.
  The GPU children die 48-1373ms after window creation while the probe
  answers 0.9-3.0s in, so an icacls that silently no-opped (exit 0, tree
  untouched) opened exactly that interval to --in-process-gpu on a
  still-poisoned tree - and a 'keep safe graphics' answer then pinned a
  userConfirmed marker no later repair may clear, with the poison marker
  already deleted so no later launch gates. The claim now stays provisional
  until this launch's probe corroborates it or the grace window lapses.

- A probe reading that disproves a 'repaired' claim re-armed the poison
  marker but never restored the unconfirmed safe-graphics marker the claim
  had cleared, so the next launch relaunched hardware-accelerated into the
  re-armed gate. The clear is now captured and handed back on disproof.

* test(windows): pin the nested and update-inherited grants against real icacls

The live spec asserted the grant landed on the root-level module file only.
It now also pins that the flagless /T pass reaches a nested file carrying
its own protected DACL (the shape app.asar.unpacked and node_modules have),
and that a file written after the repair inherits the (OI)(CI) root grant -
the stated reason that grant form exists.

* fix(windows): keep the recovered-launch prompt silent while the tree is the suspect

Round-8 fresh-eyes finding, driven against the real modules: the prompt
re-read the marker the pre-window gate may have retired, but never consulted
isInstallDirAclSuspect() - so after a FAILED gate (tree still a live suspect,
window blank behind the 10s reveal fallback, Keep as both defaultId and
cancelId) a 'keep it' answer pinned a userConfirmed marker no later repair
may clear, on the exact victim class the repair cannot help. The guard now
covers both gate outcomes; staying silent leaves the marker unconfirmed,
which a successful repair still retires.

---------

Co-authored-by: Orca Worker <orca-worker@localhost>
Co-authored-by: OrcaWin <alpha-eng@stably.ai>
2026-09-03 21:39:34 -07:00

1044 lines
48 KiB
YAML

name: PR Checks
on:
pull_request:
types:
- opened
- synchronize
- reopened
- ready_for_review
concurrency:
group: pr-checks-${{ github.event.pull_request.number }}
cancel-in-progress: true
permissions:
contents: read
jobs:
# Why: a README/docs-only PR used to start the full matrix (test shards,
# two package jobs, typecheck, git compat, xterm, shell contracts). Path
# filters on `on.pull_request` would drop the `verify` check entirely; this
# detector keeps verify as the required aggregate and skips the expensive jobs.
# Per-job outputs also skip git-compat/xterm/packaging/shell when those
# inputs are unchanged; empty diffs fail closed and run everything.
code_paths:
name: detect code-relevant changes
runs-on: ubuntu-latest
outputs:
should_run: ${{ steps.filter.outputs.should_run }}
native_cache_changed: ${{ steps.filter.outputs.native_cache_changed }}
mobile_dependencies: ${{ steps.filter.outputs.mobile_dependencies }}
static_analysis: ${{ steps.filter.outputs.static_analysis }}
typecheck: ${{ steps.filter.outputs.typecheck }}
git_compatibility: ${{ steps.filter.outputs.git_compatibility }}
codex_index_heal_contract: ${{ steps.filter.outputs.codex_index_heal_contract }}
xterm_patch_sync: ${{ steps.filter.outputs.xterm_patch_sync }}
shell_contracts: ${{ steps.filter.outputs.shell_contracts }}
test: ${{ steps.filter.outputs.test }}
orcad_browser: ${{ steps.filter.outputs.orcad_browser }}
cross-version-wire: ${{ steps.filter.outputs.cross-version-wire }}
managed_hook_node18: ${{ steps.filter.outputs.managed_hook_node18 }}
package: ${{ steps.filter.outputs.package }}
package_windows: ${{ steps.filter.outputs.package_windows }}
steps:
- name: Checkout
uses: actions/checkout@v6
with:
# Why blob:none: full history is needed for the merge-base diff, but historical
# file contents are not. Blobs are ~89% of this repo's pack, and Git fetches the
# few this job actually reads on demand.
fetch-depth: 0
filter: blob:none
persist-credentials: false
- name: Classify changed paths
id: filter
env:
BASE_SHA: ${{ github.event.pull_request.base.sha }}
HEAD_SHA: ${{ github.event.pull_request.head.sha }}
run: |
set -euo pipefail
# Why --no-renames: name-only rename detection can report only the destination.
# A code file moved under docs/ must still expose its code-side deletion.
CHANGED="$(git diff --name-only --no-renames --diff-filter=ACDMR --merge-base "$BASE_SHA" "$HEAD_SHA")"
echo "Changed paths:"
printf '%s\n' "$CHANGED"
printf '%s\n' "$CHANGED" | node config/scripts/pr-code-change-scope.mjs | tee -a "$GITHUB_OUTPUT"
static_analysis:
name: static analysis
needs: [code_paths]
if: needs.code_paths.outputs.static_analysis == 'true'
runs-on: ubuntu-latest
steps:
- name: Checkout
uses: actions/checkout@v6
with:
# Why blob:none: full history is needed for the merge-base diff, but historical
# file contents are not. Blobs are ~89% of this repo's pack, and Git fetches the
# few this job actually reads on demand.
fetch-depth: 0
filter: blob:none
persist-credentials: false
- uses: ./.github/actions/install-node-dependencies
with:
native-runtime: node
- name: Lint
run: pnpm exec oxlint --format github
- name: Enforce focused code-quality plugins
run: pnpm run audit:code-quality:native
- name: Enforce type-aware code-quality baseline
run: pnpm run audit:code-quality:type-aware
# Why: the changed-code gate lints mobile files too, and its type-aware pass
# resolves types from mobile/node_modules. Mobile is a separate pnpm project,
# so the root install above leaves it empty and every mobile type degrades to
# an `error` type — reported as phantom findings against the changed lines.
# Why no --ignore-scripts, unlike the root install: mobile's postinstall generates
# the gitignored terminal/mermaid webview engine modules that tracked source imports,
# and skipping it degrades those very types the step exists to resolve. The drift
# guard mirrors the root install so a stale mobile lockfile fails by name — mobile's
# lockfile carries patchedDependencies that a silent rewrite would drop.
- name: Install mobile dependencies
if: needs.code_paths.outputs.mobile_dependencies == 'true'
working-directory: mobile
run: |
pnpm install --frozen-lockfile
if [ "$(git -C "$GITHUB_WORKSPACE" rev-parse --is-inside-work-tree 2>/dev/null)" = true ]; then
git -C "$GITHUB_WORKSPACE" diff --exit-code -- \
mobile/package.json mobile/pnpm-lock.yaml mobile/pnpm-workspace.yaml
fi
- name: Enforce changed-code quality
run: pnpm run check:code-quality:changed -- "${{ github.event.pull_request.base.sha }}"
- name: Enforce React Doctor on changed lines
run: pnpm run check:react-doctor:changed -- "${{ github.event.pull_request.base.sha }}"
- name: Check Zustand selector fan-out budget
run: pnpm run check:zustand-selector-fanout
- name: Check reliability gate manifest
run: pnpm run check:reliability-gates
- name: Check VM runtime rollback compatibility
env:
BASE_SHA: ${{ github.event.pull_request.base.sha }}
HEAD_SHA: ${{ github.event.pull_request.head.sha }}
run: |
if git diff --quiet --merge-base "$BASE_SHA" "$HEAD_SHA" -- \
src/shared/ephemeral-vm-runtime-store.ts \
src/shared/ephemeral-vm-runtime-feature-store.ts \
src/shared/ephemeral-vm-runtime-rollback-projection.ts \
src/shared/ephemeral-vm-runtimes.ts \
src/shared/ephemeral-vm-recipes.ts \
src/shared/orca-yaml-hook-types.ts \
src/main/ephemeral-vm-runtime-service.ts \
src/main/ephemeral-vm-runtime-provisioning-persistence.ts \
src/main/ephemeral-vm-failed-start-cleanup.ts; then
echo "VM runtime persistence is unchanged."
exit 0
fi
node config/scripts/run-ephemeral-vm-runtime-store-rollback-repro.mjs \
config/scripts/ephemeral-vm-runtime-store-cross-version.test.ts
- name: Enforce max-lines ratchet
run: pnpm run check:max-lines-ratchet
- name: Enforce ts-nocheck ratchet
run: pnpm run check:ts-nocheck-ratchet
- name: Enforce runtime Electron-import ratchet
run: pnpm run check:runtime-electron-ratchet
# Why both: the ratchet proves nothing reachable from the runtime imports electron,
# which is a property of the import graph. This proves the Node artifact it enables
# actually boots, pairs, creates a worktree and round-trips a real PTY.
- name: Boot orcad and round-trip a terminal
run: pnpm run smoke:orcad-terminal
- name: Verify bundled skill guides
run: pnpm run verify:bundled-skill-guides
- name: Verify skill freshness manifest
run: pnpm run verify:skill-bundle-manifest
- name: Verify localization catalog
run: pnpm run verify:localization-catalog
# Why: the renderer ships only the English entries i18next cannot rebuild
# from each call site's inline default, so the generated subset has to
# track en.json and those defaults.
- name: Verify runtime-required localization catalog
run: pnpm run verify:localization-runtime-catalog
# Why: extraction writes sorted evidence to an isolated temporary path,
# so feature PRs need one normalized AST pass rather than a three-OS matrix.
- name: Verify localization extraction
run: pnpm run verify:localization-extraction
- name: Verify localization coverage
run: pnpm run verify:localization-coverage
# Why: project-owned type declarations must live in .ts so tsc
# actually checks them. TypeScript's skipLibCheck: true (inherited
# from @electron-toolkit/tsconfig) silently widens unresolved names
# in .d.ts to `any`, which is how #1186 shipped a broken IPC signature
# past typecheck. See .github/CONTRIBUTING.md#type-declarations-prefer-ts-over-dts.
- name: Guard against project-owned .d.ts in preload/shared
run: |
matches=$(find src/preload src/shared -name '*.d.ts' 2>/dev/null || true)
if [ -n "$matches" ]; then
echo "::error::Project-owned .d.ts files are not allowed under src/preload or src/shared."
echo "Move type declarations into a .ts file so skipLibCheck does not hide errors."
echo "See .github/CONTRIBUTING.md#type-declarations-prefer-ts-over-dts."
echo "Found:"
echo "$matches"
exit 1
fi
- name: Check feature wall asset budget
run: pnpm check:feature-wall-assets
- name: Verify macOS entitlements
run: pnpm verify:macos-entitlements
root_directory_guard:
name: root directory guard
runs-on: ubuntu-latest
steps:
- name: Checkout
uses: actions/checkout@v6
with:
# Why blob:none: full history is needed for the merge-base diff, but historical
# file contents are not. Blobs are ~89% of this repo's pack, and Git fetches the
# few this job actually reads on demand.
fetch-depth: 0
filter: blob:none
persist-credentials: false
- name: Reject new root-level files and folders
env:
BASE_SHA: ${{ github.event.pull_request.base.sha }}
HEAD_SHA: ${{ github.event.pull_request.head.sha }}
run: node .github/scripts/check-root-directory-entries.mjs "$BASE_SHA" "$HEAD_SHA"
typecheck:
needs: [code_paths]
if: needs.code_paths.outputs.typecheck == 'true'
runs-on: ubuntu-latest
steps:
- name: Checkout
uses: actions/checkout@v6
with:
persist-credentials: false
- uses: ./.github/actions/install-node-dependencies
# Why: every project is `composite`, so tsc already writes a .tsbuildinfo that lets
# the next run skip unchanged files. Share one cache entry across commits while the
# PR base stays stable; actions/cache keeps the first successful graph and the
# compiler still invalidates stale files from its content hashes.
- name: Cache TypeScript incremental state
uses: actions/cache@v5
with:
path: config/*.tsbuildinfo
key: tsbuildinfo-${{ runner.os }}-${{ hashFiles('pnpm-lock.yaml', 'config/tsconfig*.json') }}-${{ github.event.pull_request.base.sha }}
restore-keys: |
tsbuildinfo-${{ runner.os }}-${{ hashFiles('pnpm-lock.yaml', 'config/tsconfig*.json') }}-
- run: pnpm run typecheck
git_compatibility:
name: Git compatibility
needs: [code_paths]
if: needs.code_paths.outputs.git_compatibility == 'true'
runs-on: ubuntu-latest
steps:
- name: Checkout
uses: actions/checkout@v6
with:
persist-credentials: false
- uses: ./.github/actions/install-node-dependencies
# Why: the 2.25.5 lane is a source build of a pinned tarball, so it produced the
# same binary on every PR for minutes of runner time. The key carries the version
# because that is the only input; the sha256 assertion below still guards the
# tarball on the miss path that actually builds.
- name: Cache baseline Git build
uses: actions/cache@v5
with:
path: ~/.cache/orca-git-compat/git-2.25.5
key: git-compat-baseline-${{ runner.os }}-${{ runner.arch }}-2.25.5
- name: Verify Git binary compatibility matrix
run: |
pids=()
(
archive="$RUNNER_TEMP/git-2.25.5.tar.gz"
source="$HOME/.cache/orca-git-compat/git-2.25.5"
if [ ! -x "$source/git" ]; then
curl -fsSL https://www.kernel.org/pub/software/scm/git/git-2.25.5.tar.gz -o "$archive"
echo "41662c52fc16fec4963bfc41075e71f8ead6b5e386797eb6f9a1111ff95a8ddf $archive" \
| sha256sum --check
mkdir -p "$source"
tar -xzf "$archive" -C "$source" --strip-components=1
make -C "$source" -j"$(nproc)" \
NO_GETTEXT=YesPlease NO_TCLTK=YesPlease NO_PYTHON=YesPlease git
# Why: the linked binaries are what the next run needs; the objects that
# produced them are most of the tree and would bloat the cache entry.
find "$source" -name '*.o' -delete
fi
ORCA_GIT_COMPAT_BINARY="$source/git" ORCA_GIT_COMPAT_VERSION="2.25.5" \
pnpm exec vitest run --config config/vitest.config.ts \
src/shared/git-binary-compatibility.test.ts
) &
pids+=("$!")
for spec in \
"alpine/git:edge-2.38.1|2.38.1" \
"alpine/git:v2.49.1|2.49.1"; do
(
image="${spec%%|*}"
version="${spec#*|}"
ORCA_GIT_COMPAT_IMAGE="$image" ORCA_GIT_COMPAT_VERSION="$version" \
pnpm exec vitest run --config config/vitest.config.ts \
src/shared/git-binary-compatibility.test.ts
) &
pids+=("$!")
done
status=0
for pid in "${pids[@]}"; do
wait "$pid" || status=1
done
exit "$status"
# Why this job: Orca's session index-heal depends on a Codex behavior — a
# `thread/read` of an unindexed rollout performs a read-repair that inserts the
# `threads` row. Every unit test drives a stub app-server and asserts only that the
# call did not error, so if Codex dropped the repair they would all stay green while
# the subsystem went inert. This runs the pinned real binary and fails when the
# repair stops happening. Pinned because the binary is the thing expected to drift.
codex_index_heal_contract:
name: Codex index-heal contract
needs: [code_paths]
if: needs.code_paths.outputs.codex_index_heal_contract == 'true'
runs-on: ubuntu-latest
env:
CODEX_CLI_VERSION: '0.150.1'
steps:
- name: Checkout
uses: actions/checkout@v6
with:
persist-credentials: false
- uses: ./.github/actions/install-node-dependencies
- name: Install pinned Codex CLI
run: |
set -euo pipefail
npm install --no-audit --no-fund --prefix "$RUNNER_TEMP/codex-cli" \
"@openai/codex@$CODEX_CLI_VERSION"
- name: Verify Codex index-heal contract
env:
# Why REQUIRED: without a binary the suite skips, and a job that skips
# reports success. This turns a failed or missing install into a red test
# instead of a green no-op.
ORCA_CODEX_CONTRACT_REQUIRED: '1'
ORCA_CODEX_CONTRACT_VERSION: ${{ env.CODEX_CLI_VERSION }}
run: |
set -euo pipefail
ORCA_CODEX_CONTRACT_BINARY="$RUNNER_TEMP/codex-cli/node_modules/.bin/codex" \
pnpm exec vitest run --config config/vitest.config.ts \
src/main/codex/codex-index-heal-binary-contract.test.ts
xterm_patch_sync:
name: xterm patch sync
needs: [code_paths]
if: needs.code_paths.outputs.xterm_patch_sync == 'true'
runs-on: ubuntu-latest
steps:
- name: Checkout
uses: actions/checkout@v6
with:
persist-credentials: false
- uses: ./.github/actions/install-node-dependencies
# Why: the check rebuilds every package in the manifest from a pinned upstream
# commit — @xterm/xterm and its three addons, each built twice (once unmodified to
# prove the toolchain still reproduces the published bundles, once patched). Caching
# the npm metadata and the shallow clone keeps the repeated cost to the builds
# themselves; the key is the manifest, so a commit, package or toolchain bump
# invalidates it.
- name: Restore upstream xterm build inputs
uses: actions/cache@v5
with:
path: |
~/.npm
${{ runner.temp }}/xterm-patch-build/upstream/.git
key: xterm-upstream-${{ hashFiles('config/patches/xterm-upstream.json') }}
- name: Verify xterm patches match the pinned upstream build
env:
WORK_DIR: ${{ runner.temp }}/xterm-patch-build
run: node config/scripts/regenerate-xterm-patches.mjs --check --work-dir="$WORK_DIR"
shell_contracts:
name: shell contracts
needs: [code_paths]
if: needs.code_paths.outputs.shell_contracts == 'true'
runs-on: ubuntu-latest
# Why: this job's cost is almost entirely package download, and a stalled mirror has
# no wall-clock bound of its own. A successful run finishes in ~4.5 minutes, so this
# is generous; it exists so a wedge fails the job instead of holding the whole run
# open for the 6h GitHub default — which also blocks `gh run rerun --failed`.
timeout-minutes: 15
env:
# Why: the suites below gate their live fish tests on the binary, which is
# right on a developer machine and wrong here — this job is a required check
# and its fish lane is the only end-to-end guard for #9993, so a skip would
# report green with nothing exercised. Turns those skips into failures.
ORCA_REQUIRE_FISH: '1'
steps:
- name: Checkout
uses: actions/checkout@v6
with:
persist-credentials: false
# Why fish: shell-ready.test.ts gates its live fish test on the binary being
# present, so without this the fish barrier is only covered by config-shape
# assertions and never actually exercised.
# Why release-4: DECSET 2031 arming lives in the fish 4.0 Rust tty_handoff, and
# fish-color-scheme-child-stdin.node-pty.test.ts (#9993) needs it. Noble ships
# 3.7, so the PPA is what makes that lane real.
- name: Install zsh and fish
run: |
# Why the update/PPA/fish steps are tolerant: a repo the runner image already
# ships can lack a Release file for this suite, and a failed add-apt-repository
# still leaves its list entry behind — either makes `apt-get update` exit
# non-zero and would red this required check over something unrelated to the
# PR. Every fish outcome is judged by the version gate below instead, so only
# the zsh install (which has no such gate) stays fatal here.
# Why retry only here: adding the PPA is the network-flaky step, and the
# version gate below is fatal, so a transient Launchpad blip would
# otherwise red a required check on PRs unrelated to shells.
# Why -n: add-apt-repository refreshes every configured repo on its own. With
# an update on each side of it this step refreshed them three times over, and
# the Azure archive mirror alone costs ~15-30s a pass. The PPA index is the
# only thing the repo list gains here, and the single update below fetches it.
# Why bound acquisition: measured on a *passing* run, this step spent 40s
# fetching 11.4 MB of index and then 2m17s fetching 8.9 MB of packages at
# 65 kB/s — it is dominated by download throughput, not by work. apt applies
# no wall-clock bound to a stalled mirror, so a slow Launchpad or archive
# host wedges the step for tens of minutes. This job is a required check, so
# a wedge holds the entire run open and blocks `gh run rerun --failed`.
# Bounded timeouts plus retries turn an unbounded hang into a fast, legible
# failure. Set in apt.conf.d rather than on each command line so the two
# invocations below stay exactly as pr-workflow-parallelism.test.mjs parses
# them. Retries are 1, not 3: a first attempt at these bounds already multiplied
# 30s x 3 retries across every index file into a ~15 minute stall on a dead
# mirror, which is worse than failing once and moving on.
sudo tee /etc/apt/apt.conf.d/99-orca-shell-contracts >/dev/null <<'APTCONF'
Acquire::http::Timeout "15";
Acquire::https::Timeout "15";
Acquire::Retries "1";
APTCONF
for attempt in 1 2 3; do
sudo add-apt-repository -y -n ppa:fish-shell/release-4 && break
echo "add-apt-repository attempt ${attempt} failed; retrying" >&2
sudo add-apt-repository -y -n -r ppa:fish-shell/release-4 || true
sleep 5
done
# Why a wall-clock bound on each command: apt's Acquire timeouts are per-connection,
# so a dead mirror costs timeout x retries x every index file. Measured: the archive
# mirror stalled with zero bytes and the step burned 14m26s before the job bound
# killed it. `timeout` is the only thing that bounds the command as a whole.
# The update is already tolerant by design (see above), so bounding it just caps
# what a dead mirror can cost before the install runs against whatever index exists.
timeout 120 sudo apt-get update || true
# Why both shells on one line: pr-workflow-parallelism.test.mjs parses only the
# first install command in this step to prove the lane really installs them.
timeout 300 sudo apt-get install -y zsh fish
# Separate from the install so the failure names the contract, not an apt error.
# ORCA_REQUIRE_FISH re-checks this at test time; this step just fails in seconds
# instead of after a full dependency install.
- name: Require fish 4+
run: |
version="$(fish --version 2>/dev/null || true)"
major="${version##*version }"
major="${major%%.*}"
case "$major" in '' | *[!0-9]*) major=0 ;; esac
echo "${version:-<fish not installed>}"
if [ "$major" -lt 4 ]; then
echo "::error::shell contracts needs fish 4+ (DECSET 2031 arming, #9993) but got '${version:-none}'. Fix the ppa:fish-shell/release-4 install rather than letting the fish lane skip." >&2
exit 1
fi
- uses: ./.github/actions/install-node-dependencies
with:
native-runtime: node
- name: Test real shell contracts
run: |
pnpm exec vitest run --config config/vitest.config.ts --maxWorkers=1 \
src/main/daemon/repro-13767-shell-ready-marker-lost-to-exec.test.ts \
src/main/daemon/shell-ready.test.ts \
src/main/daemon/node-pty-fd-leak.test.ts \
src/main/providers/local-pty-shell-ready-zsh-launch-environment.test.ts \
src/main/providers/__tests__/shell-ready-framework-example.test.ts \
src/main/pty/codex-shell-launch-preflight.test.ts \
src/main/pty/omp-shell-wrapper-alias-safety.test.ts \
src/main/pty/omp-shell-wrapper.node-pty.test.ts \
src/main/shell-startup-feature-channel.test.ts \
src/main/terminal-history-fish-session.node-pty.test.ts \
src/main/zsh-scoped-histfile.live-shell.test.ts \
src/main/zsh-startup-hook-user-config-equivalence.live-shell.test.ts \
src/main/zsh-wrapper-version-mismatch.live-shell.test.ts \
src/renderer/src/components/terminal-pane/fish-color-scheme-child-stdin.node-pty.test.ts \
src/shared/fish-query-reply-child-stdin.node-pty.test.ts \
src/shared/pty-reply-echo-shapes.node-pty.test.ts \
src/shared/startup-shell-portability.live-shell.test.ts \
src/shared/posix-command-path-lookup.test.ts
# Cache-key input changes would otherwise make every shard compile the same
# native addon concurrently. Prime the supported Node ABI before the matrix fans out.
test_native_cache:
name: prepare test native cache node 24
needs: [code_paths]
if: needs.code_paths.outputs.native_cache_changed == 'true'
runs-on: ubuntu-latest
steps:
- name: Checkout
uses: actions/checkout@v6
with:
persist-credentials: false
- uses: ./.github/actions/install-node-dependencies
with:
native-runtime: node
node-version: '24'
test:
needs: [code_paths, test_native_cache]
if: >-
always() &&
needs.code_paths.outputs.test == 'true' &&
(needs.test_native_cache.result == 'success' || needs.test_native_cache.result == 'skipped')
uses: ./.github/workflows/unit-tests.yml
with:
node_versions: '["24"]'
# Why a separate job: the test needs a real Chrome, and the sharded `test` matrix
# would pay for it on every shard to run one file in whichever shard it landed in.
orcad_browser:
name: orcad browser provider
needs: [code_paths]
if: needs.code_paths.outputs.orcad_browser == 'true'
runs-on: ubuntu-latest
steps:
- name: Checkout
uses: actions/checkout@v6
with:
persist-credentials: false
# Why no native-runtime: the provider drives the prebuilt agent-browser binary
# shipped in node_modules and never touches node-pty.
- uses: ./.github/actions/install-node-dependencies
# Why the runner's Google Chrome and not its chromium: Ubuntu 24.04 only ships an
# AppArmor userns profile for the Chrome .deb, so chromium dies with "No usable
# sandbox" and the provider passes no --no-sandbox. Why fail instead of skip: an
# unset ORCA_BROWSER_EXECUTABLE is exactly how this test went uncovered for so long.
- name: Resolve Chrome for the browser provider
run: |
set -euo pipefail
chrome="$(command -v google-chrome || command -v google-chrome-stable || true)"
if [ -z "$chrome" ]; then
echo "::error::No Google Chrome on the runner; the browser provider test would silently skip."
exit 1
fi
"$chrome" --version
echo "ORCA_BROWSER_EXECUTABLE=$chrome" >> "$GITHUB_ENV"
- name: Test external Chromium browser provider
run: |
pnpm exec vitest run --config config/vitest.config.ts \
src/main/orcad/external-chromium-browser-process.integration.test.ts
cross-version-wire:
name: cross-version wire compatibility
needs: [code_paths]
if: needs.code_paths.outputs.cross-version-wire == 'true'
runs-on: ubuntu-latest
steps:
# Why fetch-depth 0: the harness extracts the newest release tag to skew
# current code against it. The default shallow clone has no tags, which is
# why this cannot ride along in the sharded `test` job.
- name: Checkout
uses: actions/checkout@v6
with:
# Why blob:none: full history is needed for the merge-base diff, but historical
# file contents are not. Blobs are ~89% of this repo's pack, and Git fetches the
# few this job actually reads on demand.
fetch-depth: 0
filter: blob:none
persist-credentials: false
- uses: ./.github/actions/install-node-dependencies
with:
native-runtime: node
# A path filter that matches nothing exits 1 ("No test files found"), so this
# lane cannot report success while running zero tests.
- name: Old/new client and server compatibility journeys
run: >-
pnpm exec vitest run --config config/vitest.config.ts
tests/e2e/cross-version-wire/release-checkout.unit.test.ts
tests/e2e/cross-version-wire/cross-version-browser-placement.unit.test.ts
tests/e2e/cross-version-wire/cross-version-terminal-wire.unit.test.ts
tests/e2e/cross-version-wire/reported-lossy-initial-snapshot.unit.test.ts
tests/e2e/cross-version-wire/cross-version-agent-session-wire.unit.test.ts
managed_hook_node18:
name: managed hooks on Node 18
needs: [code_paths]
if: needs.code_paths.outputs.managed_hook_node18 == 'true'
runs-on: ubuntu-latest
steps:
- name: Checkout
uses: actions/checkout@v6
with:
persist-credentials: false
- uses: ./.github/actions/install-node-dependencies
- name: Build relay companions
run: pnpm run build:relay
- name: Setup Node 18 runtime
uses: actions/setup-node@v6
with:
node-version: '18'
- name: Smoke managed-hook companions
run: node config/scripts/smoke-managed-hook-runtime-node18.mjs
package:
name: package
needs: [code_paths]
if: needs.code_paths.outputs.package == 'true'
runs-on: ubuntu-latest
# Let the serial Docker gates reach their own deadlines and report cleanup failures.
timeout-minutes: 90
steps:
- name: Checkout
uses: actions/checkout@v6
with:
persist-credentials: false
- name: Cache electron-builder downloads
uses: actions/cache@v5
with:
path: ~/.cache/electron-builder
key: electron-builder-linux-${{ hashFiles('pnpm-lock.yaml') }}
restore-keys: |
electron-builder-linux-
- uses: ./.github/actions/install-node-dependencies
with:
native-runtime: electron
# Why --no-file-parallelism: every file here launches a full Electron stack twice, and each
# probe carries its own in-process deadline. Four at once on a 4-vCPU runner starve each other
# past those deadlines; serial, every probe owns the runner.
- name: Test Linux Electron lifecycle boundary
run: >-
xvfb-run --auto-servernum pnpm exec vitest run --config config/vitest.config.ts
--no-file-parallelism
src/main/browser/browser-client-page-renderer-lifecycle.electron.test.ts
src/main/browser/browser-route-tcp-egress.electron.test.ts
src/main/browser/browser-route-webrtc-egress.electron.test.ts
src/main/browser/browser-route-h3-egress.electron.test.ts
src/main/browser/browser-route-dns-prefetch.electron.test.ts
- name: Build package inputs
run: |
status=0
pnpm run build:cli || status=1
scripts=(build:relay build:electron-vite:parallel)
pids=()
for script in "${scripts[@]}"; do
pnpm run "$script" &
pids+=("$!")
done
for pid in "${pids[@]}"; do
wait "$pid" || status=1
done
exit "$status"
- name: Project web client from renderer build
run: pnpm run build:web-from-renderer
- name: Build native components
run: pnpm run build:native
- name: Install Linux package tooling
run: sudo apt-get update && sudo apt-get install -y cpio rpm
- name: Package unpacked app
env:
ORCA_REUSE_PREPARED_NATIVE_RUNTIME: '1'
run: pnpm exec electron-builder --config config/electron-builder.config.cjs --linux AppImage deb rpm --x64 --publish never
- name: Verify root-package marker payloads
run: |
set -euo pipefail
version="$(node -p "require('./package.json').version")"
deb="dist/orca-ide_${version}_amd64.deb"
rpm="dist/orca-ide-${version}.x86_64.rpm"
test -s "$deb"
test -s "$rpm"
deb_marker="$(dpkg-deb --fsys-tarfile "$deb" | tar -xOf - ./opt/Orca/resources/package-type)"
rpm_marker="$(rpm2cpio "$rpm" | cpio --quiet --extract --to-stdout ./opt/Orca/resources/package-type)"
[[ "$deb_marker" == deb ]] || { echo "Expected deb marker, got: $deb_marker"; exit 1; }
[[ "$rpm_marker" == rpm ]] || { echo "Expected rpm marker, got: $rpm_marker"; exit 1; }
- name: Verify headless serve signal shutdown
run: node config/scripts/run-headless-serve-shutdown-docker.mjs --appimage dist/orca-linux.AppImage
- name: Verify extracted launcher serve signal shutdown
run: >-
node config/scripts/run-headless-serve-shutdown-docker.mjs
--appimage dist/orca-linux.AppImage --entrypoint launcher
- name: Verify AppImage CLI registration and serve signal shutdown
run: >-
node config/scripts/run-headless-serve-shutdown-docker.mjs
--appimage dist/orca-linux.AppImage --entrypoint appimage
--signal-target serving-electron --int-delivery pid
# A default container reproduces the hostile AppImage launch environment.
- name: Verify Linux CLI launch contract
run: node config/scripts/run-linux-cli-launch-contract-docker.mjs --appimage dist/orca-linux.AppImage
- name: Smoke packaged CLI
run: node config/scripts/smoke-packaged-cli.mjs --app-dir=dist/linux-unpacked
- name: Smoke packaged hang watchdog worker
run: xvfb-run --auto-servernum node config/scripts/smoke-packaged-hang-watchdog-worker.mjs --app-dir=dist/linux-unpacked
package_windows:
name: package (windows)
needs: [code_paths]
if: needs.code_paths.outputs.package_windows == 'true'
runs-on: windows-2022
timeout-minutes: 30
steps:
- name: Checkout
uses: actions/checkout@v6
with:
persist-credentials: false
- name: Cache electron-builder downloads
uses: actions/cache@v5
with:
path: |
~\AppData\Local\electron\Cache
~\AppData\Local\electron-builder\Cache
key: electron-builder-windows-${{ hashFiles('pnpm-lock.yaml') }}
restore-keys: |
electron-builder-windows-
# Why persist-native-cache false: this job later rebuilds the same path for
# Electron. A post-job save would store the Electron ABI under the Node key.
- uses: ./.github/actions/install-node-dependencies
id: deps
with:
native-runtime: node
persist-native-cache: 'false'
- name: Save compiled Node native modules
if: steps.deps.outputs.native-cache-hit != 'true'
uses: actions/cache/save@v5
with:
path: |
node_modules/.pnpm/node-pty@*/node_modules/node-pty/build
node_modules/.pnpm/windows-native-registry@*/node_modules/windows-native-registry/build
node_modules/.pnpm/@vscode+windows-process-tree@*/node_modules/@vscode/windows-process-tree/build
key: native-modules-${{ runner.os }}-${{ steps.deps.outputs.native-cache-scope }}-${{ runner.arch }}-node-node${{ steps.deps.outputs.node-version }}-${{ hashFiles('pnpm-lock.yaml', '.github/actions/install-node-dependencies/action.yml', 'config/scripts/ensure-native-runtime.mjs', 'config/scripts/rebuild-native-deps.mjs', 'config/patches/node-pty@1.1.0.patch', 'config/patches/@vscode__windows-process-tree@0.8.0.patch') }}
- name: Test Windows-specific boundaries
run: >-
pnpm exec vitest run --config config/vitest.config.ts
config/scripts/rebuild-native-deps.test.mjs
src/main/browser/browser-client-page-renderer-lifecycle.electron.test.ts
src/main/browser/browser-route-tcp-egress.electron.test.ts
src/main/browser/browser-route-webrtc-egress.electron.test.ts
src/main/browser/browser-route-h3-egress.electron.test.ts
src/main/browser/browser-route-dns-prefetch.electron.test.ts
src/main/providers/windows-conpty-wide-char-duplication.node-pty.test.ts
src/main/providers/pty-repaint-wide-char-buffer.node-pty.test.ts
src/shared/child-process/windows-command-line.win32.test.ts
src/main/agent-hooks/windows-hook-payload-delivery.test.ts
src/main/windows/windows-pty-job.win32.test.ts
src/main/windows/windows-host-job.win32.test.ts
src/main/wsl/wsl-runner.test.ts
src/main/wsl/wsl-guest-environment.test.ts
src/main/wsl/wsl-invocation-boundary.test.ts
src/main/wsl/wsl-executable-path.win32.test.ts
src/main/wsl/wsl-w1-w3-contract.test.ts
src/shared/source-scan/source-tree-scan.test.ts
src/main/cli/wsl-cli-powershell-boundary.test.ts
src/main/cursor/hook-service.test.ts
src/main/orca-profiles/profile-index-store.test.ts
src/main/startup/windows-install-dir-acl-repair.win32.test.ts
src/main/runtime/repo-worktree-admin-fingerprint.test.ts
src/main/runtime/worktree-scan-admin-fingerprint-gate.test.ts
src/shared/secure-file-fsync-flags.test.ts
src/main/ipc/pty-codex-account-attribution.test.ts
src/main/ipc/pty-spawn-env-codex-resume-provenance.test.ts
# Why the :parallel variant: identical to build:release except the three
# electron-vite targets overlap instead of running back to back. The Linux package
# job already packages and smoke-tests an AppImage built that way.
- name: Cache Windows CLI launcher
uses: actions/cache@v5
with:
path: native/windows-cli-launcher/.build
key: windows-cli-launcher-${{ runner.os }}-${{ runner.arch }}-${{ hashFiles('native/windows-cli-launcher/**', 'config/scripts/build-windows-cli-launcher.mjs') }}
- name: Build package inputs
env:
ORCA_REUSE_WINDOWS_CLI_LAUNCHER: '1'
run: pnpm run build:release:parallel
- name: Restore compiled Electron native modules
uses: actions/cache@v5
with:
path: |
node_modules/.pnpm/node-pty@*/node_modules/node-pty/build
node_modules/.pnpm/windows-native-registry@*/node_modules/windows-native-registry/build
node_modules/.pnpm/@vscode+windows-process-tree@*/node_modules/@vscode/windows-process-tree/build
key: native-modules-${{ runner.os }}-${{ steps.deps.outputs.native-cache-scope }}-${{ runner.arch }}-electron-node${{ steps.deps.outputs.node-version }}-${{ hashFiles('pnpm-lock.yaml', '.github/actions/install-node-dependencies/action.yml', 'config/scripts/ensure-native-runtime.mjs', 'config/scripts/rebuild-native-deps.mjs', 'config/patches/node-pty@1.1.0.patch', 'config/patches/@vscode__windows-process-tree@0.8.0.patch') }}
- name: Prepare Electron native runtime
run: node config/scripts/ensure-native-runtime.mjs --runtime=electron
- name: Package unpacked app
env:
ORCA_REUSE_PREPARED_NATIVE_RUNTIME: '1'
run: pnpm exec electron-builder --config config/electron-builder.config.cjs --dir
- name: Smoke packaged Windows PTY native capability
run: pnpm run smoke:windows-pty-native-capability -- --exe=dist/win-unpacked/Orca.exe
- name: Smoke packaged CLI
run: node config/scripts/smoke-packaged-cli.mjs --app-dir=dist/win-unpacked
# Why: PR E2E is advisory and only validates changed specs; scheduled and
# release runs retain full-suite coverage.
e2e-paths:
name: detect changed e2e specs
needs: [code_paths]
runs-on: ubuntu-latest
if: github.event.pull_request.draft != true && needs.code_paths.outputs.should_run == 'true'
# Why: detector only needs to read the checkout; do not inherit repo defaults.
permissions:
contents: read
outputs:
should_run: ${{ steps.filter.outputs.should_run }}
test_files: ${{ steps.filter.outputs.test_files }}
ssh_source_changed: ${{ steps.filter.outputs.ssh_source_changed }}
native_ime_source_changed: ${{ steps.filter.outputs.native_ime_source_changed }}
steps:
- name: Checkout
uses: actions/checkout@v6
with:
# Why blob:none: full history is needed for the merge-base diff, but historical
# file contents are not. Blobs are ~89% of this repo's pack, and Git fetches the
# few this job actually reads on demand.
fetch-depth: 0
filter: blob:none
persist-credentials: false
- name: Filter changed E2E specs
id: filter
run: |
set -euo pipefail
BASE="${{ github.event.pull_request.base.sha }}"
HEAD="${{ github.event.pull_request.head.sha }}"
CHANGED="$(git diff --name-only --diff-filter=AMCR --merge-base "$BASE" "$HEAD")"
# Source routes are executable contracts so a test can prove exact
# authorities, exclusions, and sentinels without evaluating workflow shell.
TEST_FILES_JSON="$(printf '%s\n' "$CHANGED" | node config/scripts/pr-e2e-source-routing.mjs)"
echo "test_files=$TEST_FILES_JSON" >> "$GITHUB_OUTPUT"
# Why a separate signal: the Docker-SSH lane must trigger on SSH source, not on a
# spec name surviving in a route's list. Same routes, so the two cannot drift.
SSH_SOURCE_CHANGED="$(printf '%s\n' "$CHANGED" | node config/scripts/pr-e2e-source-routing.mjs --ssh-source)"
echo "ssh_source_changed=$SSH_SOURCE_CHANGED" >> "$GITHUB_OUTPUT"
echo "SSH source changed: $SSH_SOURCE_CHANGED"
# Why its own signal: the real-IME lane is a whole ibus session, not a spec, so it must
# trigger on IME source rather than on a spec name in some route's list.
NATIVE_IME_SOURCE_CHANGED="$(printf '%s\n' "$CHANGED" | node config/scripts/pr-e2e-source-routing.mjs --native-ime-source)"
echo "native_ime_source_changed=$NATIVE_IME_SOURCE_CHANGED" >> "$GITHUB_OUTPUT"
echo "Native IME source changed: $NATIVE_IME_SOURCE_CHANGED"
if [ "$TEST_FILES_JSON" != '[]' ]; then
echo "should_run=true" >> "$GITHUB_OUTPUT"
echo "Changed E2E specs: $TEST_FILES_JSON"
else
echo "should_run=false" >> "$GITHUB_OUTPUT"
echo "No changed E2E specs"
fi
e2e:
name: e2e
needs: e2e-paths
if: needs.e2e-paths.outputs.should_run == 'true'
# Why: reusable e2e.yml only checkouts, builds, and uploads artifacts.
permissions:
contents: read
uses: ./.github/workflows/e2e.yml
with:
# The synthetic pull-request merge ref can disappear while this reusable
# workflow is queued. The head SHA is immutable and works for every PR.
ref: ${{ github.event.pull_request.head.sha }}
test_files: ${{ needs.e2e-paths.outputs.test_files }}
ssh_source_changed: ${{ needs.e2e-paths.outputs.ssh_source_changed }}
# Why this is not in verify's needs: it is the first PR-gate run of a harness whose reliability
# is only known from nightly main runs (20/20 green, 2026-08-09..2026-08-29, p50 3m25s). It
# reports a red X on the PR without blocking, exactly like `e2e` above. Deliberately no
# continue-on-error: that renders the check green and hides the signal it exists to give. To
# make it blocking, add it to verify.needs, add TERMINAL_IME_NATIVE to the env below, and
# require `success || skipped` outside the strict loop — see the note on `e2e`.
terminal_ime_native:
name: real IME
needs: e2e-paths
if: needs.e2e-paths.outputs.native_ime_source_changed == 'true'
# Why: the reusable workflow only checks out, builds, and uploads artifacts.
permissions:
contents: read
uses: ./.github/workflows/terminal-ime-e2e.yml
verify:
if: always()
needs:
- code_paths
- static_analysis
- root_directory_guard
- typecheck
- git_compatibility
- codex_index_heal_contract
- xterm_patch_sync
- shell_contracts
- test
- orcad_browser
- cross-version-wire
- managed_hook_node18
- package
- package_windows
runs-on: ubuntu-latest
steps:
# Why: e2e is deliberately absent from needs. The suite is currently red on
# main (every scheduled run), so gating merges on it would block any PR that
# touches tests/e2e/** — including the ones fixing the suite. Until it is
# green the job runs and reports for E2E-path PRs without blocking. To flip
# it on: add `e2e` to needs, add E2E to the env below, and require
# `"$E2E" = success || skipped` after the loop — skipped is the normal
# result for a path-filtered job and must keep passing, so it has to be
# checked outside the loop or it would excuse the jobs above.
- name: Require successful checks
env:
CODE_PATHS: ${{ needs.code_paths.result }}
SHOULD_RUN: ${{ needs.code_paths.outputs.should_run }}
STATIC_ANALYSIS: ${{ needs.static_analysis.result }}
STATIC_ANALYSIS_SHOULD_RUN: ${{ needs.code_paths.outputs.static_analysis }}
ROOT_DIRECTORY_GUARD: ${{ needs.root_directory_guard.result }}
TYPECHECK: ${{ needs.typecheck.result }}
TYPECHECK_SHOULD_RUN: ${{ needs.code_paths.outputs.typecheck }}
GIT_COMPATIBILITY: ${{ needs.git_compatibility.result }}
GIT_COMPATIBILITY_SHOULD_RUN: ${{ needs.code_paths.outputs.git_compatibility }}
CODEX_INDEX_HEAL_CONTRACT: ${{ needs.codex_index_heal_contract.result }}
CODEX_INDEX_HEAL_CONTRACT_SHOULD_RUN: ${{ needs.code_paths.outputs.codex_index_heal_contract }}
XTERM_PATCH_SYNC: ${{ needs.xterm_patch_sync.result }}
XTERM_PATCH_SYNC_SHOULD_RUN: ${{ needs.code_paths.outputs.xterm_patch_sync }}
SHELL_CONTRACTS: ${{ needs.shell_contracts.result }}
SHELL_CONTRACTS_SHOULD_RUN: ${{ needs.code_paths.outputs.shell_contracts }}
TEST: ${{ needs.test.result }}
TEST_SHOULD_RUN: ${{ needs.code_paths.outputs.test }}
ORCAD_BROWSER: ${{ needs.orcad_browser.result }}
ORCAD_BROWSER_SHOULD_RUN: ${{ needs.code_paths.outputs.orcad_browser }}
CROSS_VERSION_WIRE: ${{ needs.cross-version-wire.result }}
CROSS_VERSION_WIRE_SHOULD_RUN: ${{ needs.code_paths.outputs.cross-version-wire }}
MANAGED_HOOK_NODE18: ${{ needs.managed_hook_node18.result }}
MANAGED_HOOK_NODE18_SHOULD_RUN: ${{ needs.code_paths.outputs.managed_hook_node18 }}
PACKAGE: ${{ needs.package.result }}
PACKAGE_SHOULD_RUN: ${{ needs.code_paths.outputs.package }}
PACKAGE_WINDOWS: ${{ needs.package_windows.result }}
PACKAGE_WINDOWS_SHOULD_RUN: ${{ needs.code_paths.outputs.package_windows }}
run: |
if [ "$CODE_PATHS" != "success" ]; then
exit 1
fi
if [ "$ROOT_DIRECTORY_GUARD" != "success" ]; then
exit 1
fi
if [ "$SHOULD_RUN" != "true" ]; then
echo "Docs-only change; expensive PR checks skipped."
fi
failed=0
check_job() {
local name="$1" result="$2" should="$3"
if [ "$should" = "true" ]; then
if [ "$result" != "success" ]; then
echo "$name: expected success, got $result"
failed=1
fi
else
if [ "$result" != "skipped" ]; then
echo "$name: expected skipped, got $result"
failed=1
fi
fi
}
# Require success when the PR has code-relevant changes
check_job static_analysis "$STATIC_ANALYSIS" "$STATIC_ANALYSIS_SHOULD_RUN"
check_job typecheck "$TYPECHECK" "$TYPECHECK_SHOULD_RUN"
check_job git_compatibility "$GIT_COMPATIBILITY" "$GIT_COMPATIBILITY_SHOULD_RUN"
check_job codex_index_heal_contract "$CODEX_INDEX_HEAL_CONTRACT" "$CODEX_INDEX_HEAL_CONTRACT_SHOULD_RUN"
check_job xterm_patch_sync "$XTERM_PATCH_SYNC" "$XTERM_PATCH_SYNC_SHOULD_RUN"
check_job shell_contracts "$SHELL_CONTRACTS" "$SHELL_CONTRACTS_SHOULD_RUN"
check_job test "$TEST" "$TEST_SHOULD_RUN"
check_job orcad_browser "$ORCAD_BROWSER" "$ORCAD_BROWSER_SHOULD_RUN"
check_job cross-version-wire "$CROSS_VERSION_WIRE" "$CROSS_VERSION_WIRE_SHOULD_RUN"
check_job managed_hook_node18 "$MANAGED_HOOK_NODE18" "$MANAGED_HOOK_NODE18_SHOULD_RUN"
check_job package "$PACKAGE" "$PACKAGE_SHOULD_RUN"
check_job package_windows "$PACKAGE_WINDOWS" "$PACKAGE_WINDOWS_SHOULD_RUN"
exit "$failed"