check

fix: make the chain grouping comment-aware so a gap line comment never welds onto a trailer #2137

Workflow file for this run

	# Three jobs run on every push to main and every PR:
	#
	# check — runs the local `deno task check` gate verbatim, so the two cannot
	# drift: rustfmt, typecheck, the audit suite, cargo + deno tests,
	# AST-type drift, clippy with -D warnings. The authoritative list is
	# the `check` task in deno.json — deliberately NOT re-enumerated here,
	# because a hand-copied list goes stale every time an audit is added.
	# pins:audit here is the pin-AGREEMENT half only (a repo fact, so it
	# holds on this clean checkout); checkout alignment is the separate
	# pins:audit:checkouts task, gated at conformance cadence where the
	# sibling checkouts are actually read. The pinned-count guards that run
	# in CI are the committed-tree ones (fixtures_validate via cargo test,
	# swallow_audit). Likewise roundtrip:audit:prettier warn-SKIPS here —
	# it widens onto a ../prettier checkout, which this runner has no
	# reason to clone — so a green CI run does not cover that leg; it
	# gates on a dev machine and at conformance cadence (audit:corpus).
	# artifacts — publish-parity: builds the WASM/npm packages, checks size
	# bounds, runs the Node package tests (see the job comment).
	# platforms — macOS + Windows confidence for the native builds: the full
	# `cargo test --workspace` suite plus the N-API JS-boundary and
	# npm-package tests, per OS (see the job comment). Linux is
	# covered by `check`.
	#
	# `check` needs Rust and Deno: the fixture suite drives prettier + svelte
	# through a Deno sidecar (npm: specifiers pinned in
	# crates/tsv_debug/src/deno/sidecar.ts and locked by its sibling deno.lock,
	# frozen at runtime — hence the cache-hash on both below), so
	# `cargo test --workspace` panics without Deno — tests/fixtures_tests.rs
	# health-checks the sidecar first. wasm-pack is not needed by `check`.

	name: check

	on:
	push:
	branches: [main]
	pull_request:
	branches: ['**']
	# Allow manually re-running checks from the Actions tab.
	workflow_dispatch:

	concurrency:
	group: check-${{ github.ref }}
	cancel-in-progress: ${{ github.event_name == 'pull_request' }}

	# This job only reads the repo; deny everything else (least privilege).
	permissions:
	contents: read

	jobs:
	check:
	runs-on: ubuntu-latest

	# The fixture suite fetches pinned npm packages at runtime (frozen to the
	# sidecar deno.lock), so bound the job to fail fast on a registry stall or hung sidecar
	# instead of running to the 6h default.
	timeout-minutes: 30

	steps:
	# persist-credentials: false keeps the GITHUB_TOKEN out of .git/config (no step here pushes via git).
	- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
	with:
	persist-credentials: false

	# Pinned, not `stable`: the workspace runs clippy::nursery at warn and the
	# `lint` task denies warnings, so a floating toolchain would fail CI
	# whenever a new Rust release adds a nursery lint. Edition 2024 needs
	# >= 1.85. Bump deliberately.
	- name: Install Rust
	uses: dtolnay/rust-toolchain@6c977a6ca4077a0ceb28ffbe03f59d46e9ac8772 # master (no semver tags; bump manually)
	with:
	toolchain: 1.94.1
	components: clippy, rustfmt

	- uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2

	- name: Install Deno
	uses: denoland/setup-deno@22d081ff2d3a40755e97629de92e3bcbfa7cf2ed # v2.0.5
	with:
	# Pinned to match local: deno drives the prettier/svelte fixture oracle
	# and `deno test`, so a floating minor could shift behavior the pinned
	# Rust toolchain can't guard. Bump deliberately.
	deno-version: 2.8.0
	# Round-trips DENO_DIR across runs — captures the sidecar's runtime
	# npm fetches (prettier/svelte/acorn) since those land in DENO_DIR.
	cache: true
	# Key on the sidecar pins (fetched at runtime, not lockfile-pinned) so a
	# bump invalidates the cached deps. The sidecar's prettier/svelte fetches
	# are the only runtime npm `check` does: `test:deno` runs the
	# dependency-free divergence suite (node: builtins + relative imports, no
	# node_modules — proven by a clean checkout), and the bench's heavy npm
	# tree (biome/oxc/oxfmt) is never installed here. The only deno.lock is the
	# sidecar's (crates/tsv_debug/src/deno/deno.lock, embedded + frozen); the
	# bench resolves npm from package-lock.json and the root has no external deps.
	cache-hash: ${{ hashFiles('crates/tsv_debug/src/deno/sidecar.ts', 'crates/tsv_debug/src/deno/actor.rs', 'crates/tsv_debug/src/deno/deno.lock') }}

	- name: Check
	run: deno task check

	# Publish-parity gate: builds every WASM/npm artifact, checks the size
	# bounds, and runs the Node package tests — none of which `check` exercises,
	# so a packaging break surfaces on PRs instead of at release time. Runs the
	# same `build:packages` task as scripts/publish.ts (6 wasm-pack release builds
	# + wasm-opt), so the two can't drift.
	artifacts:
	runs-on: ubuntu-latest
	# Runs in parallel with `check` (no `needs:`) so PR wall-clock is
	# max(check, artifacts), not the sum. Tradeoff: a PR that fails `check`
	# still burns the artifact build. Re-add `needs: check` to trade that
	# latency back for not spending build minutes on red PRs.
	# Generous headroom over the cold wasm-only build (6 wasm-pack release builds
	# + wasm-opt; wasm-pack itself is now a prebuilt download, not a source
	# compile) — well short of the 6h default so a hang fails fast.
	timeout-minutes: 45

	steps:
	# persist-credentials: false keeps the GITHUB_TOKEN out of .git/config (no step here pushes via git).
	- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
	with:
	persist-credentials: false

	- name: Install Rust
	uses: dtolnay/rust-toolchain@6c977a6ca4077a0ceb28ffbe03f59d46e9ac8772 # master (no semver tags; bump manually)
	with:
	toolchain: 1.94.1
	targets: wasm32-unknown-unknown

	# Caches the wasm32 dependency builds across runs (keyed on Cargo.lock).
	- uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2

	# Prebuilt binary, not `cargo install` from source: a source build is keyed
	# by rust-cache on Cargo.lock, so any dep bump forced a multi-minute wasm-pack
	# recompile. The download is independent of the cache and runs in seconds.
	# Version pinned to match local: wasm-pack bundles wasm-opt, whose output size
	# the ~±8% validate:artifacts bounds are calibrated against — a float would trip
	# the size gate on an unrelated binaryen release. Bump the SHA and wasm-pack
	# version manually.
	- name: Install wasm-pack
	uses: taiki-e/install-action@7f4eb899022d8fe70b20c4f3de697aa85c309026 # v2.85.11
	with:
	tool: wasm-pack@0.15.0

	- name: Install Deno
	uses: denoland/setup-deno@22d081ff2d3a40755e97629de92e3bcbfa7cf2ed # v2.0.5
	with:
	# Pinned to match the check job / local. Bump deliberately.
	deno-version: 2.8.0
	cache: true

	- name: Install Node
	uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
	with:
	# Pinned to match local. Needs >= 22.18 for native .ts type stripping
	# (scripts/test_npm.ts runs under node --test directly). Bump deliberately.
	node-version: '24.14.1'

	# Builds only the publishable wasm bundles (npm + deno) — exactly what
	# scripts/publish.ts builds and what validate:artifacts / test:npm consume.
	# NOT `build:all`: that prepends an LTO release workspace build + FFI lib that
	# nothing here validates (publish skips them too), so they were pure overhead.
	- name: Build packages
	run: deno task build:packages

	- name: Validate artifact sizes
	run: deno task validate:artifacts

	# All three published shapes differ (format excludes convert, parse
	# bundles the .d.ts, all ships the tsv bin), so test each. The `:run`
	# variants skip the per-package rebuild the bare `test:npm*` wrappers do —
	# `build:packages` above already built all six bundles.
	- name: Node package tests
	run: \|
	deno task test:npm:run
	deno task test:npm:parse:run
	deno task test:npm:all:run

	# Platform confidence for the native builds: the committed-tree test suite
	# (fixtures, CLI integration, the napi in-crate tests) plus the real N-API
	# JS-boundary test, on each OS a native artifact will ship for. Linux is
	# already covered by `check` (`deno task check` runs `cargo test --workspace`
	# on ubuntu), so the matrix is macOS + Windows; macos-15 runners are arm64,
	# which is also aarch64 coverage. Deliberately NOT the full `deno task check`:
	# rustfmt/clippy/typecheck/the audits are platform-independent and already
	# gate on ubuntu — duplicating them here is wall-clock for no new signal.
	# The highest-yield surfaces per OS are path handling (tsv_ignore /
	# tsv_discover / tsv_cli discovery, exercised by cli_tests) and the native
	# cdylib boundary (test:napi). Byte-exact fixtures survive Windows checkout
	# because .gitattributes pins eol=lf.
	platforms:
	strategy:
	# A Windows failure must not cancel the macOS signal (or vice versa) —
	# per-OS triage needs both results every run.
	fail-fast: false
	matrix:
	os: [macos-15, windows-2025]
	runs-on: ${{ matrix.os }}
	# Windows runners build roughly 2x slower than ubuntu; generous headroom,
	# still far from the 6h default so a hung sidecar fails fast.
	timeout-minutes: 60

	steps:
	# persist-credentials: false keeps the GITHUB_TOKEN out of .git/config (no step here pushes via git).
	- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
	with:
	persist-credentials: false

	# Same pin as `check`; no clippy/rustfmt components — lints gate on ubuntu.
	- name: Install Rust
	uses: dtolnay/rust-toolchain@6c977a6ca4077a0ceb28ffbe03f59d46e9ac8772 # master (no semver tags; bump manually)
	with:
	toolchain: 1.94.1

	- uses: Swatinem/rust-cache@6323deb102c322ba6fcbdcafc7e3dddab59af2b6 # v2.9.2

	# `cargo test --workspace` needs Deno: the fixture suite drives prettier +
	# svelte through the Deno sidecar (see the `check` job comment; same
	# cache-hash reasoning — the sidecar's npm pins are fetched at runtime).
	- name: Install Deno
	uses: denoland/setup-deno@22d081ff2d3a40755e97629de92e3bcbfa7cf2ed # v2.0.5
	with:
	# Pinned to match the check job / local. Bump deliberately.
	deno-version: 2.8.0
	cache: true
	cache-hash: ${{ hashFiles('crates/tsv_debug/src/deno/sidecar.ts', 'crates/tsv_debug/src/deno/actor.rs', 'crates/tsv_debug/src/deno/deno.lock') }}

	# For test:napi's runner (`node --test` executes the .ts directly via
	# native type stripping, >= 22.18). Pinned to match the artifacts job.
	- name: Install Node
	uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
	with:
	node-version: '24.14.1'

	# --no-fail-fast: cargo test stops at the first failing test BINARY by
	# default, hiding every later suite — the platform burn-down wants all
	# failures per run, not one per push.
	- name: Test
	run: cargo test --workspace --no-fail-fast

	# The real N-API JS boundary against this OS's napi-profile cdylib
	# (release + unwind, built with the test-only panic probe so the
	# panic-contract test runs; per-OS naming — .dll / .dylib — handled by
	# benches/js/lib/runtime.ts). This is the only tsv_napi build on every PR
	# (release_napi.yml builds it on tags, dispatch, and a weekly cron).
	# Runs even when Test fails (independent signal
	# during the burn-down; its own failure still fails the job).
	- name: N-API boundary test
	if: ${{ !cancelled() }}
	run: deno task test:napi

	# The staged npm shape on this OS: loader platform-resolution (per-OS
	# triple detection), the wasm-parity options surface, and the packaged
	# .node — what a real `npm i @fuzdev/tsv` consumer executes.
	- name: N-API npm package test
	if: ${{ !cancelled() }}
	run: deno task test:napi:npm

Navigation Menu

Uh oh!

Uh oh!

fix: make the chain grouping comment-aware so a gap line comment never welds onto a trailer #2137

Workflow file

fix: make the chain grouping comment-aware so a gap line comment never welds onto a trailer #2137

Uh oh!

Workflow file for this run