Commit Graph
2 Commits
Author SHA1 Message Date
Neil 8ff6296bc7 Speed up serializer checks and keep native caches stable (#24476)
* Reuse serializer oracle cells and isolate native cache policy

* Preserve native cache post-save paths and record hosted oracle gain

* Record native cache reuse and separate cancel-test startup budget
2026-10-01 21:43:56 -07:00
OrcaWinandm4air 6593d7d194 feat(orcad): run orcad on the pinned Node instead of Bun (#24110)
* ci(daemon): gate PRs on daemon protocol crossing from the newest release

Lands daemon-protocol-facts.mjs from the Windows update diagnostic branch with a
stricter parser, and adds check-daemon-protocol-crossing.mjs (rule R1): the working
tree must attach the newest release tag's daemon. Rollback crossing is reported only.
Runs in the cross-version-wire job, which already has full tags; tag selection moves
to config/scripts/stable-release-tags.mjs so both use one rule.

* feat(persistence): run profile backups in the worker whenever its entry is bundled

* refactor(orcad): make profile and native preflight runtime-neutral

The profile preflight parser now takes the expected runtime identity from the
caller (shipped callers pass the pinned Bun identity), and the native
preflight is renamed to orcad-runtime-native-preflight with neutral wording.

* feat(runtime): pin the Node 24.21.0 server runtime with an offline CI check

Add src/shared/node-runtime-pin.ts (NODE_RUNTIME_PIN, SERVER_TARGETS,
NODE_RUNTIME_ASSETS for all 8 server targets plus the headers tarball),
generated by config/scripts/update-node-runtime-pin.mjs from the nodejs.org
and unofficial-builds SHASUMS. check-node-runtime-pin.mjs verifies, with no
network, that the pin tracks the locked Electron, matches engines.node's
major, and covers exactly SERVER_TARGETS; it runs in the static analysis job.

ORCAD_BUN_TARGETS consumers now read SERVER_TARGETS so there is one target
list; orcad's Bun runtime and build output are unchanged.

* test(persistence): skip plain-Node backup selection tests in the Bun profile suite

* fix(runtime): reject a pinned archive that belongs to another target

* ci(daemon): fail PRs that swap a runtime launcher and bump the daemon protocol

D7.1 R3: hosting orcad or the daemon on another runtime is not a protocol change,
so one PR must not do both. The launcher file list lives in the check script; the
allow-runtime-launcher-protocol-bump label overrides it.

* feat(orcad): select pinned-Node slots by a .runtime-node marker

D7.1 R5: a Node slot names its shared runtimes/node-<sha256>/node through
.runtime-node instead of .build-target, so Bun-era clients read it as a legacy
slot rather than exiting 78 on a missing bun-runtime. Nothing builds the marker yet.

* fix(runtime): load the Node pin without the typeless-module warning

check-node-runtime-pin.mjs now requires the pin and takes nodeDistArchiveName from
its own module, so it no longer loads the update script's build graph.

* fix(orcad): resolve Node slots to the design's runtimes/node-<sha>/bin/node layout

* feat(orcad): 8-slot node-pty prebuilds against the pinned Node headers at N-API 8

- build-orcad-prebuilds.mjs adds win32-x64/arm64 (conpty.node, the vendored
  conpty.dll/OpenConsole.exe, upstream's N-API conpty_console_list.node), compiles
  in a scratch copy against the hash-verified pinned headers (node.lib pinned per
  Windows arch) with NAPI_VERSION=8, rejects post-8 node_api_* imports, and writes
  a schema 2 manifest with per-file sha256, N-API level and the glibc need.
- --require-slots [slots] verifies files against hashes; --smoke loads the slot
  under the pinned Node and spawns a PTY; --print-slot names the host slot.
- The slot installer gates on N-API, libc, arch, glibc and file hashes instead of
  the exact NODE_MODULE_VERSION, and installs nested files (conpty/).
- bun-profile-tests.yml builds, verifies and smokes each runner's slot.

* fix(orcad): scope node-pty's glibc .symver pins to glibc on musl prebuild slots

musl's unversioned libc cannot satisfy openpty@GLIBC_* references at link
time, so the Alpine slot compile would fail. Pin the staged pty.cc guard to
__GLIBC__ and assert both musl transforms against the installed patch.

* feat(orcad): run orcad on the pinned Node instead of Bun

A packaged orcad slot now references the pinned Node 24.21.0 by its
executableSha256 (`.runtime-node`, `.server-target`) instead of carrying
bun-runtime, and ships node-pty from the slot's prebuild, only its own
ripgrep, and no Windows Bun PTY gate. The runtime lives beside the slots
at runtimes/node-<sha>/bin/node (node.exe on Windows, upstream name).

- build:orcad (build-orcad-node.mjs) builds the host slot's prebuild when
  missing and places the pinned runtime; the template is schema 3 with
  per-target files.
- handoffToBundledOrcad() resolves the slot's runtime reference and checks
  process.versions.node against the pin; a host Node >= 18 still hands off.
  Startup preflight keys on running as that runtime; callers expect 'node'.
- orcad and its daemon use node-pty (ConPTY + windows-pty-job on Windows);
  the Bun PTY sources, gate entry and canUseBunPty branches are removed.
- SSH deploy uploads the official archive once per pin, extracts and
  hash-checks it on the host, and self-tests it before publishing. Bun
  slots stay launchable for rollback; Node slots never use host Node.
- The runtime materializer is generic over pinned assets; the Bun wrapper
  remains only for the OpenCode vault reader (design Phase 2).
- Cross-runtime test: a profile DB written by Bun 1.4.2 (WAL left by
  SIGKILL) opens and backs up under the pinned Node, and the reverse.

No daemon PROTOCOL_VERSION change (design D7.1 R3).

* docs(ci): name the headless lanes after the pinned Node, drop Bun shard timings

Design D10: ci-demand-rollout.md and ci-runner-efficiency.md follow the
bun-profile-tests.yml -> node-server-tests.yml rename; shard timings drop the
deleted Bun PTY tests and follow the renamed ones.

* chore(ci): count the runtime archive download as a runtime launcher path

* fix(orcad): pin the macOS C++ standard for node-pty prebuilds

The official Node headers' config.gypi sets clang: 0, so common.gypi skips its
gnu++20 xcode_settings and Apple clang 15 (macos-14 runners) compiles
node-addon-api as C++98.

* fix(orcad): resolve the preflight's slot through realpath, as the handoff does

A symlinked orcad.js handed off to its real slot's pinned Node, but the
startup and profile preflights read the symlink's directory, found no
runtime marker there, and silently skipped the readiness check.

* refactor(ssh): drop materializeCachedNodeRuntime, which nothing calls

Deploys upload the verified official archive (design D5); no client path
needs an extracted Node executable cached by digest.

* test(orcad): gate the Bun-to-Node upgrade and Node-to-Bun rollback with live terminals

Design D7.1 R1/R3/R4 and D7.2. The last Bun orcad and this checkout's Node
slot are installed side by side under ~/.orca-remote, launched and stopped
with the client's own deploy commands, and share one data root. Each
direction proves the incoming orcad adopts the outgoing runtime's daemon
(same PID, same shell, output continues), opens its profile database and
backs it up with its own shipped worker, and that GC keeps the slot the
live daemon was forked from.

The node-server Linux lanes provide Bun 1.4.2 and build that Bun orcad
from main, and run with --cross-runtime. --artifact and --cross-runtime
now make their tests fail on a missing input instead of skipping.

* ci(node-server): pin node:24.21.0-alpine by its multi-arch index digest

* test(ssh): name the runtime archive fixture after its role

* test(node-server): load node-pty from the packaged slot in artifact runs

The node-server lane installs dependencies without building node-pty, and
Linux has no upstream prebuild, so the real-PTY failed-I/O teardown test
(picked up by the pty-subprocess selector) could not load pty.node. In
--artifact runs, alias node-pty to out/orcad's shipped slot so the test
exercises the addon orcad actually runs under the pinned Node.

* fix(orcad): let the Windows profile preflight exit after its PTY probe

On Windows, node-pty keeps the conout worker thread and pseudoconsole alive
until kill(), even after the shell exits. The PTY health probe never killed a
cleanly exited probe, so the packaged preflight printed its readiness line
and then hung until the build's 30s timeout, reported with an empty stderr.

- The probe kills its PTY on Windows after exit and uses the bundled ConPTY
  the daemon spawns with.
- The preflight exits once stdout is flushed; its owner reads to EOF.
- Preflight failures now report code, signal, timeout, stdout and stderr.

* test(node-server): load the slot's node-pty in the real-PTY test, not by alias

A vite alias redirected only ESM imports of node-pty; windows-pty-job and
local-pty-utils resolve it through require, so Windows loaded two conpty.node
copies and the Git Bash job-membership proof read an empty job. The failed-I/O
teardown test now loads node-pty through a fixture that picks the packaged slot
in artifact lanes.

The pty-subprocess selector was a prefix that also pulled in its POSIX-host
sibling unit tests, which pr.yml runs and which were never qualified on
Windows. Select the directory plus the two sibling files that belong here.

---------

Co-authored-by: m4air <m4air@m4airs-Air.localdomain>
2026-10-01 00:39:00 -07:00