Add "Headless Mode"

John McCardle 2025-12-02 02:59:09 +00:00
parent f0d8599fe1
commit 47d8592685
1 changed files with 270 additions and 0 deletions

270
Headless-Mode.md Normal file

@ -0,0 +1,270 @@
# Headless Mode
McRogueFace supports headless operation for automated testing, CI pipelines, and LLM agent orchestration. In headless mode, no window is created and simulation time is controlled programmatically via Python.
**Related Pages:**
- [[Writing-Tests]] - Test patterns using headless mode
- [[Input-and-Events]] - Timer and event system
- [[Animation-System]] - Animations work with step()
**Key Files:**
- `src/GameEngine.cpp::step()` - Simulation advancement
- `src/McRFPy_Automation.cpp::_screenshot()` - Synchronous capture
- `tests/unit/test_step_function.py` - step() tests
- `tests/unit/test_synchronous_screenshot.py` - Screenshot tests
---
## Running in Headless Mode
Launch McRogueFace with the `--headless` flag:
```bash
# Run a script in headless mode
./mcrogueface --headless --exec my_script.py
# Run inline Python
./mcrogueface --headless -c "import mcrfpy; print('Hello headless')"
```
In headless mode:
- No window is created (uses RenderTexture internally)
- Simulation time is frozen until `step()` is called
- Screenshots capture current state synchronously
---
## Simulation Control with step()
The `mcrfpy.step()` function advances simulation time in headless mode:
```python
import mcrfpy
# Advance by specific duration (seconds)
dt = mcrfpy.step(0.1) # Advance 100ms
print(f"Advanced by {dt} seconds")
# Advance by integer (converts to float)
dt = mcrfpy.step(1) # Advance 1 second
# Advance to next scheduled event (timer or animation)
dt = mcrfpy.step(None) # or mcrfpy.step()
print(f"Advanced {dt} seconds to next event")
```
### Return Value
`step()` returns the actual time advanced (float, in seconds):
- In headless mode: Returns the requested dt (or time to next event)
- In windowed mode: Returns 0.0 (no-op, simulation runs via game loop)
### What step() Does
1. Advances internal `simulation_time` by the specified duration
2. Updates all active animations
3. Fires any timers whose intervals have elapsed
4. Does NOT render (call `screenshot()` to trigger render)
---
## Timers in Headless Mode
Timers work with simulation time in headless mode:
```python
import mcrfpy
import sys
fired = [False]
def on_timer(runtime):
"""Timer callback receives simulation time in milliseconds"""
fired[0] = True
print(f"Timer fired at {runtime}ms")
# Set timer for 500ms
mcrfpy.setTimer("my_timer", on_timer, 500)
# Advance past the timer interval
mcrfpy.step(0.6) # 600ms - timer will fire
if fired[0]:
print("Success!")
mcrfpy.delTimer("my_timer")
```
### Timer Behavior
- Timers use `simulation_time` in headless mode (not wall-clock time)
- Timer callbacks receive the current simulation time in milliseconds
- Multiple timers can fire in a single `step()` call if intervals overlap
---
## Synchronous Screenshots
In headless mode, `automation.screenshot()` is synchronous - it renders the current scene state before capturing:
```python
import mcrfpy
from mcrfpy import automation
mcrfpy.createScene("test")
ui = mcrfpy.sceneUI("test")
# Add a red frame
frame = mcrfpy.Frame(pos=(100, 100), size=(200, 200))
frame.fill_color = mcrfpy.Color(255, 0, 0)
ui.append(frame)
mcrfpy.setScene("test")
# Screenshot captures current state immediately (no timer needed)
automation.screenshot("/tmp/red_frame.png")
# Change to green
frame.fill_color = mcrfpy.Color(0, 255, 0)
# Next screenshot shows green (not red!)
automation.screenshot("/tmp/green_frame.png")
```
### Windowed vs Headless Screenshot Behavior
| Mode | Behavior |
|------|----------|
| Windowed | Captures previous frame's buffer (async) |
| Headless | Renders current state, then captures (sync) |
In windowed mode, you need timer callbacks to ensure the frame has rendered:
```python
# Windowed mode pattern (NOT needed in headless)
def capture(dt):
automation.screenshot("output.png")
sys.exit(0)
mcrfpy.setTimer("capture", capture, 100)
```
---
## Animations with step()
Animations update when `step()` is called:
```python
import mcrfpy
frame = mcrfpy.Frame(pos=(0, 0), size=(100, 100))
ui.append(frame)
# Start animation: move x from 0 to 500 over 2 seconds
anim = mcrfpy.Animation("x", 500.0, 2.0, "easeInOut")
anim.start(frame)
# Advance halfway through animation
mcrfpy.step(1.0)
print(f"Frame x: {frame.x}") # ~250 (halfway)
# Complete the animation
mcrfpy.step(1.0)
print(f"Frame x: {frame.x}") # 500 (complete)
```
---
## Use Cases
### Automated Testing
```python
#!/usr/bin/env python3
"""Headless test example"""
import mcrfpy
from mcrfpy import automation
import sys
# Setup
mcrfpy.createScene("test")
mcrfpy.setScene("test")
# Test logic
ui = mcrfpy.sceneUI("test")
frame = mcrfpy.Frame(pos=(50, 50), size=(100, 100))
ui.append(frame)
# Verify
assert frame.x == 50
assert frame.w == 100
# Take verification screenshot
automation.screenshot("/tmp/test_output.png")
print("PASS")
sys.exit(0)
```
### LLM Agent Orchestration
The `step()` function enables external agents to control simulation pacing:
```python
import mcrfpy
from mcrfpy import automation
def agent_loop():
"""External agent controls simulation"""
while not game_over():
# Agent observes current state
automation.screenshot("/tmp/current_state.png")
# Agent decides action (external LLM call)
action = get_agent_action("/tmp/current_state.png")
# Execute action
execute_action(action)
# Advance simulation to see results
mcrfpy.step(0.1)
```
### Batch Rendering
Generate multiple frames without real-time constraints:
```python
import mcrfpy
from mcrfpy import automation
for i in range(100):
# Update scene state
update_scene(i)
# Advance animations
mcrfpy.step(1/60) # 60 FPS equivalent
# Capture frame
automation.screenshot(f"/tmp/frame_{i:04d}.png")
```
---
## API Reference
**Module Functions:**
- `mcrfpy.step(dt=None)` - Advance simulation time
- `dt` (float or None): Seconds to advance, or None for next event
- Returns: Actual time advanced (float)
- In windowed mode: Returns 0.0 (no-op)
**Automation Functions:**
- `automation.screenshot(filename)` - Capture current frame
- In headless: Synchronous (renders then captures)
- In windowed: Asynchronous (captures previous frame buffer)
---
*Last updated: 2025-12-01 - Added for #153*