Repository navigation

Remove surface gate; fix Anthropic/OpenAI capability, AgentTask retry/budget, FetchUrl error leak #3663

Workflow file for this run

	# Every `setup-bun` here takes `latest`, deliberately. The suites are what this
	# repository ships against a moving runtime: `bunfig.toml`'s compatibility shim
	# is written against the gaps in Bun's own `vitest` shim and each of its guards
	# is meant to stand down once Bun ships its own, which only running a new Bun
	# can ever reveal. `engines` already states the floor (`bun >= 1.4.0`, where
	# `node:sqlite` arrives) and `latest` is always above it.
	#
	# The cost is real and accepted: a Bun regression can redden every PR on a day
	# nobody touched the code. That is the trade — finding out on the day Bun ships
	# it, rather than at the next manual bump. One spelling in every job, so there
	# is no version to keep in sync; `packageManager` still pins local installs.
	#
	# Coverage is NOT collected here. Every vitest job below runs one slice of the
	# suite, so a per-job number only means something once the fragments are merged
	# — an artifact per job, a merge job and a cleanup job, paid for on every push.
	# The nightly workflow runs the suite in one job and reports it instead.
	name: Build & Test
	permissions:
	contents: read

	on:
	push:
	branches: [main]
	pull_request:
	workflow_dispatch:

	concurrency:
	group: ${{ github.workflow }}-${{ github.event.pull_request.head.ref \|\| (github.event_name == 'workflow_dispatch' && github.run_id \|\| github.ref_name) }}
	cancel-in-progress: true

	env:
	CI: "true"
	DO_NOT_TRACK: "1"
	TURBO_TELEMETRY_DISABLED: "1"
	# PRs: only packages the change can reach. Push/dispatch: full suite.
	TEST_CHANGED: ${{ github.event_name == 'pull_request' && format('--changed {0}', github.event.pull_request.base.sha) \|\| '' }}

	jobs:
	typecheck-budget:
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v6
	- uses: actions/setup-node@v6
	with:
	node-version: 24
	- uses: oven-sh/setup-bun@v2
	with:
	bun-version: latest
	- run: bun i
	# Fails the PR if any package's type-instantiation count grows past its
	# committed budget (scripts/typecheck-budget.json). Re-baseline intentional
	# growth with: bun run typecheck:budget --update
	- name: Typecheck budget guard
	run: bun run typecheck:budget
	# Co-located `__tests__` are excluded from each package's build program so
	# they never reach `dist`, which also takes them out of every tsc run.
	# `tsconfig.test.json` is the program that still checks them; without this
	# step a type error in a test file is invisible until someone runs it.
	# Must follow the budget guard, which is what emits the `dist/*.d.ts` these
	# programs resolve their own package against.
	- name: Typecheck co-located tests
	run: bun run typecheck:tests

	test-discovery:
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v6
	- uses: oven-sh/setup-bun@v2
	with:
	bun-version: latest
	- run: bun i
	# Fails if any test file is unreachable by section+kind selection. Every
	# other job passes a kind, so an unreachable file silently never runs.
	- name: Test discovery guard
	run: bun scripts/test.ts --check-sections

	lint:
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v6
	- uses: actions/setup-node@v6
	with:
	node-version: 24
	- uses: oven-sh/setup-bun@v2
	with:
	bun-version: latest
	- run: bun i
	# The only thing that runs oxlint under CI. `prepare` skips husky when $CI
	# is set, so the pre-commit hook -- which `--no-verify` bypasses anyway --
	# is otherwise oxlint's sole invoker, and the `error`-level rules in
	# .oxlintrc.json can never fail a build.
	#
	# Still not `needs: build`, but no longer because nothing reads types: the
	# `lint` script builds them itself (turbo-cached `build:types`), because
	# `--type-aware` resolves cross-package imports through `dist/*.d.ts` and
	# types every one of them as `any` when they are missing -- so tsgolint
	# silently finds nothing rather than failing.
	- name: Lint
	run: bun run lint
	# `format-check` had no CI invoker at all -- `prepare` skips husky when
	# $CI is set, so the pre-commit hook was its only one, and `--no-verify`
	# walks around that. Cheap, and the tree is clean today.
	- name: Format check
	run: bun run format-check

	build:
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v6
	- uses: actions/setup-node@v6
	with:
	node-version: 24
	- uses: oven-sh/setup-bun@v2
	with:
	bun-version: latest
	- run: bun i
	- run: bun run build
	- name: Upload build artifacts
	uses: actions/upload-artifact@v7
	with:
	name: build-output
	path: \|
	packages//dist/*
	providers//dist/*
	examples//dist/*
	retention-days: 1

	# Sharded, and it is the only job here that is. The unit slice is ~700 files
	# and ran four times the wall clock of every other job in this workflow, so
	# the run finished when this job did and nothing else on the board mattered.
	# Most of that is not test execution: vitest accounts the slice as ~51% setup
	# (a fresh module registry per file, `vitest.setup.ts` bootstrapping the task
	# registry into each), ~21% transform, ~5% import and ~22% tests — so ~78% is
	# paid per FILE, and dividing the file list divides it. Measured 2.9x across
	# four shards rather than 4x because transform is the part that does not
	# shrink: it is mostly the shared module graph, which every shard still pays
	# in full. `scripts/lib/testShards.ts` does the partition and says why it is
	# dealt from a sorted list rather than sliced.
	#
	# `fail-fast: false` because the shards are one job's worth of tests split up:
	# cancelling the rest on the first red one would report a fraction of the
	# failures and send someone back for a second full run to see the others.
	test-vitest-unit:
	runs-on: ubuntu-latest
	needs: build
	strategy:
	fail-fast: false
	matrix:
	# `i/N` verbatim, so the count lives in one place and the job name
	# reads as the shard it ran.
	shard: ["1/4", "2/4", "3/4", "4/4"]
	steps:
	- uses: actions/checkout@v6
	with:
	fetch-depth: 0
	- uses: actions/setup-node@v6
	with:
	node-version: 24
	- uses: oven-sh/setup-bun@v2
	with:
	bun-version: latest
	- run: bun i
	- name: Download build artifacts
	uses: actions/download-artifact@v8
	with:
	name: build-output
	path: .
	# An empty shard passes: `--changed` narrows a pull request to a handful
	# of files, and the shards past the first then hold none of them.
	- name: Run unit tests via vitest (shard ${{ matrix.shard }})
	run: bun run test:vitest:unit -- --shard ${{ matrix.shard }} ${{ env.TEST_CHANGED }}
	test-vitest-integration:
	runs-on: ubuntu-latest
	needs: build
	steps:
	- uses: actions/checkout@v6
	with:
	fetch-depth: 0
	- uses: actions/setup-node@v6
	with:
	node-version: 24
	- uses: oven-sh/setup-bun@v2
	with:
	bun-version: latest
	- run: bun i
	- name: Download build artifacts
	uses: actions/download-artifact@v8
	with:
	name: build-output
	path: .
	- name: Run integration tests via vitest
	env:
	WORKGLOW_SECRETS_PASSPHRASE: ${{ secrets.WORKGLOW_SECRETS_PASSPHRASE }}
	run: bun run test:vitest:integration -- ${{ env.TEST_CHANGED }}
	test-vitest-rag:
	runs-on: ubuntu-latest
	needs: build
	timeout-minutes: 25
	steps:
	- uses: actions/checkout@v6
	with:
	fetch-depth: 0
	- uses: actions/setup-node@v6
	with:
	node-version: 24
	- uses: oven-sh/setup-bun@v2
	with:
	bun-version: latest
	- run: bun i
	- name: Cache Hugging Face model downloads
	uses: actions/cache@v4
	with:
	path: models
	key: hf-models-rag-${{ runner.os }}-${{ hashFiles('packages/test/src/samples/**') }}
	restore-keys: \|
	hf-models-rag-${{ runner.os }}-
	- name: Download build artifacts
	uses: actions/download-artifact@v8
	with:
	name: build-output
	path: .
	- name: Run rag tests via vitest
	env:
	WORKGLOW_SECRETS_PASSPHRASE: ${{ secrets.WORKGLOW_SECRETS_PASSPHRASE }}
	run: bun run test:vitest:rag -- ${{ env.TEST_CHANGED }}

	test-vitest-ai-provider-hft:
	runs-on: ubuntu-latest
	needs: build
	steps:
	- uses: actions/checkout@v6
	with:
	fetch-depth: 0
	- uses: actions/setup-node@v6
	with:
	node-version: 24
	- uses: oven-sh/setup-bun@v2
	with:
	bun-version: latest
	- run: bun i
	- name: Cache Hugging Face model downloads
	uses: actions/cache@v4
	with:
	path: models
	key: hf-models-hft-${{ runner.os }}-${{ hashFiles('packages/test/src/samples/**') }}
	restore-keys: \|
	hf-models-hft-${{ runner.os }}-
	- name: Download build artifacts
	uses: actions/download-artifact@v8
	with:
	name: build-output
	path: .
	- name: Run HuggingFace Transformers provider tests via vitest
	env:
	WORKGLOW_SECRETS_PASSPHRASE: ${{ secrets.WORKGLOW_SECRETS_PASSPHRASE }}
	run: bun run test:vitest:ai-provider-hft -- ${{ env.TEST_CHANGED }}

	test-vitest-ai-provider-nodellama:
	runs-on: ubuntu-latest
	needs: build
	steps:
	- uses: actions/checkout@v6
	with:
	fetch-depth: 0
	- uses: actions/setup-node@v6
	with:
	node-version: 24
	- uses: oven-sh/setup-bun@v2
	with:
	bun-version: latest
	- run: bun i
	- name: Cache Hugging Face / GGUF model downloads
	uses: actions/cache@v4
	with:
	path: models
	key: hf-models-nodellama-${{ runner.os }}-${{ hashFiles('packages/test/src/samples/**') }}
	restore-keys: \|
	hf-models-nodellama-${{ runner.os }}-
	- name: Download build artifacts
	uses: actions/download-artifact@v8
	with:
	name: build-output
	path: .
	- name: Run LlamaCpp provider tests via vitest
	env:
	WORKGLOW_SECRETS_PASSPHRASE: ${{ secrets.WORKGLOW_SECRETS_PASSPHRASE }}
	run: bun run test:vitest:ai-provider-nodellama -- ${{ env.TEST_CHANGED }}

	test-vitest-ai-provider-api:
	runs-on: ubuntu-latest
	needs: build
	steps:
	- uses: actions/checkout@v6
	with:
	fetch-depth: 0
	- uses: actions/setup-node@v6
	with:
	node-version: 24
	- uses: oven-sh/setup-bun@v2
	with:
	bun-version: latest
	- run: bun i
	- name: Download build artifacts
	uses: actions/download-artifact@v8
	with:
	name: build-output
	path: .
	- name: Run API provider tests via vitest
	env:
	WORKGLOW_SECRETS_PASSPHRASE: ${{ secrets.WORKGLOW_SECRETS_PASSPHRASE }}
	run: bun run test:vitest:ai-provider-api -- ${{ env.TEST_CHANGED }}

	cleanup:
	runs-on: ubuntu-latest
	needs:
	[
	test-vitest-unit,
	test-vitest-integration,
	test-vitest-rag,
	test-vitest-ai-provider-hft,
	test-vitest-ai-provider-nodellama,
	test-vitest-ai-provider-api,
	]
	steps:
	- name: Delete build artifacts
	uses: geekyeggo/delete-artifact@v6
	with:
	name: build-output

Navigation Menu

Repository navigation

Remove surface gate; fix Anthropic/OpenAI capability, AgentTask retry/budget, FetchUrl error leak #3663

Workflow file

Remove surface gate; fix Anthropic/OpenAI capability, AgentTask retry/budget, FetchUrl error leak #3663

Uh oh!

Workflow file for this run