scar-chat7/PROGRESS-SCREENCAPTURE.md

Screen Capture Implementation Progress

Overview

Implementation of screen capture functionality for SCAR Chat with support for multiple backends:

• Wayland/Hyprland: xdg-desktop-portal + Pipewire
• X11: FFmpeg with x11grab
• Windows: FFmpeg with GDI

Current Status: 🟡 In Progress

---

Architecture

Backend Detection

• ✅ Auto-detection via environment variables (WAYLAND_DISPLAY, HYPRLAND_INSTANCE_SIGNATURE)
• ✅ Fallback mechanism (Portal → X11 → Windows)
• ✅ Manual backend selection support

Wayland/Hyprland Implementation (Priority)

Status: 🟡 In Progress - CORRECTED TO USE HYPRLAND PORTAL

Critical Architecture Understanding:

• xdg-desktop-portal-hyprland implements org.freedesktop.portal.ScreenCast (standard API)
• We use sdbus-c++ library (NOT libdbus-1) to communicate with the portal
• The portal handles Hyprland-specific details internally (via hyprland-share-picker)
• From client perspective: call standard portal API → portal shows hyprland-share-picker → get Pipewire stream

Dependencies:

• sdbus-c++ (required for DBus communication) ✅ Installed
• libpipewire-0.3 (for stream handling) ✅ Installed
• xdg-desktop-portal-hyprland (runtime requirement - provides hyprland-share-picker)
• spa-utils (for spa_hook structure)

Implementation Tasks:

• Switch from libdbus-1 to sdbus-c++
• Use standard org.freedesktop.portal.Desktop.ScreenCast interface
• Update CMakeLists.txt with pkg-config for sdbus-c++ and pipewire
• Screen capture class structure with backend selection
• initPortalConnection() - sdbus session bus connection
• cleanupPortalConnection() - Resource cleanup
• createPortalSession() - Use sdbus-c++ for CreateSession method with unique handles
• selectPortalSources() - Use sdbus-c++ for SelectSources (portal shows hyprland-share-picker)
• startPortalSession() - Use sdbus-c++ for Start method
• openPipeWireRemote() - Get file descriptor from portal using UnixFd
• getStreamsNodeId() - Query session Streams property to get actual node_id
• initPipewire() - Complete implementation with stream connection, listeners, and thread loop
• onStreamProcess() - Frame callback implementation that dequeues buffers and invokes user callback
• onStreamParamChanged() - Handle resolution/format changes and update dimensions
• cleanupPipewire() - Stop thread loop and cleanup resources properly
• UI Integration - Add floating screen share button to VideoGridWidget
• Test end-to-end screen capture flow
• Frame buffer memory management optimization
• Error handling and session recovery
• Restore token support for session persistence

UI Integration Details:

• Added "Share Screen" button to VideoGridWidget
• Button floats at bottom center of video grid, above video streams
• Positioned via resize events to stay centered
• Styled with Discord-like blue theme
• Toggles between "Share Screen" (blue) and "Stop Sharing" (red)
• Clicking button calls ScreenCapture::start() which opens xdg-desktop-portal dialog
• ScreenCapture instance managed by VideoGridWidget
• Signal screenShareRequested() emitted when sharing starts

Notes:

• Standard portal API is service name: org.freedesktop.portal.Desktop
• Object path: /org/freedesktop/portal/desktop
• Interface: org.freedesktop.portal.ScreenCast
• When SelectSources is called, xdph automatically launches hyprland-share-picker GUI
• User selection is handled transparently - we just get back session handle + Pipewire node

Technical Details:

// XDG Desktop Portal ScreenCast API Workflow:
// 1. org.freedesktop.portal.ScreenCast.CreateSession(options) -> session_handle
//    - Creates session object for this screen cast
//
// 2. org.freedesktop.portal.ScreenCast.SelectSources(session_handle, options)
//    - options.types: MONITOR(1), WINDOW(2), VIRTUAL(4)
//    - options.multiple: allow selecting multiple sources
//    - options.cursor_mode: Hidden(1), Embedded(2), Metadata(4)
//    - options.persist_mode: DoNotPersist(0), WhileRunning(1), UntilRevoked(2)
//
// 3. org.freedesktop.portal.ScreenCast.Start(session_handle, parent_window, options)
//    - User selects screen/window via portal UI
//    - Response includes: streams array with [(node_id, properties)]
//    - Each stream has: id, position, size, source_type
//    - Returns restore_token for future sessions
//
// 4. org.freedesktop.portal.ScreenCast.OpenPipeWireRemote(session_handle) -> fd
//    - Returns file descriptor for PipeWire connection
//
// 5. Pipewire Connection:
//    - pw_context_connect_fd(fd) creates pw_core
//    - pw_stream_new() with node_id from Step 3
//    - pw_stream_add_listener() for frame callbacks
//    - pw_stream_connect() to start streaming
//
// 6. Frame Processing:
//    - on_process() callback receives spa_buffer with frame data
//    - Extract video/raw format (RGB, YUV, etc.)
//    - Invoke FrameCallback with decoded data

X11 Implementation (Fallback)

Status: 🔴 Not Started

Dependencies:

• FFmpeg (libavformat, libavcodec, libavutil, libavdevice)
• X11 libraries

Implementation Tasks:

• FFmpeg context initialization
• x11grab input device configuration
• Frame extraction and decoding
• Frame callback integration
• Display selection (multi-monitor support)

Windows Implementation (Future)

Status: 🔴 Not Started

Dependencies:

• FFmpeg with GDI support

Implementation Tasks:

• FFmpeg GDI grabber setup
• Frame processing pipeline
• Display enumeration

---

Testing Plan

Unit Tests

• Backend detection on different environments
• Frame callback invocation
• Start/stop lifecycle
• Memory leak verification

Integration Tests

• Wayland/Hyprland capture on real desktop
• X11 capture verification
• Multi-monitor scenarios
• Permission denial handling

Performance Tests

• Frame rate consistency (target: 30 FPS)
• CPU usage profiling
• Memory usage under continuous capture

---

Known Issues & Limitations

Current

• All backends are stubs (no actual implementation)
• No frame encoding/compression
• No multi-monitor selection UI

Future Considerations

• Portal permissions may require user interaction each session
• Hyprland-specific optimizations possible via hyprland-share-picker
• Frame rate limiting needed to prevent CPU overload
• Consider hardware encoding for lower CPU usage

---

Code Locations

• Header: client/media/screen_capture.h
• Implementation: client/media/screen_capture.cpp
• Dependencies: CMakeLists.txt (client section)

---

Next Steps

1. Implement Pipewire + Portal screen capture for Wayland/Hyprland
2. Test on Hyprland environment
3. Implement X11 fallback
4. Add frame encoding for network transmission
5. Integrate with video streaming protocol

---

Session Log

Session 1 - December 7, 2025

• Completed: Authentication system fully working (plaintext → salt → argon2 verification)
• Fixed: Message deserialization bug (async buffer capture issue in both client and server)
• Status: Ready to begin screen capture implementation
• Decision: Prioritize Wayland/Hyprland implementation due to target environment

Session 2 - December 7, 2025 (Current)

• Researched: xdg-desktop-portal-hyprland specifications and org.freedesktop.portal.ScreenCast API
• Implemented:
  • Screen capture header with forward declarations for DBus/Pipewire types
  • Basic structure with backend detection and selection
  • DBus initialization and cleanup functions
  • Pipewire initialization skeleton (loop, context creation)
  • Platform-specific compilation (#ifdef linux)
  • startPortalCapture() workflow outline (6-step process)
• TODO: Implement actual DBus method calls for portal communication
• Next: Implement createPortalSession() with proper DBus message building