Publish .github/ from marimo

839c1d7 verified 4 days ago

3.45 kB

name: live desktop + capability gallery

# Real-OS coverage for the parts a synthetic screenshot cannot reach: capture
# backends, native input, and DPI scaling. Linux runs the live test under Xvfb,
# Windows runs it on an interactive desktop session.
#
# Trigger it by hand from the Actions tab, or on every push to main.
#
# The same jobs also build the capability gallery (showcase.py) on each real OS,
# so the screenshots in the model card can be regenerated - including on Windows,
# which the card itself was not built on.

on:
  workflow_dispatch:
  push:
    branches: [main]

jobs:
  linux-xvfb:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: "3.12"
      - name: Install system bits (virtual X server + tkinter)
        run: sudo apt-get update && sudo apt-get install -y xvfb python3-tk
      - name: Install the harness
        run: python -m pip install -e ".[desktop,ocr,dev]"
      - name: Unit suite (needs no display)
        run: python -m pytest tests -q --no-header
      - name: Live desktop test under Xvfb
        run: |
          Xvfb :99 -screen 0 1280x800x24 &
          sleep 2
          export DISPLAY=:99
          python -m tiny_computer_agent --doctor --live || true
          TINY_COMPUTER_AGENT_LIVE_TEST=1 python -m pytest tests/test_live_desktop.py -q -rA
      - name: Capability gallery on a real X server
        run: |
          export DISPLAY=:99
          python showcase.py
      - name: Upload the Linux gallery
        uses: actions/upload-artifact@v4
        with:
          name: gallery-linux
          path: gallery/*.png

  windows:
    runs-on: windows-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: "3.12"
      - name: Install the harness
        run: python -m pip install -e ".[desktop,ocr,dev]"
      - name: Preflight on real Windows
        run: python -m tiny_computer_agent --doctor --live
      - name: Unit suite
        run: python -m pytest tests -q --no-header
      - name: Live desktop test (real Windows desktop)
        # continue-on-error on purpose: the first run tells us whether a hosted
        # runner's session accepts injected input. Read the log either way.
        continue-on-error: true
        env:
          TINY_COMPUTER_AGENT_LIVE_TEST: "1"
        run: python -m pytest tests/test_live_desktop.py -q -rA
      - name: Capability gallery on a real Windows desktop
        # The point of this job: the same showcase.py, on Windows pixels.
        continue-on-error: true
        run: python showcase.py
      - name: Upload the Windows gallery
        uses: actions/upload-artifact@v4
        with:
          name: gallery-windows
          path: gallery/*.png

  macos:
    runs-on: macos-latest
    # macOS gates screen capture and input behind Accessibility permissions that a
    # hosted runner cannot grant, so this job only reports what it can: the doctor.
    continue-on-error: true
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: "3.12"
      - name: Install the harness
        run: python -m pip install -e ".[desktop,ocr,dev]"
      - name: Preflight on macOS
        run: python -m tiny_computer_agent --doctor
      - name: Unit suite
        run: python -m pytest tests -q --no-header

	name: live desktop + capability gallery

	# Real-OS coverage for the parts a synthetic screenshot cannot reach: capture
	# backends, native input, and DPI scaling. Linux runs the live test under Xvfb,
	# Windows runs it on an interactive desktop session.
	#
	# Trigger it by hand from the Actions tab, or on every push to main.
	#
	# The same jobs also build the capability gallery (showcase.py) on each real OS,
	# so the screenshots in the model card can be regenerated - including on Windows,
	# which the card itself was not built on.

	on:
	workflow_dispatch:
	push:
	branches: [main]

	jobs:
	linux-xvfb:
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v4
	- uses: actions/setup-python@v5
	with:
	python-version: "3.12"
	- name: Install system bits (virtual X server + tkinter)
	run: sudo apt-get update && sudo apt-get install -y xvfb python3-tk
	- name: Install the harness
	run: python -m pip install -e ".[desktop,ocr,dev]"
	- name: Unit suite (needs no display)
	run: python -m pytest tests -q --no-header
	- name: Live desktop test under Xvfb
	run: \|
	Xvfb :99 -screen 0 1280x800x24 &
	sleep 2
	export DISPLAY=:99
	python -m tiny_computer_agent --doctor --live \|\| true
	TINY_COMPUTER_AGENT_LIVE_TEST=1 python -m pytest tests/test_live_desktop.py -q -rA
	- name: Capability gallery on a real X server
	run: \|
	export DISPLAY=:99
	python showcase.py
	- name: Upload the Linux gallery
	uses: actions/upload-artifact@v4
	with:
	name: gallery-linux
	path: gallery/*.png

	windows:
	runs-on: windows-latest
	steps:
	- uses: actions/checkout@v4
	- uses: actions/setup-python@v5
	with:
	python-version: "3.12"
	- name: Install the harness
	run: python -m pip install -e ".[desktop,ocr,dev]"
	- name: Preflight on real Windows
	run: python -m tiny_computer_agent --doctor --live
	- name: Unit suite
	run: python -m pytest tests -q --no-header
	- name: Live desktop test (real Windows desktop)
	# continue-on-error on purpose: the first run tells us whether a hosted
	# runner's session accepts injected input. Read the log either way.
	continue-on-error: true
	env:
	TINY_COMPUTER_AGENT_LIVE_TEST: "1"
	run: python -m pytest tests/test_live_desktop.py -q -rA
	- name: Capability gallery on a real Windows desktop
	# The point of this job: the same showcase.py, on Windows pixels.
	continue-on-error: true
	run: python showcase.py
	- name: Upload the Windows gallery
	uses: actions/upload-artifact@v4
	with:
	name: gallery-windows
	path: gallery/*.png

	macos:
	runs-on: macos-latest
	# macOS gates screen capture and input behind Accessibility permissions that a
	# hosted runner cannot grant, so this job only reports what it can: the doctor.
	continue-on-error: true
	steps:
	- uses: actions/checkout@v4
	- uses: actions/setup-python@v5
	with:
	python-version: "3.12"
	- name: Install the harness
	run: python -m pip install -e ".[desktop,ocr,dev]"
	- name: Preflight on macOS
	run: python -m tiny_computer_agent --doctor
	- name: Unit suite
	run: python -m pytest tests -q --no-header