Files
orca/.github/workflows/adhoc-mac-build.yml
T
Jinwoo HongandNeil 3f8d296b8d fix(macos): preserve terminal permissions across app updates with a small signed helper (#27021)
* fix(macos): preserve terminal privacy identity across app updates

* test(macos): make launchd fixtures portable across host platforms

* fix(macos): probe protected contents and close handles on denial

* fix(i18n): register terminal permission status labels

* fix(macos): fork the daemon when no stable-bundle job can claim the endpoint

A copy, signature, plist or deadline failure before bootstrap, a bootstrap
launchd refused with the job absent, a job whose daemon exited before
answering, and a foreign endpoint all threw out of the launcher. The app then
lost its persistent daemon and fell back to local PTYs, where main forks one.
These now fall through to the existing fork path, which also adopts a foreign
occupant normally instead of as a degraded endpoint. A timed-out bootstrap or
a job still running/unreadable stays fatal, so no second daemon races it.

* feat(telemetry): split daemon origin events by stable-copy launch method

daemon_adopted, daemon_pty_cwd_denied and daemon_pty_cwd_readable gain a
closed-enum launch_method (stable-copy / app-fork / unknown), read from the
spawner path the daemon recorded, so the update fix is measurable per daemon
even when an old forked daemon is adopted across updates. The reliability
gate is a test manifest with no runtime switch, so there is no gate state to
report; spawner_path_class reports the private copy as 'other' alongside
other non-/Applications installs.

* refactor(telemetry): report the stable copy as a spawner path class

The separate launch_method property duplicated what the recorded spawner
path already says. spawner_path_class gains `stable-copy`, classified from
the same path under the private runtime root, and launch_method, its enum
and classifier are removed from all three .strict() daemon origin events.
A deleted copy still reports `missing`.

* refactor(settings): gate terminal-host Full Disk Access on macOS once

The permissions handler already asks for the terminal-host status only on
macOS, so the status function's own platform check and the pane's
'unsupported' branch could never fire. The off-macOS guarantee is now
tested at the handler.

* fix(macos): keep the fork fallback inside the startup PTY gate

The stable launch could run until 55 s after the launch started, so a late
"unavailable" verdict left the fork fallback's own 10 s readiness wait to
finish after the 60 s gate had already moved the app to local PTYs.

The launcher now gives the stable attempt one handoff deadline: the gate
minus the fork's readiness timeout and the 5 s lease/adapter allowance
(45 s). Copy, signature check and plist conversion share one abort signal
that ends early enough for bootstrap and the job's own readiness wait to
meet that deadline; every command otherwise keeps a fixed timeout. An
unavailable verdict past the deadline is reported instead of forking a
daemon that cannot arrive in time. The per-step budget module is gone (its
deletion landed in the previous commit by a staging slip).

The launch body is now two blocks: preparation (plist, plutil, deadline
check) fails as unavailable after removing the copy and job file, and
bootstrap runs in its own try/finally that always deletes the job file.

Retirement is simpler and claims more orphans: one in-flight collection per
process instead of a per-root map and rotating 20-record scan, and the job
record (written before copying) no longer carries a bundle name because
lsof +D scans the whole runtime directory (measured equal cost, same
result on a live copy). A never-submitted copy is retired once its producer
has exited and its job is missing, so a bootstrap timeout or a crash
mid-copy no longer leaves a copy forever; the open-file check still guards
deletion. Collection now starts from the launcher after the attempt.

* docs(macos): ship the daemon launch rules, not the investigation

The signed investigation harness left the tree in the previous telemetry
commit by a staging slip; its results and pinned link stay in the PR. This
trims the reference doc to why the copy and launchd exist, the fallback
states and the cleanup rule, trims the reliability gate to what its
validator needs, and points AGENTS.md at the reference doc next to the
Windows relocation entry.

* fix(macos): hand off to the fork as soon as the stable job exits

A copied daemon refused at exec (for example by Gatekeeper) never answers,
so every fresh start paid the full 10 s readiness wait before the fork
fallback. The readiness loop now reads the job state every 500 ms and
abandons as soon as the job is stopped or missing; a running or
unreadable job is still waited out and never forked beside.

Also drops the dead forkEntryPath the stable launch never read (its
options type now omits it) and un-exports the handoff constant.

* fix(macos): keep daemon copies out of Spotlight and unregister retired ones

Each runtime copy is a full Orca.app, so Spotlight indexed it as an app
and LaunchServices kept a record for it that outlived the copy. A
registered copy claims orca: links and Markdown/CSV files, and becomes
their default handler whenever /Applications/Orca.app is not registered.

New copies go under runtime-*/app.noindex/, which Spotlight skips and which
stops the record LaunchServices made while the bundle was being copied.
Retirement now runs lsregister -u on every bundle in the copy (both the
app.noindex layout and the earlier runtime-*/Orca.app layout) once lsof
shows it unused, then deletes it. A running copy stays registered,
because Local Network access may be resolved through that record.

Collection and stable-copy telemetry match the runtime-* directory, so
copies made before this change are still adopted, classified and retired.

* docs(macos): record the LaunchServices trade-off and the rollback plan

The reference doc now explains why copies sit in a .noindex folder, why
only retired copies are unregistered, and how to turn the copy launch off:
make launchMacDaemonFromStableBundle return null, and keep the retirement
code until no copy-launched daemon can remain, because the old code never
deletes copies.

A test pins that rollback: with the stable launcher returning null, the
launcher forks and still runs the copy collection.

* fix(macos): keep daemon copies registered so Local Network can resolve them

The `.noindex` folder kept copies out of Spotlight and LaunchServices, but
Local Network access is resolved from the signing identifier through
LaunchServices to the executable's Mach-O UUID. After an update that changes
Electron only the copy carries the running daemon's UUID, so an unregistered
copy risks the same Local Network loss main has today. Copies go back to
`runtime-*/Orca.app`; retirement still unregisters each copy before deleting
it, so deleted copies leave no LaunchServices entry. The remaining trade-off:
while `/Applications/Orca.app` is missing, a running registered copy can be
picked for `orca:` links and documents.

* fix(macos): hide daemon copies from Spotlight and tie the job to Orca

Spotlight listed each runtime copy as a second Orca app. New copies go
back under runtime-*/app.noindex/Orca.app, which Spotlight skips.
Retirement and stable-copy telemetry match the runtime-* directory, so
copies in the earlier runtime-*/Orca.app layout still retire and classify
the same way.

Local Network access identifies the daemon by its code signature and
main-executable UUID, not by a LaunchServices record, so an unregistered
copy keeps it. The launchd job now sets AssociatedBundleIdentifiers to
com.stablyai.orca, which is how a job not installed through SMAppService
names the app it belongs to. Copies are never registered explicitly;
retirement still runs lsregister -u before deleting each one.

* feat(macos): package a signed terminal host helper that runs the daemon on the pinned Node

Contents/Helpers/Orca Terminal Host.app carries the pinned Node, the daemon's
out/main JS and this slice's node-pty, signed inside-out as com.stablyai.orca.
afterPack boots it through one PTY round trip, then fails the build on layout,
Info.plist, macOS floor (NODE_RUNTIME_DARWIN_MINIMUM_OS vs Node's minos),
architecture or signature drift.

* ci(macos): cache and prefetch the pinned Node before packaging

Packaging now fetches the darwin Node from nodejs.org, whose download has no
retry. Each mac workflow caches it with the electron-builder downloads, keyed on
the pin, and warms both darwin targets inside the existing retry wrapper.

* feat(macos): launch the daemon from a copy of the terminal host helper, not the whole app

The stable launch now clones Contents/Helpers/Orca Terminal Host.app (~124 MiB
instead of ~628 MiB) and runs its Node on its daemon entry, without
ELECTRON_RUN_AS_NODE. The copy must still match the running app's designated
requirement. A missing helper falls back to the fork, and macOS below the
helper's Node floor forks without copying.

* docs(macos): describe the terminal host helper, its floor and its identity

* fix(macos): bound the terminal host boot check by its deadline and always clean up

A daemon that stayed silent hung packaging past the 20 s deadline; one that closed the
connection left the hello unsettled so Node exited 0 without cleanup; a malformed line threw
uncaught. Every connect, hello and PTY wait now races the deadline, close/error/parse failures
reject, and the daemon and scratch dir are removed on every path.

* ci(macos): cache the pinned Node on its own key with no restore-keys fallback

Sharing the electron-builder cache let old Node archives accumulate through restore-keys.
Also moves release-mac-build's Electron cache comment back onto its step.

* fix(macos): reject a universal slice for the terminal host and pin its name in one test

packageMacTerminalHost turned an unknown arch into darwin-undefined; it now throws through the
shared mac arch mapping. A test asserts the packager and the desktop agree on the helper bundle
and executable names, replacing a Linux/Windows test that could not fail.

* fix(macos): pass the launchd adoption lease its options object after the main merge

* test(macos): accept a reset as the daemon closing the boot-check connection; keep only passing gate commands

* fix(daemon): bump the terminal daemon protocol to v45 so new macOS tabs start in the terminal host helper

* fix(macos): ship the daemon's required packages in the terminal host and boot it outside the build tree

* style(macos): use a template literal in the terminal host resolution check

* fix(macos): derive the terminal host's packages from the daemon's own module graph

* fix(macos): treat any non-literal daemon load as computed, follow directory loads, and keep nested dependencies

* refactor(macos): move terminal-host telemetry, Settings status and Helper folder strings to a follow-up

Keeps this change to the update-safe terminal host itself. The follow-up restores:
- the `stable-copy` spawner path class in daemon adoption telemetry;
- the per-terminal-host Full Disk Access row in Settings and its daemon RPC;
- writing folder usage descriptions into the app's Electron Helpers for the fork fallback.

The helper still takes its Desktop, Documents and Downloads descriptions from
`macos-folder-usage-descriptions.cjs`, so that constant stays.

* fix(i18n): restore en.json to this branch's base, not newer main

The previous commit took the file from a newer main, which pulled unrelated strings into this PR.

* refactor(macos): publish one terminal-host copy per helper build under a fixed launchd label

Name the copy after the helper's CDHash, reuse it when present, and
publish a new one by cloning into a staging dir, verifying it once
against the app's designated requirement and renaming it into place.
One launchd label per profile and protocol replaces per-launch labels,
so job.json, producer pids and per-label collection go away. Retirement
deletes any other copy lsof shows unused. The stable launcher returns
null for every fork fallback and logs its reason.

* fix(packaging): copy terminal-host dependencies from where Node resolves them

A nested package's own dependencies were never walked, so a hoisted
dependency of a nested package was left out of the helper. Resolve each
dependency through the node_modules lookup Node performs from its parent
and copy it to the same relative place, unless a copied package already
carries it.

* fix(packaging): fail release and CI mac builds that cannot boot a terminal-host slice

An arm64 runner without Rosetta skipped the x64 helper's PTY boot check
and shipped the slice untested. Release channels and CI now fail there;
local cross-arch builds still skip with a log line.

* docs(macos): describe the per-build terminal-host copy and its fallback reasons

* docs(macos): keep the codesign command on one line

* test(macos): type the launchd test state without an assertion

* docs(macos): name what the login-session flag gates in the stable launcher

* fix(macos): wait for launchd to drop a booted-out terminal job before reusing its label

---------

Co-authored-by: Neil <neil@stably.ai>
2026-10-11 14:54:36 -04:00

624 lines
31 KiB
YAML

name: Adhoc macOS + Windows Dev Build
# Why: lets anyone cut installable macOS and Windows builds of a trusted repo ref
# so the team can run an experimental feature instead of reasoning from a diff.
# Hourly covers current main; this covers experiments and on-demand validation.
#
# Deliberately narrow scope:
# - macOS and Windows desktop installers. Linux keeps using RC/stable.
# - No tests, no lint, no e2e. PR CI and release-cut remain the gates. The one
# exception is the orcad template: like release-cut, it is merged from the
# node-server lanes that build and qualify each SSH target's slot.
# - macOS is signed and notarized so TCC grants survive updates.
# - Windows is unsigned; the published release notes explain the one-time
# SmartScreen/manual-install requirement.
#
# Artifacts publish to stablyai/orca-adhoc — separate from both orca and
# orca-hourly. Separate from orca because the main repo's releases atom feed
# exposes only its 10 newest entries. Separate from orca-hourly because a build
# from someone's branch must never be picked up by a developer who only meant to
# ride main; the two are different levels of "unvetted".
#
# From the Actions tab: pick this workflow, "Run workflow", leave "Use workflow
# from" on main, and type your branch in the first field. Or from the CLI:
#
# gh workflow run adhoc-mac-build.yml --ref main -f ref=my-branch -f label=wasm-terminal
#
# Leaving the ref field empty builds whatever "Use workflow from" is set to, which
# is what someone who only touched that picker means. Naming the branch explicitly
# is still better: the workflow file is always read from the dispatch ref, so a
# branch carrying a stale copy of this file would otherwise run that copy.
#
# GITHUB_TOKEN is scoped to this repo and cannot publish there, so writes use the
# same GitHub App as hourly, additionally installed on orca-adhoc with
# Contents: Read and write. The secret names below are historical — one App, one
# private key, both dev-channel repos — and rotating it stays a single operation.
# Provision with `bash config/scripts/setup-hourly-release-token.sh`.
#
# The requested ref is vetted before checkout: it must be a branch or tag of this
# repo, or a commit reachable from one. PR refs are refused outright — this
# workflow runs the checked-out code next to MAC_CERTS and the notary password,
# so "just build that community PR" must not become a way to hand fork code the
# release identity. A branch here always belongs to someone with write access,
# which is the same trust the dispatch button itself already requires.
on:
workflow_dispatch:
inputs:
ref:
# Why optional: the Actions UI already shows its own "Use workflow from"
# branch picker directly above this field, and picking a branch there is
# what most people will read as "build this". Defaulting to that branch
# makes the obvious action correct. Fill this in only to build a ref other
# than the one the workflow file itself is read from — normally leave the
# picker on main and name your branch here, so a stale copy of this
# workflow on an old branch is not what runs.
description: 'Branch, tag, or SHA to build — must live in stablyai/orca; PR refs are refused (default: the branch selected above)'
required: false
default: ''
type: string
label:
description: 'Short name shown in the release title (default: the ref)'
required: false
default: ''
type: string
permissions:
contents: read
concurrency:
# Why keyed on the ref rather than global: two people cutting builds from two
# different branches at the same time is the ordinary case here, and serialising
# them would make each wait out the other's notary queue. Re-dispatching the
# *same* branch still queues, so a push mid-build cannot race itself.
group: adhoc-mac-build-${{ inputs.ref || github.ref_name }}
cancel-in-progress: false
env:
ADHOC_REPO: stablyai/orca-adhoc
# Why age and not a count like hourly: this channel is low-volume and bursty, so
# a count would either hold one week's experiments forever or evict a build
# someone is still running after a busy afternoon. A month is well past the "few
# days" these exist for, and by then the branch has landed or been abandoned.
ADHOC_RETAIN_DAYS: 30
jobs:
# Why: every job below checks out its own copy, and a branch name read per job builds whatever
# the branch points at when that job starts. A push mid-run then mixes commits, e.g. slot lanes
# from one commit merged by a template step from the next. Resolving once pins the whole run.
# The mac job still vets this commit's reachability before anything is signed.
resolve-ref:
if: github.repository == 'stablyai/orca'
runs-on: ubuntu-latest
timeout-minutes: 5
permissions:
contents: read
outputs:
sha: ${{ steps.resolve.outputs.sha }}
steps:
- name: Resolve the requested ref to one commit
id: resolve
shell: bash
env:
REQUESTED_REF: ${{ inputs.ref || github.ref_name }}
REPO_URL: https://github.com/${{ github.repository }}
run: |
set -euo pipefail
case "$REQUESTED_REF" in
refs/pull/*|pull/*)
echo "::error::Refusing to build PR ref '$REQUESTED_REF'; push the code to a branch of stablyai/orca instead."
exit 1
;;
esac
if [[ "$REQUESTED_REF" =~ ^[0-9a-f]{40}$ ]]; then
sha="$REQUESTED_REF"
else
# Branch first, as the mac job's vet does; a peeled annotated tag names its commit.
refs="$(git ls-remote "$REPO_URL" "refs/heads/$REQUESTED_REF" "refs/tags/$REQUESTED_REF" "refs/tags/$REQUESTED_REF^{}")"
sha=""
for name in "refs/heads/$REQUESTED_REF" "refs/tags/$REQUESTED_REF^{}" "refs/tags/$REQUESTED_REF"; do
sha="$(awk -v ref="$name" '$2 == ref { print $1; exit }' <<<"$refs")"
[[ -n "$sha" ]] && break
done
fi
if [[ -z "$sha" ]]; then
echo "::error::'$REQUESTED_REF' is not a branch, tag, or full commit SHA of stablyai/orca."
exit 1
fi
echo "Resolved $REQUESTED_REF -> $sha"
echo "sha=$sha" >>"$GITHUB_OUTPUT"
# Why: a branch cut before the orcad template landed has none to build or ship.
orcad-template-support:
needs: resolve-ref
runs-on: ubuntu-latest
timeout-minutes: 5
permissions:
contents: read
outputs:
ships: ${{ steps.detect.outputs.ships }}
steps:
- name: Checkout the template packager only
uses: actions/checkout@v6
with:
ref: ${{ needs.resolve-ref.outputs.sha }}
sparse-checkout: |
/config/scripts/packaged-orcad-template.cjs
/config/tsconfig.relay.json
sparse-checkout-cone-mode: false
persist-credentials: false
# Why: a relay-era ref (before #27215) needs the Windows relay addon this build no longer compiles.
- name: Refuse a ref that still ships the SSH relay
shell: bash
run: |
if [[ -e config/tsconfig.relay.json ]]; then
echo "::error::This ref predates #27215 and still ships the SSH relay; build a commit at or after 2b4b60a87e."
exit 1
fi
- name: Detect whether the ref ships the orcad template
id: detect
shell: bash
run: |
if [[ -f config/scripts/packaged-orcad-template.cjs ]]; then
echo "ships=true" >>"$GITHUB_OUTPUT"
else
echo "ships=false" >>"$GITHUB_OUTPUT"
echo "::notice::This ref predates the orcad template; the build ships without it."
fi
# Design D2, as release-cut does: every desktop build ships the orcad template (server JS plus
# every target's addons), merged from the node-server lanes that qualify each slot at this ref.
# Without it an adhoc build cannot deploy managed orcad to an SSH host. No secrets, and the mac
# job vets the ref before anything is signed.
orcad-template:
needs: [resolve-ref, orcad-template-support]
if: needs.orcad-template-support.outputs.ships == 'true'
permissions:
contents: read
uses: ./.github/workflows/node-server-tests.yml
with:
ref: ${{ needs.resolve-ref.outputs.sha }}
build_template: true
build-adhoc-mac:
needs: [resolve-ref, orcad-template-support, orcad-template]
# Why not the implicit success(): a ref without the orcad template skips that job on purpose.
if: >-
!cancelled() && github.repository == 'stablyai/orca' &&
needs.orcad-template-support.result == 'success' &&
(needs.orcad-template.result == 'success' ||
(needs.orcad-template.result == 'skipped' &&
needs.orcad-template-support.outputs.ships == 'false'))
# Why an environment: it gives the signing/notary/App secrets somewhere to
# live that a stale copy of this workflow on an old branch cannot reach.
# Referencing it is a no-op until repo settings give it teeth; the intended
# follow-up is to move MAC_CERTS, MAC_CERTS_PASSWORD, APPLE_ID,
# APPLE_APP_SPECIFIC_PASSWORD, APPLE_TEAM_ID, HOURLY_RELEASE_APP_ID, and
# HOURLY_RELEASE_APP_PRIVATE_KEY into it, then pin its deployment branch
# policy to main so only main's copy of this file can read them.
environment: adhoc-mac-build
outputs:
tag: ${{ steps.release.outputs.tag }}
version: ${{ steps.adhoc.outputs.version }}
head_sha: ${{ steps.adhoc.outputs.head_sha }}
published: ${{ steps.publish_live.outcome == 'success' && 'true' || 'false' }}
ships_orcad_template: ${{ needs.orcad-template-support.outputs.ships }}
runs-on: blacksmith-6vcpu-macos-15
# Why 150: it must exceed the worst case the retry budgets below can produce
# (install 3x10 + publish 2x45 = 120, plus ~25 for checkout/build/verify), or
# the job is killed mid-retry and no cleanup step runs at all.
timeout-minutes: 150
env:
NODE_OPTIONS: --max-old-space-size=4096
steps:
# Why vet before checkout: everything after this step runs the checked-out
# code with release signing credentials in reach. Branches and tags of this
# repo are the intended audience; refs/pull/* would smuggle in fork code,
# and a raw SHA is only accepted when some branch or tag of this repo can
# actually reach it. Resolving to a pinned SHA here also means the commit
# that was vetted is the commit that gets checked out — a push to the
# branch between the two steps cannot swap it.
- name: Vet the requested ref
id: vetted
shell: bash
env:
# The commit resolve-ref pinned for every job; vetting it keeps the reachability test.
REQUESTED_REF: ${{ needs.resolve-ref.outputs.sha }}
REPO_URL: https://github.com/${{ github.repository }}
run: |
set -euo pipefail
case "$REQUESTED_REF" in
refs/pull/*|pull/*)
echo "::error::Refusing to build PR ref '$REQUESTED_REF': this workflow signs with release credentials, so it only builds branches, tags, or commits of stablyai/orca. Push the code to a branch of this repo instead."
exit 1
;;
esac
# Bare: a work-tree repo refuses to fetch over its own checked-out
# branch. tree:0 keeps the fetch to the commit graph — no trees, no
# blobs — so this stays cheap next to the build it fronts. reftable
# because this repo has branches that differ only in casing, and the
# files backend cannot store both on a case-insensitive runner disk —
# it fails the entire fetch, not just the one ref.
scratch="$RUNNER_TEMP/vet-requested-ref"
git init -q --bare --ref-format=reftable "$scratch"
git -C "$scratch" fetch -q --filter=tree:0 "$REPO_URL" '+refs/heads/*:refs/heads/*' '+refs/tags/*:refs/tags/*'
# Branch first to keep actions/checkout's old tie-break: bare
# rev-parse would prefer the tag when a branch shares its name.
sha=""
for cand in "refs/heads/$REQUESTED_REF" "refs/tags/$REQUESTED_REF" "$REQUESTED_REF"; do
if sha="$(git -C "$scratch" rev-parse --verify --quiet "$cand^{commit}")"; then
break
fi
sha=""
done
if [[ -z "$sha" ]]; then
echo "::error::'$REQUESTED_REF' does not resolve to a branch, tag, or commit of stablyai/orca."
exit 1
fi
# The object resolving locally is not proof a branch or tag reaches
# it: a partial clone can lazily fetch a bare SHA on demand, and
# GitHub serves PR-only commits by SHA. Reachability is the actual
# trust test.
if [[ -z "$(git -C "$scratch" for-each-ref --contains "$sha" refs/heads refs/tags | head -1)" ]]; then
echo "::error::Commit $REQUESTED_REF is not reachable from any branch or tag of stablyai/orca; refusing to build it with release credentials."
exit 1
fi
echo "Vetted $REQUESTED_REF -> $sha"
echo "sha=$sha" >>"$GITHUB_OUTPUT"
- name: Checkout the requested ref
uses: actions/checkout@v6
with:
# Why an input at all rather than just github.ref: the whole point is to
# build code that has not landed, and the workflow definition itself
# always comes from the dispatch ref — naming the branch here instead
# applies main's current copy of this file to an arbitrary branch.
ref: ${{ steps.vetted.outputs.sha }}
# Version helpers only read HEAD; published versions come from the release API.
fetch-depth: 1
# This job only reads stablyai/orca and never pushes; every write goes
# to the adhoc repo through a minted App token passed by env. Not
# persisting the checkout credential shrinks the blast radius if a build
# step is compromised (zizmor: artipacked).
persist-credentials: false
- name: Setup pnpm
uses: pnpm/setup@v2
with:
install: false
- name: Setup Node.js
uses: actions/setup-node@v6
with:
node-version-file: package.json
cache: pnpm
cache-dependency-path: |
pnpm-lock.yaml
mobile/pnpm-lock.yaml
# Why: packaging bundles the pinned Node into the macOS terminal host. Keyed on the pin alone
# with no restore-keys, so a pin bump starts empty instead of carrying old archives forward.
- name: Locate the pinned Node runtime cache
shell: bash
run: echo "ORCA_NODE_RUNTIME_CACHE_DIR=$HOME/Library/Caches/orca-node-runtime" >>"$GITHUB_ENV"
- name: Cache the pinned Node runtime
uses: actions/cache@v5
with:
path: ~/Library/Caches/orca-node-runtime
key: orca-node-runtime-mac-${{ hashFiles('src/shared/node-runtime-pin.ts') }}
- name: Cache electron-builder downloads
uses: actions/cache@v5
with:
path: |
~/Library/Caches/electron
~/Library/Caches/electron-builder
key: electron-builder-mac-${{ hashFiles('pnpm-lock.yaml') }}
restore-keys: |
electron-builder-mac-
# Why both CPUs: the mac config packages x64 and arm64 from this arm64
# runner, so the install must carry both variants of the native optional deps.
- name: Install dependencies
uses: nick-fields/retry@v4
with:
timeout_minutes: 10
max_attempts: 3
retry_wait_seconds: 30
command: pnpm install --frozen-lockfile --cpu=current,x64,arm64
# Why here: electron-builder's beforePack requires out/mobile-web, and the bundle
# build resolves React Native and Expo from mobile/node_modules.
- uses: ./.github/actions/install-mobile-dependencies
# Why: signing is what makes an adhoc build installable over an existing
# Orca, so a missing cert must fail here rather than after a 20-minute build.
- name: Verify macOS signing environment
run: node config/scripts/verify-macos-release-env.mjs
env:
CSC_LINK: ${{ secrets.MAC_CERTS }}
CSC_KEY_PASSWORD: ${{ secrets.MAC_CERTS_PASSWORD }}
APPLE_ID: ${{ secrets.APPLE_ID }}
APPLE_APP_SPECIFIC_PASSWORD: ${{ secrets.APPLE_APP_SPECIFIC_PASSWORD }}
APPLE_TEAM_ID: ${{ secrets.APPLE_TEAM_ID }}
- name: Compute adhoc version
id: adhoc
shell: bash
env:
REF: ${{ inputs.ref || github.ref_name }}
LABEL: ${{ inputs.label }}
MAIN_REPO_TOKEN: ${{ github.token }}
run: |
set -euo pipefail
# Why this check: the version scripts are read from the branch being built,
# not from main, so a branch cut before the adhoc channel landed has no
# copy of them. Say that plainly instead of failing with a module-not-found.
for script in adhoc-build-version dev-channel-base-version; do
if [[ ! -f "config/scripts/$script.mjs" ]]; then
echo "::error::$REF has no config/scripts/$script.mjs; rebase it onto a main that has the adhoc channel."
exit 1
fi
done
echo "head_sha=$(git rev-parse HEAD)" >>"$GITHUB_OUTPUT"
# Why git tags, not GitHub releases: unpublishing a buggy cut deletes the
# GitHub release and leaves the tag, which still owns that number.
# Releases-only let adhoc sit on a number already taken, so the updater
# would not install it. Empty on failure — the script then falls back
# to package.json.
published="$(GH_TOKEN="$MAIN_REPO_TOKEN" gh api \
"repos/$GITHUB_REPOSITORY/git/matching-refs/tags/v" \
--jq '.[].ref | sub("^refs/tags/"; "")' || true)"
ORCA_PUBLISHED_VERSIONS="$published" ORCA_ADHOC_LABEL="${LABEL:-$REF}" \
node config/scripts/adhoc-build-version.mjs \
>"$RUNNER_TEMP/adhoc-identity.txt"
if ! grep -q '^name=' "$RUNNER_TEMP/adhoc-identity.txt"; then
echo "::error::adhoc-build-version.mjs emitted no release name; $REF's copy of the script is out of sync with this workflow."
exit 1
fi
cat "$RUNNER_TEMP/adhoc-identity.txt" >>"$GITHUB_OUTPUT"
- name: Build app
run: pnpm build:release
env:
NODE_OPTIONS: --max-old-space-size=4096
# Why: adhoc builds are not an official channel — telemetry's transport
# gate accepts only 'stable' or 'rc', so leaving this unset keeps them
# silent, which is correct for unvetted branch artifacts.
ORCA_DIAGNOSTICS_TOKEN_URL: https://www.onorca.dev/diagnostics/token
# After the app build so nothing that cleans out/ can drop it; electron-builder ships it.
- name: Download the orcad deployment template
if: needs.orcad-template-support.outputs.ships == 'true'
uses: actions/download-artifact@v8
with:
name: orcad-template
path: out/orcad-template
# Why the token is minted here and not at the top: installation tokens live
# one hour, everything before this point writes nothing, and the notary round
# trip inside the publish step can be tens of minutes. Minting after the build
# starts the clock at the first call that actually uses it.
- name: Mint adhoc repo token
id: app_token
uses: actions/create-github-app-token@v2
with:
app-id: ${{ secrets.HOURLY_RELEASE_APP_ID }}
private-key: ${{ secrets.HOURLY_RELEASE_APP_PRIVATE_KEY }}
owner: stablyai
repositories: orca-adhoc
- name: Create adhoc release
id: release
shell: bash
env:
GH_TOKEN: ${{ steps.app_token.outputs.token }}
TAG: v${{ steps.adhoc.outputs.version }}
NAME: ${{ steps.adhoc.outputs.name }}
SHA: ${{ steps.adhoc.outputs.head_sha }}
REF: ${{ inputs.ref || github.ref_name }}
# Via env, not inline `${{ }}`: both land inside a shell string, and an
# expression expanded there is substituted before bash parses the line
# (zizmor: template-injection).
ACTOR: ${{ github.actor }}
run: |
set -euo pipefail
short_sha="${SHA:0:12}"
# Why create it up front: electron-builder then uploads into a known tag
# rather than inferring one from package.json.
#
# Why --draft: everything between here and the manifest check is a window
# where the release exists but has no installable assets. A draft is
# absent from the releases list and from listReleaseBuilds, so a job that
# dies in that window — including a hard kill by the job timeout, which
# runs no cleanup step at all — leaves something invisible rather than a
# tag the picker offers and the download 404s on.
gh release create "$TAG" \
--repo "$ADHOC_REPO" \
--title "$NAME" \
--draft \
--notes "Adhoc macOS and Windows dev build of \`$REF\` at commit \`$short_sha\`.
Built from [\`stablyai/orca@$short_sha\`](https://github.com/stablyai/orca/commit/$SHA), cut by @$ACTOR.
**Unlanded and unvetted.** This is somebody's branch, not main. No tests
ran. The macOS build is signed and notarized like a release, so it installs
through Orca's in-app updater and opens without a Gatekeeper prompt — but the
branch may never merge, and this build is deleted after $ADHOC_RETAIN_DAYS days.
**Windows builds are unsigned.** They install and update normally once you
are on one, but a signed Stable or RC build cannot install one through the
in-app updater — download \`orca-windows-setup.exe\` below and run it once
(SmartScreen will warn about an unknown publisher). Every later switch,
including back to Stable, works in-app from there."
echo "tag=$TAG" >>"$GITHUB_OUTPUT"
- name: Publish adhoc macOS artifacts
uses: nick-fields/retry@v4
with:
# Why 45: an attempt is pack + notarize + upload, and the notary queue is
# the unbounded part. Two attempts, because a failed adhoc build has a
# person waiting on it who can simply dispatch again.
timeout_minutes: 45
max_attempts: 2
retry_wait_seconds: 30
command: node --disable-warning=MODULE_TYPELESS_PACKAGE_JSON config/scripts/pinned-node-downloads.mjs darwin-arm64 darwin-x64 && node config/scripts/ensure-native-runtime.mjs --runtime=electron && ORCA_MAC_ADHOC=1 pnpm exec electron-builder --config config/electron-builder.config.cjs --mac --publish always
env:
# Why: electron-builder's github publisher targets the repo named in the
# config; the token must therefore carry write access to orca-adhoc.
GH_TOKEN: ${{ steps.app_token.outputs.token }}
ORCA_ADHOC_BUILD_VERSION: ${{ steps.adhoc.outputs.version }}
ORCA_BUILD_COMMIT: ${{ steps.adhoc.outputs.commit }}
# beforePack and afterPack fail the package when the template is absent.
ORCA_REQUIRE_ORCAD_TEMPLATE: ${{ needs.orcad-template-support.outputs.ships == 'true' && '1' || '' }}
CSC_LINK: ${{ secrets.MAC_CERTS }}
CSC_KEY_PASSWORD: ${{ secrets.MAC_CERTS_PASSWORD }}
# Why all three: electron-builder's notarize step authenticates to the
# Apple notary service with the app-specific password, not with the
# signing cert. Omitting them fails the build rather than skipping it.
APPLE_ID: ${{ secrets.APPLE_ID }}
APPLE_APP_SPECIFIC_PASSWORD: ${{ secrets.APPLE_APP_SPECIFIC_PASSWORD }}
APPLE_TEAM_ID: ${{ secrets.APPLE_TEAM_ID }}
# Why: the updater resolves a tag, then fetches latest-mac.yml from it. A
# release missing that manifest is a tag the picker offers and the download
# 404s on, so fail loudly instead of leaving a broken entry.
- name: Verify update manifest published
shell: bash
env:
GH_TOKEN: ${{ steps.app_token.outputs.token }}
TAG: ${{ steps.release.outputs.tag }}
run: |
set -euo pipefail
assets="$(gh release view "$TAG" --repo "$ADHOC_REPO" --json assets --jq '.assets[].name')"
echo "Published assets:"
echo "$assets"
# Why exit 1 without deleting here: the release is still a draft, so it is
# already invisible to users, and the failure handler below owns cleanup.
for required in latest-mac.yml; do
if ! grep -qx "$required" <<<"$assets"; then
echo "::error::Adhoc draft $TAG is missing $required; the updater could not install it."
exit 1
fi
done
if ! grep -q '\.zip$' <<<"$assets"; then
echo "::error::Adhoc draft $TAG has no ZIP artifact for the updater to download."
exit 1
fi
# Why this is the last mutating step: publishing the draft is what makes the
# build visible to listReleaseBuilds. Doing it only after the manifest check
# means the picker can never offer a release whose assets are incomplete.
- name: Publish the verified release
id: publish_live
shell: bash
env:
GH_TOKEN: ${{ steps.app_token.outputs.token }}
TAG: ${{ steps.release.outputs.tag }}
NAME: ${{ steps.adhoc.outputs.name }}
run: |
set -euo pipefail
# --title again: electron-builder resolves this draft by tag and may
# rewrite its title on upload. Re-asserting here means the name the
# picker reads is the one composed above, whatever it did in between.
gh release edit "$TAG" --repo "$ADHOC_REPO" --draft=false --prerelease --title "$NAME"
echo "Published $TAG as \"$NAME\""
# Why: a draft left behind by a failed publish is invisible to users but still
# holds its tag name. Gated on publish_live not having succeeded so a later
# failure (the prune step) cannot delete a release that already went live and
# that people may already be installing. Why cancelled() too: a run stopped
# from the Actions UI is not a failure(), so without it a manual cancel
# mid-publish would strand the draft.
- name: Discard the draft release on failure
if: >-
(failure() || cancelled()) && steps.release.outputs.tag != '' &&
steps.publish_live.outcome != 'success'
shell: bash
env:
GH_TOKEN: ${{ steps.app_token.outputs.token }}
TAG: ${{ steps.release.outputs.tag }}
run: |
set -uo pipefail
# No --cleanup-tag: an unpublished draft never created a git tag.
echo "Run failed before publish; discarding draft $TAG"
gh release delete "$TAG" --repo "$ADHOC_REPO" --yes ||
echo "::warning::Could not discard draft $TAG; remove it manually."
- name: Prune expired adhoc releases
# Only after a live publish: $TAG is then a non-draft this step must not
# delete, and a run that failed before publishing has nothing to retire.
if: steps.publish_live.outcome == 'success'
shell: bash
env:
GH_TOKEN: ${{ steps.app_token.outputs.token }}
# Protect the tag this run just shipped, so no filter mistake can delete
# a build minutes after the person who cut it was told it exists.
TAG: ${{ steps.release.outputs.tag }}
run: |
set -euo pipefail
# Why compute the cutoff in bash rather than with jq's `now`: this runs
# once per dispatch, and a fixed epoch makes the threshold visible in the
# log when someone asks where their build went.
cutoff=$(( $(date -u +%s) - ADHOC_RETAIN_DAYS * 86400 ))
echo "Pruning adhoc releases published before $(date -u -r "$cutoff" '+%Y-%m-%dT%H:%M:%SZ')"
# --cleanup-tag so pruning does not leave orphan tags with no release or
# assets attached. Drafts are excluded: a stale draft is the failure
# path's business, not the retention window's.
#
# Age comes from publishedAt, never createdAt. GitHub reports createdAt
# as the date of the *commit* a release's tag points at, and every tag
# here is cut from this repo's one seed commit — so all of them carry
# that same createdAt, and the day the window rolled past it the entire
# channel expired at once and a single run deleted it. A release with no
# publishedAt is kept rather than aged by guesswork.
jq_filter='map(select(.isDraft | not))'
if [[ -n "${TAG:-}" ]]; then
jq_filter+=" | map(select(.tagName != \"${TAG//\"/\\\"}\"))"
fi
jq_filter+=" | map(select((.publishedAt // \"\") != \"\"))"
jq_filter+=" | map(select((.publishedAt | fromdateiso8601) < $cutoff)) | .[].tagName"
stale="$(gh release list --repo "$ADHOC_REPO" --limit 200 --json tagName,publishedAt,isDraft \
--jq "$jq_filter")"
if [[ -z "$stale" ]]; then
echo "Nothing to prune."
exit 0
fi
while read -r tag; do
[[ -n "$tag" ]] || continue
# Belt-and-suspenders: the filter above should already exclude $TAG.
if [[ -n "${TAG:-}" && "$tag" == "$TAG" ]]; then
echo "::warning::Prune list still included just-published $tag after protect; skipping delete."
continue
fi
echo "Pruning $tag"
gh release delete "$tag" --repo "$ADHOC_REPO" --yes --cleanup-tag || \
echo "::warning::Could not prune $tag"
done <<<"$stale"
# Why this runs after the mac leg rather than beside it: the tag and version
# are computed inside that job, so until it has run nothing else can name the
# release to upload into. The cost is small enough not to matter — the Windows
# leg measures ~7.5 min (install 2m45, build 35s, NSIS package 3m) against a
# ~9.5 min mac run, which keeps a adhoc run far inside its interval.
#
# Why `./` rather than a pinned `@main`: `uses:` resolves against the ref this
# file itself came from, which for an ordinary dispatch is main. Someone who
# deliberately points the Actions "Use workflow from" picker at a branch
# already gets that branch's copy of this entire file, so this follows the same
# rule instead of inventing a second one.
build-adhoc-win:
needs: build-adhoc-mac
# Only once the mac release is actually live: there is no release to upload
# into otherwise, and the Windows workflow refuses to create one.
if: needs.build-adhoc-mac.outputs.published == 'true'
uses: ./.github/workflows/dev-channel-win-build.yml
secrets: inherit
with:
channel: adhoc
tag: ${{ needs.build-adhoc-mac.outputs.tag }}
ref: ${{ needs.build-adhoc-mac.outputs.head_sha }}
version: ${{ needs.build-adhoc-mac.outputs.version }}
orcad_template: ${{ needs.build-adhoc-mac.outputs.ships_orcad_template == 'true' }}