The singleton/global-state bucket was the only bucket left unchunked, and
it is the one that got OOM-killed: 79 files in a single `bun test`
process exited 137. Chunking exists precisely to hold peak RSS under the
runner's ceiling, so the bucket that opts out is the bucket that hits it.
The stated reason for leaving it whole does not hold. That bucket is
selected because its suites mutate process-wide state, and what keeps
them from colliding is sequencing, which `parallel: 1` already
guarantees. Splitting them across processes strictly increases isolation
rather than reducing it. Use 10 files per chunk, the width the 650-file
native bucket already sustains on the same runner.
Separately, the sequential path could not say why a chunk was SIGKILLed.
It arms a watchdog but never records when the watchdog fires, so an OOM
kill and a watchdog kill both surfaced as a bare "failed with exit code
137" - the exact message this failure produced, which points at neither
remedy. The parallel path already tracks this; apply the same idea here
and name the cause.
Drives real subprocesses so the 137 under test is a genuine SIGKILL
rather than a hand-written constant: `kill -9 $$` for the OOM-killer
shape, and a watchdog-killed `sleep` for the timeout shape.
The runner stacked two independent parallelism knobs: OMP_TEST_CONCURRENCY
spawned N `bun test` processes while each process ran its own --parallel=M
test files, so the workspace bucket put 4 x 8 = 32 files in flight on a
4-core runner. bun's per-test timeout is wall-clock, so CPU-starved suites
crossed the 5s default and failed at random - mnemopi's sqlite/CLI files
tripped a different pair every run.
- TestCommand now carries a `parallel` request instead of baking the flag
into argv; the dispatcher resolves it against one shared budget
(availableParallelism x 2, split by the live pool width) so total
in-flight files track the machine. A chunk that runs alone still gets
its full requested width, leaving the sequential CI path unchanged.
- Raised the per-test timeout to 30s (OMP_TEST_TIMEOUT to override).
Suites here build real SQLite schemas and spawn CLIs, already running
1-4s per case on a quiet runner; the 10-minute chunk watchdog stays the
backstop for an actual hang.
- --dry-run now resolves the same budget, so it prints the argv a real run
would use, and the pool header reports cores plus the granted width.
- Dropped the dead `{ smol: true }` argument: workspaceTestCommand never
accepted it, and this file's own findings say a smaller heap makes bun
1.3.14's GC crash more often, so it should not be wired up.
Runtime: the loader's AVX2 probe used [System.Runtime.Intrinsics.X86.Avx2]
via powershell.exe, which only exists on .NET Core — stock Windows PowerShell
5.1 raised TypeNotFound, so every Windows host silently loaded the baseline
addon and paid ~270ms for the spawn on the startup path. Ask the kernel via
bun:ffi (IsProcessorFeaturePresent(PF_AVX2_INSTRUCTIONS_AVAILABLE), ~0.5ms)
and fall back to pwsh-then-powershell for Node embeds. scripts/host-detect.ts
shared the same broken probe for build-time variant selection.
Build: `bun run build:native` on a win32 host died deep inside the bazel msvc
repo rules (linux/mac exec hosts only, by design). The `host` pseudo-target
now delegates to the local napi build against real VS Build Tools, and other
targets fail fast with guidance. build-bindings.ts resolves the @napi-rs/cli
JS entry from its manifest (the node_modules/.bin shim is a PE launcher Bun
cannot parse) and auto-appends VS Build Tools' bundled CMake/Ninja to PATH
via vswhere so a vcvars prompt is no longer required. Cap maudio at
opt-level=1 to dodge a rustc ICE codegenning MaybeUninit<ma_fence> for
x86_64-pc-windows-msvc under the pinned nightly (Bazel's older pin is
unaffected).
Compile: Bun.Glob.scan yields backslash paths on Windows; the legacy Pi
virtual module used them verbatim for export keys and generated identifiers,
producing invalid JavaScript in compiled-binary builds.
Tests: strip ConPTY negotiation escapes in the pty argv test; ignore the
bazel-oh-my-pi convenience symlink.
The stated reason for the exclusion no longer holds: no test in
packages/mnemopi loads the fastembed model. The suites that touch embeddings
inject a deterministic in-process provider via setEmbeddingProviderForTests,
and the model-cache suite mocks globalThis.fetch, so nothing downloads or
reads a ~270MB model. On a clean checkout with no model cached, all 448 tests
across 73 files pass on linux-x64 in 11.7s.
packages/mnemopi moves into fastWorkspacePackages, which the workspace bucket
runs and which local-ts already covered through localOnlyWorkspacePackages, so
the local full run is unchanged.
- Implemented in-house, zero-dependency utility modules in `pi-utils` covering DOM manipulation, markdown parsing, templating, browser automation helpers, and terminal buffers.
- Migrated packages across the repository to consume the new internal utilities and `omptype` schema validators instead of external dependencies.
- Removed multiple external runtime and development dependencies including Zod, Marked, LRU cache, Turndown, and Puppeteer browser packages.
Generate a sha256sum-compatible checksums file covering every release
binary and the browser-relay zip, and upload it as a release asset so
downloads can be verified offline without hitting the GitHub API.
Wave-4 review finding on #7586:
- release.ts took compareVersions from the pi-utils barrel, which pulls
file-lock and procmgr into the graph; those import pi-natives, whose
entrypoint calls loadNative() and throws in a checkout without the
built addon
- that also meant scripts/release.test.ts never executed under
test:scripts, so the guard wired into CI last round proved nothing
- take the dependency-free version subpath instead
Wave-2 review findings on #7586:
- the guard accepted `17.2.8-rc.1`, and the publish step passes no
--tag, so a prerelease would have become the npm `latest` every
unqualified install and `omp update` resolve
- `v17.2.8` passed the guard and reached Cargo.toml verbatim, which
cargo rejects only after every manifest was rewritten; validation now
returns the normalized version and callers write that
Review findings on #7586:
- validate an explicit release version before comparing; the shared
comparator never throws, so 999.bad previously reached every manifest
- ci-release-notes imports the comparator by relative path: the
release_github job runs without bun install
- route Bun cache pruning through compareVersions and delete
compareSemverLikeVersions
- regression test for the release-version guard
- Introduce `@oh-my-pi/omptype` as a new ArkType-compatible schema validation package featuring a lazy JIT runtime, JSON Schema emission, and compatibility adapters.
- Replace `arktype` across workspace packages and test utilities with `@oh-my-pi/omptype`.
- Add benchmark suites, tests, and documentation for the new validation engine and adapters.
- Update workspace build, test runner, and release configurations to include the new package.
install_binary downloaded the release binary, chmod'd it, and printed
"✓ Installed omp" with exit 0 without ever running it. Bun's musl-target
binaries link libstdc++/libgcc dynamically, which stock Alpine/musl systems
lack, so the binary exits 127 with relocation errors while the installer
still claimed success.
Smoke-run `omp --version` after install; on failure print the captured
error, add the `apk add libstdc++ libgcc` remediation for musl targets,
and exit non-zero. Document the Alpine/musl runtime requirement in the
README install section.
Fixes#7545
- prepareNativeCorePackage's files whitelist predated the native/clipboard
module; the packed tarball lacked it and the coding-agent CLI failed with
'Cannot find module @oh-my-pi/pi-natives/clipboard' in the tarball install
smoke. Verified the packed tarball now contains clipboard.js/.d.ts.
Scoped test-runtime detection to explicit runner markers and Bun test entrypoints, so application NODE_ENV/BUN_ENV values no longer make ProcessTerminal headless.
Added subprocess regression coverage and propagated the private marker to test children.
Fixes#7261
- Replaced legacy `SWAP`, `INS`, and `PASTE` commands with unified `PUT` and `CUT` hunks across parser, grammar, tokenizer, and test suites.
- Added support for named registers and span paste operations in clipboard and block execution logic.
- Implemented indentation repair and enhanced gap locator formatting for improved patch resilience.
- Updated documentation, system prompts, and session analysis scripts to reflect the new syntax and header shapes.
`CDPATH=` already cleared the variable for the `cd` subshell, so this is
a readability-only change: `CDPATH=''` is the identical POSIX assignment
written with explicit empty quotes. No behavior change, and there was no
CDPATH leak to fix.
- Enhance CI workflows to build and reuse native addon artifacts instead of recompiling.
- Update bazel cache actions with selective backend detection and separate remote and disk modes.
- Add `--source` CLI option to install prebuilt native targets without requiring Bazel.
- Increase Kata runner memory configuration from 12Gi to 24Gi.
- Add conditional attributes to silence dead-code warnings on platform-specific code and fields.
- Update target dependencies in pi-walker/Cargo.toml with explicit windows-sys feature sets.
- Simplify time cast expressions in linux_reflink and rcopy modules.
- Replaced the napi-cli/cargo-zigbuild/cargo-xwin/sccache build path with
Bazel: rules_rust + crate_universe over Cargo.lock, hermetic zig cc
toolchains (linux-gnu pinned to glibc 2.17, linux-musl), host Xcode for
darwin, and a repo-local hermetic clang-cl + llvm-ml + xwin toolchain for
windows-msvc (bazel/toolchains/msvc).
- All eight shipped addons build as //:natives-<target> via the release
transition in bazel/defs.bzl (opt, thin LTO, cgu=16, stripped, canonical
.node naming); scripts/bazel-natives.ts is the single driver for local
dev and CI.
- Rust validation moved to bazel test + clippy aspects (strict workspace
policy for opted-in crates, default lints elsewhere, mirroring cargo
semantics) and the rustfmt aspect; cargo stays as the dev-iteration
surface, with brush-core/brush-builtins promoted to workspace members
and excluded from cargo dev tasks to keep their historical scope.
- CI caches through an in-cluster bazel-remote action cache (TLS + basic
auth, cluster-internal only); GitHub-hosted runners never touch the
infrastructure and use an actions/cache-backed disk cache instead.
- Deleted the hand-rolled caching machinery: ci-target-cache,
ci-native-artifact-cache, ci-build-native, native-source-hash,
find-native-artifacts, restore-linux-native, native-prewarm workflow,
ensure-* toolchain actions, and all sccache/Swatinem wiring.
- Warm native rebuilds drop from ~20 minutes to seconds; a cold client
with a warm remote cache rebuilds the linux x64 pair in ~2.5 minutes.
- Enhanced CI workflows and GitHub actions to support native artifact caching and parallel builds.
- Added composite actions and scripts for computing sources, finding artifacts, and managing caches.
- Updated infrastructure documentation and runner deployment scripts with revised resource limits.
- Added resolveWindowsShell to locate Git Bash, scoop installs, and path binaries with a fallback to cmd.exe.
- Updated bash-executor to prevent wrapping user commands in cmd.exe when using fallback shell paths.
- Updated installation script to report optional shell status rather than failing when bash is absent.
- Add scripts/ci-target-cache.ts to snapshot and restore Cargo target directories to S3 storage on omp-kata runners.
- Update .github/actions/build-native/action.yml to support target cache restoration, saving, and compiler launcher configurations.
- Add skip_validation input in GitHub workflow actions to bypass clippy and Rust test checks on release runs.
- Detect bun crash exits (signals 132-139) and retry up to MAX_CHUNK_ATTEMPTS in a fresh heap.
- Add `retries` field to ChunkOutcome and surface crash retries in progress output.
- Fix test assertion expecting `stopReason: "aborted"` to use `toolUse` (aborted turns are dropped from context).
- omp update's getBinaryName now detects a musl Linux host (Alpine release file or /lib/ld-musl-* loader, mirroring scripts/install.sh) and downloads omp-linux-musl-<arch>, so self-update no longer replaces a musl install with the glibc build.
- Updated musl-release dry-run assertions to the Bun.build output format the binaries script now emits; promoted the musl changelog entry to [Unreleased].
Muhammad Mustaqeem