compose.dev.yml

# Containerized development runtime. Devkit adds this checkout's PostgreSQL service.
#
# Named `compose.dev.yml` rather than `compose.yml` because the root `compose.yml` is gitignored and
# reserved for a deployment-local copy of `compose.server.example.yml` (see .gitignore). Devkit
# combines this file with the checkout's generated database service.
#
# Browser and host-bridge entry points bind to loopback only. Do not remove the bind address to
# "make it reachable": on a laptop that would expose development services to the current network.
services:
  node:
    build:
      context: .
      dockerfile: docker/dev/Dockerfile
    working_dir: /workspace
    user: "${HOST_UID:?host uid is required}:${HOST_GID:?host gid is required}"
    init: true
    env_file:
      - path: .env
        required: false
    environment:
      HOME: /tmp/devkit-home
      NODE_ENV: development
      DATABASE_URL: ${DATABASE_URL:?database URL is required}
      API_PORT: ${API_PORT:?api port is required}
      METRICS_PORT: ${METRICS_PORT:?metrics port is required}
      CLIENT_ORIGIN: ${CLIENT_ORIGIN:?client origin is required}
      # The bridge watcher runs on the host, where printer LAN routes are available.
      PRINTSTREAM_DEV_BRIDGE: host
      # Preserve the tri-state deployment override so the private source tree can exercise the
      # public self-hosted auth and UI paths with `SELF_HOSTED=true npm run dev`.
      SELF_HOSTED: ${SELF_HOSTED:-}
      VITE_API_PORT: ${VITE_API_PORT:?vite api port is required}
      VITE_DEV_ALLOWED_HOSTS: ${VITE_DEV_ALLOWED_HOSTS:?allowed hosts are required}
      VITE_DEV_HOST: 0.0.0.0
      VITE_DEV_PORT: ${VITE_DEV_PORT:?vite port is required}
      SLICER_PORT: ${SLICER_PORT:?slicer port is required}
      SLICER_DATA_ROOT: /slicer-data
      SLICER_WORK_DIR: /slicer-work
      SLICER_SERVICE_URL: http://slicer:${SLICER_PORT:?slicer port is required}
      DEVKIT_COMPOSE_PROJECT: ${DEVKIT_COMPOSE_PROJECT:?compose project is required}
    ports:
      - "127.0.0.1:${DEVKIT_WEB_PORT:?web port is required}:${VITE_DEV_PORT:?vite port is required}"
      - "[::1]:${DEVKIT_WEB_PORT:?web port is required}:${VITE_DEV_PORT:?vite port is required}"
      - "127.0.0.1:${DEVKIT_API_PORT:?api port is required}:${API_PORT:?api port is required}"
      - "[::1]:${DEVKIT_API_PORT:?api port is required}:${API_PORT:?api port is required}"
    volumes:
      - .:/workspace
      # A linked worktree's `.git` is a file pointing into the primary checkout's absolute Git
      # directory. Mount that directory at the same absolute path so Git-aware validation works
      # inside the source-mounted container too. Read-only is sufficient for diff/test discovery.
      - ${PRINTSTREAM_GIT_COMMON_DIR:?git common dir is required}:${PRINTSTREAM_GIT_COMMON_DIR:?git common dir is required}:ro
      - slicer-data:/slicer-data
      - slicer-work:/slicer-work
    depends_on:
      database:
        condition: service_healthy
    command: node scripts/dev/run-dev.mjs --container-runtime

  # Opt-in: `docker compose -f compose.dev.yml up -d slicer`, then set PRINTSTREAM_DEV_SLICER=remote
  # so `npm run dev` leaves the in-process slicer alone and the API uses SLICER_SERVICE_URL.
  #
  # Why a container rather than the in-process slicer `npm run dev` normally starts: that path needs
  # BambuStudio's x86-64 toolchain, which the development image bakes in and a
  # bare host does not have. On arm64 it needs qemu-user-static, weston and an x86-64 cross compiler
  # besides. The published image already carries all of it, including the emulation sysroot for
  # arm64, so this is the same engine without installing any of it system-wide.
  #
  # Built FROM THIS CHECKOUT, not pulled: in dev a container should run the code you are editing.
  # `npm run dev` rebuilds it when this checkout's source is newer; `dev:doctor` only reports the
  # state. Use `--build` directly when you deliberately want to force a rebuild.
  #
  #   docker compose -f compose.dev.yml --profile slicer up -d --build slicer
  #
  # The slicer belongs to this checkout's Compose project too. That costs one engine process per
  # running worktree, but guarantees each branch runs the slicer source mounted beside its API and
  # lets `devkit --down` remove the complete checkout runtime as one unit.
  #
  # To skip building entirely (you are not touching the slicer and want the fastest start), point
  # SLICER_IMAGE at the published engine and pull it instead of building:
  #
  #   SLICER_IMAGE=ghcr.io/printstreamapp/printstream-slicer:latest \
  #     docker compose -f compose.dev.yml --profile slicer up -d slicer
  slicer:
    # A LOCAL tag by default, so a build from this checkout can never be tagged over the published
    # image other things on this machine may be using.
    image: ${SLICER_IMAGE:-printstream-slicer-dev}
    build:
      context: .
      dockerfile: apps/slicer/Dockerfile
      args:
        PRINTSTREAM_SLICER_SOURCE_FINGERPRINT: ${SLICER_SOURCE_FINGERPRINT:-unknown}
    profiles:
      - slicer
    restart: unless-stopped
    # The CLI launcher spawns helpers (a per-slice weston; qemu on arm64) that become zombies when a
    # slice is cancelled. node-as-PID-1 never reaps them and they accumulate until slicing stops
    # working; tini reaps. Same reason as the production stack.
    init: true
    environment:
      SLICER_PORT: ${SLICER_PORT:?slicer port is required}
      SLICER_WORK_DIR: /work
      SLICER_BAMBUSTUDIO_HOME_DIR: /work/bambustudio-home
      SLICER_BAMBUSTUDIO_DATA_DIR: /work/bambustudio-data
      # Engines download on first use (~530 MB each). Name ids here to preload instead.
      SLICER_PRELOAD_ENGINES: ${SLICER_PRELOAD_ENGINES:-}
    volumes:
      - slicer-work:/work
      # Engines live outside the image so they survive an image update, and outside the repo so a
      # worktree cannot carry 530 MB per checkout.
      - slicer-engines:/data/engines

volumes:
  slicer-data:
  slicer-work:
  slicer-engines:
    name: ${SLICER_ENGINES_VOLUME:-printstream-slicer-compose-engines}