# CI for @freepeak/dsh-mux.
#
# Node 22 is required (`node --experimental-strip-types`). The suite is offline
# and does not need real CLI binaries or a harness profile.
#
# ---------------------------------------------------------------------------
# RUNNER: these jobs target a SELF-HOSTED runner, not a GitHub-hosted one.
#
# Why: GitHub Free gives an organization only a small monthly Actions allowance
# for *private* repositories, and this repository is private. Hosted jobs were
# therefore never scheduled at all — they failed in about three seconds with
# zero steps executed and an empty `runner_name`, which is GitHub refusing to
# provision a machine, not the workflow failing. The same pattern was confirmed
# on FreePeak/dsh-feature-loop.
#
# The runner is registered at ~/actions-runners/dsh-mux with the custom label
# `dsh-mux`, installed as a launchd service so it survives reboots. To inspect
# or restart it:
#
#   cd ~/actions-runners/dsh-mux
#   ./svc.sh status      # is it running
#   ./svc.sh stop        # pause it
#   ./svc.sh start       # resume it
#   ./svc.sh uninstall   # deregister and remove the service
#
# !! IF THIS REPOSITORY IS EVER MADE PUBLIC, CHANGE `runs-on` BACK TO
# !! `ubuntu-latest` BEFORE YOU DO. On a public repository anyone can open a
# !! pull request, and a pull-request job on a self-hosted runner executes
# !! their code on your machine, with your files and credentials in reach.
# !! Public repositories also get unlimited free hosted minutes, so the limit
# !! this runner exists to dodge stops applying the moment it is public.
# ---------------------------------------------------------------------------

name: CI

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

concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: true

permissions:
  contents: read

jobs:
  test:
    name: test (self-hosted, Node ${{ matrix.node }})
    # The custom label, not bare `self-hosted`: it targets this repository's
    # runner specifically, so a runner added later for another project can never
    # pick this job up by accident.
    runs-on: [self-hosted, dsh-mux]
    # A self-hosted runner is a finite resource — one machine, not an elastic
    # pool. A hung job would occupy it indefinitely, so cap both.
    timeout-minutes: 15
    strategy:
      fail-fast: false
      matrix:
        node: ["22"]
    steps:
      - uses: actions/checkout@v4

      - uses: actions/setup-node@v4
        with:
          node-version: ${{ matrix.node }}
          # No `cache: npm` on self-hosted: the post-job cache save talks to
          # GitHub's cache service and has hung this runner after a green
          # suite. A cold `npm ci` of ~37 packages is seconds; skip the cache.

      # .npmrc sets auto-install-peers=false so unpublished harness peers do not
      # abort the install. Only schemastery + tsdown/typescript/@types/node land.
      - name: Install dependencies
        run: npm ci --no-audit --no-fund

      - name: Unit tests
        run: npm test

      - name: Build host bundle
        run: npm run build

      - name: Re-run artifact gates after build
        run: node --experimental-strip-types --test test/artifacts.test.ts