Mercurial > libervia-backend
view libervia/frontends/tools/webrtc.py @ 4232:0fbe5c605eb6
tests (unit/webrtc,XEP-0176, XEP-0234): Fix tests and add webrtc file transfer tests:
fix 441
author | Goffi <goffi@goffi.org> |
---|---|
date | Sat, 06 Apr 2024 12:59:50 +0200 |
parents | fe29fbdabce6 |
children | d01b8d002619 |
line wrap: on
line source
#!/usr/bin/env python3 # Libervia WebRTC implementation # Copyright (C) 2009-2023 Jérôme Poisson (goffi@goffi.org) # This program is free software: you can redistribute it and/or modify # it under the terms of the GNU Affero General Public License as published by # the Free Software Foundation, either version 3 of the License, or # (at your option) any later version. # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU Affero General Public License for more details. # You should have received a copy of the GNU Affero General Public License # along with this program. If not, see <http://www.gnu.org/licenses/>. import gi gi.require_versions({ "Gst": "1.0", "GstWebRTC": "1.0" }) from gi.repository import Gst, GstWebRTC, GstSdp from libervia.backend.core import exceptions try: from gi.overrides import Gst as _ except ImportError: raise exceptions.MissingModule( "No GStreamer Python overrides available. Please install relevant packages on " "your system (e.g., `python3-gst-1.0` on Debian and derivatives)." ) import asyncio from dataclasses import dataclass from datetime import datetime import logging from random import randint import re from typing import Callable from urllib.parse import quote_plus from libervia.backend.tools.common import data_format from libervia.frontends.tools import aio, display_servers current_server = display_servers.detect() if current_server == display_servers.X11: # GSTreamer's ximagesrc documentation asks to run this function import ctypes ctypes.CDLL('libX11.so.6').XInitThreads() log = logging.getLogger(__name__) Gst.init(None) SOURCES_AUTO = "auto" SOURCES_TEST = "test" SINKS_APP = "app" SINKS_AUTO = "auto" SINKS_TEST = "test" class ScreenshareError(Exception): pass @dataclass class AppSinkData: local_video_cb: Callable remote_video_cb: Callable|None class DesktopPortal: def __init__(self, webrtc: "WebRTC"): import dbus from dbus.mainloop.glib import DBusGMainLoop # we want monitors + windows, see https://flatpak.github.io/xdg-desktop-portal/docs/doc-org.freedesktop.portal.ScreenCast.html#org-freedesktop-portal-screencast-availablesourcetypes self.dbus = dbus self.webrtc = webrtc self.sources_type = dbus.UInt32(7) DBusGMainLoop(set_as_default=True) self.session_bus = dbus.SessionBus() portal_object = self.session_bus.get_object( 'org.freedesktop.portal.Desktop', '/org/freedesktop/portal/desktop' ) self.screencast_interface = dbus.Interface( portal_object, 'org.freedesktop.portal.ScreenCast' ) self.session_interface = None self.session_signal = None self.handle_counter = 0 self.session_handle = None self.stream_data: dict|None = None @property def handle_token(self): self.handle_counter += 1 return f"libervia{self.handle_counter}" def on_session_closed(self, details: dict) -> None: if self.session_interface is not None: self.session_interface = None self.webrtc.desktop_sharing = False if self.session_signal is not None: self.session_signal.remove() self.session_signal = None async def dbus_call(self, method_name: str, *args) -> dict: """Call a screenshare portal method This method handle the signal response. @param method_name: method to call @param args: extra args `handle_token` will be automatically added to the last arg (option dict) @return: method result """ if self.session_handle is not None: self.end_screenshare() method = getattr(self.screencast_interface, method_name) options = args[-1] reply_fut = asyncio.Future() signal_fut = asyncio.Future() # cf. https://flatpak.github.io/xdg-desktop-portal/docs/doc-org.freedesktop.portal.Request.html handle_token = self.handle_token sender = self.session_bus.get_unique_name().replace(".", "_")[1:] path = f"/org/freedesktop/portal/desktop/request/{sender}/{handle_token}" signal_match = None def on_signal(response, results): assert signal_match is not None signal_match.remove() if response == 0: signal_fut.set_result(results) elif response == 1: signal_fut.set_exception( exceptions.CancelError("Cancelled by user.") ) else: signal_fut.set_exception(ScreenshareError( f"Can't get signal result" )) signal_match = self.session_bus.add_signal_receiver( on_signal, signal_name="Response", dbus_interface="org.freedesktop.portal.Request", path=path ) options["handle_token"] = handle_token method( *args, reply_handler=reply_fut.set_result, error_handler=reply_fut.set_exception ) try: await reply_fut except Exception as e: raise ScreenshareError(f"Can't ask screenshare permission: {e}") return await signal_fut async def request_screenshare(self) -> dict: session_data = await self.dbus_call( "CreateSession", { "session_handle_token": str(randint(1, 2**32)), } ) try: session_handle = session_data["session_handle"] except KeyError: raise ScreenshareError("Can't get session handle") self.session_handle = session_handle await self.dbus_call( "SelectSources", session_handle, { "multiple": True, "types": self.sources_type, "modal": True } ) screenshare_data = await self.dbus_call( "Start", session_handle, "", {} ) session_object = self.session_bus.get_object( 'org.freedesktop.portal.Desktop', session_handle ) self.session_interface = self.dbus.Interface( session_object, 'org.freedesktop.portal.Session' ) self.session_signal = self.session_bus.add_signal_receiver( self.on_session_closed, signal_name="Closed", dbus_interface="org.freedesktop.portal.Session", path=session_handle ) try: node_id, stream_data = screenshare_data["streams"][0] source_type = int(stream_data["source_type"]) except (IndexError, KeyError): raise ScreenshareError("Can't parse stream data") self.stream_data = stream_data = { "session_handle": session_handle, "node_id": node_id, "source_type": source_type } try: height = int(stream_data["size"][0]) weight = int(stream_data["size"][1]) except (IndexError, KeyError): pass else: stream_data["size"] = (height, weight) return self.stream_data def end_screenshare(self) -> None: """Close a running screenshare session, if any.""" if self.session_interface is None: return self.session_interface.Close() self.on_session_closed({}) class WebRTC: """GSTreamer based WebRTC implementation for audio and video communication. This class encapsulates the WebRTC functionalities required for initiating and handling audio and video calls. """ def __init__( self, bridge, profile: str, sources: str = SOURCES_AUTO, sinks: str = SINKS_AUTO, appsink_data: AppSinkData | None = None, reset_cb: Callable | None = None, merge_pip: bool|None = None, target_size: tuple[int, int]|None = None, ) -> None: """Initializes a new WebRTC instance. @param bridge: An instance of backend bridge. @param profile: Libervia profile. @param sources: Which kind of source to use. @param sinks: Which kind of sinks to use. @param appsink_data: configuration data for appsink (when SINKS_APP is used). Must not be used for other sinks. @param reset_cb: An optional Callable that is triggered on reset events. Can be used to reset UI data on new calls. @param merge_pip: A boolean flag indicating whether Picture-in-Picture mode is enabled. When PiP is used, local feedback is merged to remote video stream. Only one video stream is then produced (the local one). If None, PiP mode is selected automatically according to selected sink (it's used for SINKS_AUTO only for now). @param target_size: Expected size of the final sink stream. Mainly use by composer when ``merge_pip`` is set. None to autodetect (not real autodetection implemeted yet, default to (1280,720)). """ self.main_loop = asyncio.get_event_loop() self.bridge = bridge self.profile = profile self.pipeline = None self._audio_muted = False self._video_muted = False self._desktop_sharing = False self.desktop_sharing_data = None self.sources = sources self.sinks = sinks if target_size is None: target_size=(1280, 720) self.target_width, self.target_height = target_size if merge_pip is None: merge_pip = sinks == SINKS_AUTO self.merge_pip = merge_pip if sinks == SINKS_APP: if ( merge_pip and appsink_data is not None and appsink_data.remote_video_cb is not None ): raise ValueError("Remote_video_cb can't be used when merge_pip is used!") self.appsink_data = appsink_data elif appsink_data is not None: raise exceptions.InternalError( "appsink_data can only be used for SINKS_APP sinks" ) self.reset_cb = reset_cb if current_server == display_servers.WAYLAND: self.desktop_portal = DesktopPortal(self) else: self.desktop_portal = None self.reset_instance() @property def audio_muted(self): return self._audio_muted @audio_muted.setter def audio_muted(self, muted: bool) -> None: if muted != self._audio_muted: self._audio_muted = muted self.on_audio_mute(muted) @property def video_muted(self): return self._video_muted @video_muted.setter def video_muted(self, muted: bool) -> None: if muted != self._video_muted: self._video_muted = muted self.on_video_mute(muted) @property def desktop_sharing(self): return self._desktop_sharing @desktop_sharing.setter def desktop_sharing(self, active: bool) -> None: if active != self._desktop_sharing: self._desktop_sharing = active self.on_desktop_switch(active) try: cb = self.bindings["desktop_sharing"] except KeyError: pass else: cb(active) @property def sdp_set(self): return self._sdp_set @sdp_set.setter def sdp_set(self, is_set: bool): self._sdp_set = is_set if is_set: self.on_ice_candidates_new(self.remote_candidates_buffer) for data in self.remote_candidates_buffer.values(): data["candidates"].clear() @property def media_types(self): if self._media_types is None: raise Exception("self._media_types should not be None!") return self._media_types @media_types.setter def media_types(self, new_media_types: dict) -> None: self._media_types = new_media_types self._media_types_inv = {v: k for k, v in new_media_types.items()} @property def media_types_inv(self) -> dict: if self._media_types_inv is None: raise Exception("self._media_types_inv should not be None!") return self._media_types_inv def bind(self, **kwargs: Callable) -> None: self.bindings.clear() for key, cb in kwargs.items(): if key not in ("desktop_sharing",): raise ValueError( 'Only "desktop_sharing" is currently allowed for binding' ) self.bindings[key] = cb def generate_dot_file( self, filename: str = "pipeline", details: Gst.DebugGraphDetails = Gst.DebugGraphDetails.ALL, with_timestamp: bool = True, bin_: Gst.Bin|None = None, ) -> None: """Generate Dot File for debugging ``GST_DEBUG_DUMP_DOT_DIR`` environment variable must be set to destination dir. ``dot -Tpng -o <filename>.png <filename>.dot`` can be use to convert to a PNG file. See https://gstreamer.freedesktop.org/documentation/gstreamer/debugutils.html?gi-language=python#GstDebugGraphDetails for details. @param filename: name of the generated file @param details: which details to print @param with_timestamp: if True, add a timestamp to filename @param bin_: which bin to output. By default, the whole pipeline (``self.pipeline``) will be used. """ if bin_ is None: bin_ = self.pipeline if with_timestamp: timestamp = datetime.now().isoformat(timespec='milliseconds') filename = f"{timestamp}_filename" Gst.debug_bin_to_dot_file(bin_, details, filename) def get_sdp_mline_index(self, media_type: str) -> int: """Gets the sdpMLineIndex for a given media type. @param media_type: The type of the media. """ for index, m_type in self.media_types.items(): if m_type == media_type: return index raise ValueError(f"Media type '{media_type}' not found") def _set_media_types(self, offer_sdp: str) -> None: """Sets media types from offer SDP @param offer: RTC session description containing the offer """ sdp_lines = offer_sdp.splitlines() media_types = {} mline_index = 0 for line in sdp_lines: if line.startswith("m="): media_types[mline_index] = line[2 : line.find(" ")] mline_index += 1 self.media_types = media_types def _a_call(self, method, *args, **kwargs): """Call an async method in main thread""" aio.run_from_thread(method, *args, **kwargs, loop=self.main_loop) def get_payload_types( self, sdpmsg, video_encoding: str, audio_encoding: str ) -> dict[str, int | None]: """Find the payload types for the specified video and audio encoding. Very simplistically finds the first payload type matching the encoding name. More complex applications will want to match caps on profile-level-id, packetization-mode, etc. """ # method coming from gstreamer example (Matthew Waters, Nirbheek Chauhan) at # subprojects/gst-examples/webrtc/sendrecv/gst/webrtc_sendrecv.py video_pt = None audio_pt = None for i in range(0, sdpmsg.medias_len()): media = sdpmsg.get_media(i) for j in range(0, media.formats_len()): fmt = media.get_format(j) if fmt == "webrtc-datachannel": continue pt = int(fmt) caps = media.get_caps_from_media(pt) s = caps.get_structure(0) encoding_name = s["encoding-name"] if video_pt is None and encoding_name == video_encoding: video_pt = pt elif audio_pt is None and encoding_name == audio_encoding: audio_pt = pt return {video_encoding: video_pt, audio_encoding: audio_pt} def parse_ice_candidate(self, candidate_string): """Parses the ice candidate string. @param candidate_string: The ice candidate string to be parsed. """ pattern = re.compile( r"candidate:(?P<foundation>\S+) (?P<component_id>\d+) (?P<transport>\S+) " r"(?P<priority>\d+) (?P<address>\S+) (?P<port>\d+) typ " r"(?P<type>\S+)(?: raddr (?P<rel_addr>\S+) rport " r"(?P<rel_port>\d+))?(?: generation (?P<generation>\d+))?" ) match = pattern.match(candidate_string) if match: candidate_dict = match.groupdict() # Apply the correct types to the dictionary values candidate_dict["component_id"] = int(candidate_dict["component_id"]) candidate_dict["priority"] = int(candidate_dict["priority"]) candidate_dict["port"] = int(candidate_dict["port"]) if candidate_dict["rel_port"]: candidate_dict["rel_port"] = int(candidate_dict["rel_port"]) if candidate_dict["generation"]: candidate_dict["generation"] = candidate_dict["generation"] # Remove None values return {k: v for k, v in candidate_dict.items() if v is not None} else: log.warning(f"can't parse candidate: {candidate_string!r}") return None def build_ice_candidate(self, parsed_candidate): """Builds ICE candidate @param parsed_candidate: Dictionary containing parsed ICE candidate """ base_format = ( "candidate:{foundation} {component_id} {transport} {priority} " "{address} {port} typ {type}" ) if parsed_candidate.get("rel_addr") and parsed_candidate.get("rel_port"): base_format += " raddr {rel_addr} rport {rel_port}" if parsed_candidate.get("generation"): base_format += " generation {generation}" return base_format.format(**parsed_candidate) def extract_ufrag_pwd(self, sdp: str) -> tuple[str, str]: """Retrieves ICE password and user fragment for SDP offer. @param sdp: The Session Description Protocol offer string. @return: ufrag and pwd @raise ValueError: Can't extract ufrag and password """ ufrag_line = re.search(r"ice-ufrag:(\S+)", sdp) pwd_line = re.search(r"ice-pwd:(\S+)", sdp) if ufrag_line and pwd_line: ufrag = self.ufrag = ufrag_line.group(1) pwd = self.pwd = pwd_line.group(1) return ufrag, pwd else: log.error(f"SDP with missing ice-ufrag or ice-pwd:\n{sdp}") raise ValueError("Can't extract ice-ufrag and ice-pwd from SDP") def reset_instance(self): """Inits or resets the instance variables to their default state.""" self.role: str | None = None if self.pipeline is not None: self.pipeline.set_state(Gst.State.NULL) self.pipeline = None self._remote_video_pad = None self.sid: str | None = None self.offer: str | None = None self.local_candidates_buffer = {} self.ufrag: str | None = None self.pwd: str | None = None self.callee: str | None = None self._media_types = None self._media_types_inv = None self._sdp_set: bool = False self.remote_candidates_buffer: dict[str, dict[str, list]] = { "audio": {"candidates": []}, "video": {"candidates": []}, } self._media_types = None self._media_types_inv = None self.audio_valve = None self.video_valve = None if self.desktop_portal is not None: self.desktop_portal.end_screenshare() self.desktop_sharing = False self.desktop_sink_pad = None self.bindings = {} if self.reset_cb is not None: self.reset_cb() async def setup_call( self, role: str, audio_pt: int | None = 96, video_pt: int | None = 97, ) -> None: """Sets up the call. This method establishes the Gstreamer pipeline for audio and video communication. The method also manages STUN and TURN server configurations, signal watchers, and various connection handlers for the webrtcbin. @param role: The role for the call, either 'initiator' or 'responder'. @param audio_pt: The payload type for the audio stream. @param video_pt: The payload type for the video stream @raises NotImplementedError: If audio_pt or video_pt is set to None. @raises AssertionError: If the role is not 'initiator' or 'responder'. """ assert role in ("initiator", "responder") self.role = role if audio_pt is None or video_pt is None: raise NotImplementedError("None value is not handled yet") if self.sources == SOURCES_AUTO: video_source_elt = "v4l2src" audio_source_elt = "pulsesrc" elif self.sources == SOURCES_TEST: video_source_elt = "videotestsrc is-live=true pattern=ball" audio_source_elt = "audiotestsrc" else: raise exceptions.InternalError(f'Unknown "sources" value: {self.sources!r}') if self.sinks == SINKS_APP: local_video_sink_elt = ( "appsink name=local_video_sink emit-signals=true drop=true max-buffers=1 " "sync=True" ) elif self.sinks == SINKS_AUTO: local_video_sink_elt = "autovideosink" else: raise exceptions.InternalError(f"Unknown sinks value {self.sinks!r}") if self.merge_pip: extra_elt = ( "compositor name=compositor background=black " f"! video/x-raw,width={self.target_width},height={self.target_height}," "framerate=30/1 " f"! {local_video_sink_elt}" ) local_video_sink_elt = "compositor.sink_1" else: extra_elt = "" self.gst_pipe_desc = f""" webrtcbin latency=100 name=sendrecv bundle-policy=max-bundle input-selector name=video_selector ! videorate ! video/x-raw,framerate=30/1 ! tee name=t {extra_elt} {video_source_elt} name=video_src ! queue leaky=downstream ! video_selector. videotestsrc name=muted_src is-live=true pattern=black ! queue leaky=downstream ! video_selector. t. ! queue max-size-buffers=5 max-size-time=0 max-size-bytes=0 leaky=downstream ! videoconvert ! vp8enc deadline=1 keyframe-max-dist=60 ! rtpvp8pay picture-id-mode=15-bit ! application/x-rtp,media=video,encoding-name=VP8,payload={video_pt} ! sendrecv. t. ! queue max-size-buffers=5 max-size-time=0 max-size-bytes=0 leaky=downstream ! videoconvert ! {local_video_sink_elt} {audio_source_elt} name=audio_src ! valve ! queue max-size-buffers=10 max-size-time=0 max-size-bytes=0 leaky=downstream ! audioconvert ! audioresample ! opusenc audio-type=voice ! rtpopuspay ! application/x-rtp,media=audio,encoding-name=OPUS,payload={audio_pt} ! sendrecv. """ log.debug(f"Gstreamer pipeline: {self.gst_pipe_desc}") # Create the pipeline try: self.pipeline = Gst.parse_launch(self.gst_pipe_desc) except Exception: log.exception("Can't parse pipeline") self.pipeline = None if not self.pipeline: raise exceptions.InternalError("Failed to create Gstreamer pipeline.") self.webrtcbin = self.pipeline.get_by_name("sendrecv") self.video_src = self.pipeline.get_by_name("video_src") self.muted_src = self.pipeline.get_by_name("muted_src") self.video_selector = self.pipeline.get_by_name("video_selector") self.audio_valve = self.pipeline.get_by_name("audio_valve") if self.video_muted: self.on_video_mute(True) if self.audio_muted: self.on_audio_mute(True) # set STUN and TURN servers external_disco = data_format.deserialise( await self.bridge.external_disco_get("", self.profile), type_check=list ) for server in external_disco: if server["type"] == "stun": if server["transport"] == "tcp": log.info( "ignoring TCP STUN server, GStreamer only support one STUN server" ) url = f"stun://{server['host']}:{server['port']}" log.debug(f"adding stun server: {url}") self.webrtcbin.set_property("stun-server", url) elif server["type"] == "turn": url = "{scheme}://{username}:{password}@{host}:{port}".format( scheme="turns" if server["transport"] == "tcp" else "turn", username=quote_plus(server["username"]), password=quote_plus(server["password"]), host=server["host"], port=server["port"], ) log.debug(f"adding turn server: {url}") if not self.webrtcbin.emit("add-turn-server", url): log.warning(f"Erreur while adding TURN server {url}") # local video feedback if self.sinks == SINKS_APP: assert self.appsink_data is not None local_video_sink = self.pipeline.get_by_name("local_video_sink") local_video_sink.set_property("emit-signals", True) local_video_sink.connect("new-sample", self.appsink_data.local_video_cb) local_video_sink_caps = Gst.Caps.from_string(f"video/x-raw,format=RGB") local_video_sink.set_property("caps", local_video_sink_caps) # Create bus and associate signal watchers self.bus = self.pipeline.get_bus() if not self.bus: log.error("Failed to get bus from pipeline.") return self.bus.add_signal_watch() self.webrtcbin.connect("pad-added", self.on_pad_added) self.bus.connect("message::error", self.on_bus_error) self.bus.connect("message::eos", self.on_bus_eos) self.webrtcbin.connect("on-negotiation-needed", self.on_negotiation_needed) self.webrtcbin.connect("on-ice-candidate", self.on_ice_candidate) self.webrtcbin.connect( "notify::ice-gathering-state", self.on_ice_gathering_state_change ) self.webrtcbin.connect( "notify::ice-connection-state", self.on_ice_connection_state ) def start_pipeline(self) -> None: """Starts the GStreamer pipeline.""" log.debug("starting the pipeline") self.pipeline.set_state(Gst.State.PLAYING) def on_negotiation_needed(self, webrtc): """Initiate SDP offer when negotiation is needed.""" log.debug("Negotiation needed.") if self.role == "initiator": log.debug("Creating offer…") promise = Gst.Promise.new_with_change_func(self.on_offer_created) self.webrtcbin.emit("create-offer", None, promise) def on_offer_created(self, promise): """Callback for when SDP offer is created.""" log.info("on_offer_created called") assert promise.wait() == Gst.PromiseResult.REPLIED reply = promise.get_reply() if reply is None: log.error("Promise reply is None. Offer creation might have failed.") return offer = reply["offer"] self.offer = offer.sdp.as_text() log.info(f"SDP offer created: \n{self.offer}") self._set_media_types(self.offer) promise = Gst.Promise.new() self.webrtcbin.emit("set-local-description", offer, promise) promise.interrupt() self._a_call(self._start_call) def on_answer_set(self, promise): assert promise.wait() == Gst.PromiseResult.REPLIED def on_answer_created(self, promise, _, __): """Callback for when SDP answer is created.""" assert promise.wait() == Gst.PromiseResult.REPLIED reply = promise.get_reply() answer = reply["answer"] promise = Gst.Promise.new() self.webrtcbin.emit("set-local-description", answer, promise) promise.interrupt() answer_sdp = answer.sdp.as_text() log.info(f"SDP answer set: \n{answer_sdp}") self.sdp_set = True self._a_call(self.bridge.call_answer_sdp, self.sid, answer_sdp, self.profile) def on_offer_set(self, promise): assert promise.wait() == Gst.PromiseResult.REPLIED promise = Gst.Promise.new_with_change_func(self.on_answer_created, None, None) self.webrtcbin.emit("create-answer", None, promise) def link_element_or_pad( self, source: Gst.Element, dest: Gst.Element | Gst.Pad ) -> bool: """Check if dest is a pad or an element, and link appropriately""" src_pad = source.get_static_pad("src") if isinstance(dest, Gst.Pad): # If the dest is a pad, link directly if not src_pad.link(dest) == Gst.PadLinkReturn.OK: log.error( "Failed to link 'conv' to the compositor's newly requested pad!" ) return False elif isinstance(dest, Gst.Element): return source.link(dest) else: log.error(f"Unexpected type for dest: {type(sink)}") return False return True def scaled_dimensions( self, original_width: int, original_height: int, max_width: int, max_height: int ) -> tuple[int, int]: """Calculates the scaled dimensions preserving aspect ratio. @param original_width: Original width of the video stream. @param original_height: Original height of the video stream. @param max_width: Maximum desired width for the scaled video. @param max_height: Maximum desired height for the scaled video. @return: The width and height of the scaled video. """ aspect_ratio = original_width / original_height new_width = int(max_height * aspect_ratio) if new_width <= max_width: return new_width, max_height new_height = int(max_width / aspect_ratio) return max_width, new_height def on_remote_decodebin_stream(self, _, pad: Gst.Pad) -> None: """Handle the stream from the remote decodebin. This method processes the incoming stream from the remote decodebin, determining whether it's video or audio. It then sets up the appropriate GStreamer elements for video/audio processing and adds them to the pipeline. @param pad: The Gst.Pad from the remote decodebin producing the stream. """ assert self.pipeline is not None if not pad.has_current_caps(): log.error(f"{pad} has no caps, ignoring") return caps = pad.get_current_caps() assert len(caps) s = caps[0] name = s.get_name() log.debug(f"====> NAME START: {name}") q = Gst.ElementFactory.make("queue") if name.startswith("video"): log.debug("===> VIDEO OK") self._remote_video_pad = pad # Check and log the original size of the video width = self.target_width height = self.target_height log.info(f"Original video size: {width}x{height}") # This is a fix for an issue found with Movim on desktop: a non standard # resolution is used (990x557) resulting in bad alignement and no color in # rendered image adjust_resolution = width % 4 != 0 or height % 4 != 0 if adjust_resolution: log.info("non standard resolution, we need to adjust size") width = (width + 3) // 4 * 4 height = (height + 3) // 4 * 4 log.info(f"Adjusted video size: {width}x{height}") conv = Gst.ElementFactory.make("videoconvert") if self.merge_pip: # with ``merge_pip`` set, we plug the remote stream to the composer compositor = self.pipeline.get_by_name("compositor") sink1_pad = compositor.get_static_pad("sink_1") local_width, local_height = self.scaled_dimensions( sink1_pad.get_property("width"), sink1_pad.get_property("height"), width // 3, height // 3, ) sink1_pad.set_property("xpos", width - local_width) sink1_pad.set_property("ypos", height - local_height) sink1_pad.set_property("width", local_width) sink1_pad.set_property("height", local_height) sink1_pad.set_property("sizing-policy", 1) sink1_pad.set_property("zorder", 1) # Request a new pad for the remote stream sink_pad_template = compositor.get_pad_template("sink_%u") remote_video_sink = compositor.request_pad(sink_pad_template, None, None) remote_video_sink.set_property("zorder", 0) remote_video_sink.set_property("width", width) remote_video_sink.set_property("height", height) remote_video_sink.set_property("sizing-policy", 1) elif self.sinks == SINKS_APP: # ``app`` sink without ``self.merge_pip`` set, be create the sink and # connect it to the ``remote_video_cb``. assert self.appsink_data is not None remote_video_sink = Gst.ElementFactory.make("appsink") remote_video_caps = Gst.Caps.from_string("video/x-raw,format=RGB") remote_video_sink.set_property("caps", remote_video_caps) remote_video_sink.set_property("emit-signals", True) remote_video_sink.set_property("drop", True) remote_video_sink.set_property("max-buffers", 1) remote_video_sink.set_property("sync", True) remote_video_sink.connect("new-sample", self.appsink_data.remote_video_cb) self.pipeline.add(remote_video_sink) elif self.sinks == SINKS_AUTO: # if ``self.merge_pip`` is not set, we create a dedicated # ``autovideosink`` for remote stream. remote_video_sink = Gst.ElementFactory.make("autovideosink") self.pipeline.add(remote_video_sink) else: raise exceptions.InternalError(f'Unhandled "sinks" value: {self.sinks!r}') if adjust_resolution: videoscale = Gst.ElementFactory.make("videoscale") adjusted_caps = Gst.Caps.from_string( f"video/x-raw,width={width},height={height}" ) capsfilter = Gst.ElementFactory.make("capsfilter") capsfilter.set_property("caps", adjusted_caps) self.pipeline.add(q, conv, videoscale, capsfilter) self.pipeline.sync_children_states() ret = pad.link(q.get_static_pad("sink")) if ret != Gst.PadLinkReturn.OK: log.error(f"Error linking pad: {ret}") q.link(conv) conv.link(videoscale) videoscale.link(capsfilter) self.link_element_or_pad(capsfilter.link, remote_video_sink) else: self.pipeline.add(q, conv) self.pipeline.sync_children_states() ret = pad.link(q.get_static_pad("sink")) if ret != Gst.PadLinkReturn.OK: log.error(f"Error linking pad: {ret}") q.link(conv) self.link_element_or_pad(conv, remote_video_sink) elif name.startswith("audio"): log.debug("===> Audio OK") conv = Gst.ElementFactory.make("audioconvert") resample = Gst.ElementFactory.make("audioresample") remote_audio_sink = Gst.ElementFactory.make("autoaudiosink") self.pipeline.add(q, conv, resample, remote_audio_sink) self.pipeline.sync_children_states() ret = pad.link(q.get_static_pad("sink")) if ret != Gst.PadLinkReturn.OK: log.error(f"Error linking pad: {ret}") q.link(conv) conv.link(resample) resample.link(remote_audio_sink) else: log.warning(f"unmanaged name: {name!r}") def on_pad_added(self, __, pad: Gst.Pad) -> None: """Handle the addition of a new pad to the element. When a new source pad is added to the element, this method creates a decodebin, connects it to handle the stream, and links the pad to the decodebin. @param __: Placeholder for the signal source. Not used in this method. @param pad: The newly added pad. """ log.debug("on_pad_added") if pad.direction != Gst.PadDirection.SRC: return decodebin = Gst.ElementFactory.make("decodebin") decodebin.connect("pad-added", self.on_remote_decodebin_stream) self.pipeline.add(decodebin) decodebin.sync_state_with_parent() pad.link(decodebin.get_static_pad("sink")) async def _start_call(self) -> None: """Initiate the call. Initiates a call with the callee using the stored offer. If there are any buffered local ICE candidates, they are sent as part of the initiation. """ assert self.callee self.sid = await self.bridge.call_start( str(self.callee), data_format.serialise({"sdp": self.offer}), self.profile ) if self.local_candidates_buffer: log.debug( f"sending buffered local ICE candidates: {self.local_candidates_buffer}" ) if self.pwd is None: sdp = self.webrtcbin.props.local_description.sdp.as_text() self.extract_ufrag_pwd(sdp) ice_data = {} for media_type, candidates in self.local_candidates_buffer.items(): ice_data[media_type] = { "ufrag": self.ufrag, "pwd": self.pwd, "candidates": candidates, } await self.bridge.ice_candidates_add( self.sid, data_format.serialise(ice_data), self.profile ) self.local_candidates_buffer.clear() def _remote_sdp_set(self, promise) -> None: assert promise.wait() == Gst.PromiseResult.REPLIED self.sdp_set = True def on_accepted_call(self, sdp: str, profile: str) -> None: """Outgoing call has been accepted. @param sdp: The SDP answer string received from the other party. @param profile: Profile used for the call. """ log.debug(f"SDP answer received: \n{sdp}") __, sdpmsg = GstSdp.SDPMessage.new_from_text(sdp) answer = GstWebRTC.WebRTCSessionDescription.new( GstWebRTC.WebRTCSDPType.ANSWER, sdpmsg ) promise = Gst.Promise.new_with_change_func(self._remote_sdp_set) self.webrtcbin.emit("set-remote-description", answer, promise) async def answer_call(self, sdp: str, profile: str) -> None: """Answer an incoming call @param sdp: The SDP offer string received from the initiator. @param profile: Profile used for the call. @raise AssertionError: Raised when either "VP8" or "OPUS" is not present in payload types. """ log.debug(f"SDP offer received: \n{sdp}") self._set_media_types(sdp) __, offer_sdp_msg = GstSdp.SDPMessage.new_from_text(sdp) payload_types = self.get_payload_types( offer_sdp_msg, video_encoding="VP8", audio_encoding="OPUS" ) assert "VP8" in payload_types assert "OPUS" in payload_types await self.setup_call( "responder", audio_pt=payload_types["OPUS"], video_pt=payload_types["VP8"] ) self.start_pipeline() offer = GstWebRTC.WebRTCSessionDescription.new( GstWebRTC.WebRTCSDPType.OFFER, offer_sdp_msg ) promise = Gst.Promise.new_with_change_func(self.on_offer_set) self.webrtcbin.emit("set-remote-description", offer, promise) def on_ice_candidate(self, webrtc, mline_index, candidate_sdp): """Handles the on-ice-candidate signal of webrtcbin. @param webrtc: The webrtcbin element. @param mlineindex: The mline index. @param candidate: The ICE candidate. """ log.debug( f"Local ICE candidate. MLine Index: {mline_index}, Candidate: {candidate_sdp}" ) parsed_candidate = self.parse_ice_candidate(candidate_sdp) try: media_type = self.media_types[mline_index] except KeyError: raise exceptions.InternalError("can't find media type") if self.sid is None: log.debug("buffering local ICE candidate") self.local_candidates_buffer.setdefault(media_type, []).append( parsed_candidate ) else: sdp = self.webrtcbin.props.local_description.sdp.as_text() assert sdp is not None ufrag, pwd = self.extract_ufrag_pwd(sdp) ice_data = {"ufrag": ufrag, "pwd": pwd, "candidates": [parsed_candidate]} self._a_call( self.bridge.ice_candidates_add, self.sid, data_format.serialise({media_type: ice_data}), self.profile, ) def on_ice_candidates_new(self, candidates: dict) -> None: """Handle new ICE candidates. @param candidates: A dictionary containing media types ("audio" or "video") as keys and corresponding ICE data as values. @raise exceptions.InternalError: Raised when sdp mline index is not found. """ if not self.sdp_set: log.debug("buffering remote ICE candidate") for media_type in ("audio", "video"): media_candidates = candidates.get(media_type) if media_candidates: buffer = self.remote_candidates_buffer[media_type] buffer["candidates"].extend(media_candidates["candidates"]) return for media_type, ice_data in candidates.items(): for candidate in ice_data["candidates"]: candidate_sdp = self.build_ice_candidate(candidate) try: mline_index = self.get_sdp_mline_index(media_type) except Exception as e: raise exceptions.InternalError(f"Can't find sdp mline index: {e}") self.webrtcbin.emit("add-ice-candidate", mline_index, candidate_sdp) log.debug( f"Remote ICE candidate added. MLine Index: {mline_index}, " f"{candidate_sdp}" ) def on_ice_gathering_state_change(self, pspec, __): state = self.webrtcbin.get_property("ice-gathering-state") log.debug(f"ICE gathering state changed to {state}") def on_ice_connection_state(self, pspec, __): state = self.webrtcbin.props.ice_connection_state if state == GstWebRTC.WebRTCICEConnectionState.FAILED: log.error("ICE connection failed") log.info(f"ICE connection state changed to {state}") def on_bus_error(self, bus: Gst.Bus, message: Gst.Message) -> None: """Handles the GStreamer bus error messages. @param bus: The GStreamer bus. @param message: The error message. """ err, debug = message.parse_error() log.error(f"Error from {message.src.get_name()}: {err.message}") log.error(f"Debugging info: {debug}") def on_bus_eos(self, bus: Gst.Bus, message: Gst.Message) -> None: """Handles the GStreamer bus eos messages. @param bus: The GStreamer bus. @param message: The eos message. """ log.info("End of stream") def on_audio_mute(self, muted: bool) -> None: """Handles (un)muting of audio. @param muted: True if audio is muted. """ if self.audio_valve is not None: self.audio_valve.set_property("drop", muted) state = "muted" if muted else "unmuted" log.info(f"audio is now {state}") def on_video_mute(self, muted: bool) -> None: """Handles (un)muting of video. @param muted: True if video is muted. """ if self.video_selector is not None: current_source = None if muted else "desktop" if self.desktop_sharing else "video" self.switch_video_source(current_source) state = "muted" if muted else "unmuted" log.info(f"Video is now {state}") def on_desktop_switch(self, desktop_active: bool) -> None: """Switches the video source between desktop and video. @param desktop_active: True if desktop must be active. False for video. """ if desktop_active and self.desktop_portal is not None: aio.run_async(self.on_desktop_switch_portal(desktop_active)) else: self.do_desktop_switch(desktop_active) async def on_desktop_switch_portal(self, desktop_active: bool) -> None: """Call freedesktop screenshare portal and the activate the shared stream""" assert self.desktop_portal is not None try: screenshare_data = await self.desktop_portal.request_screenshare() except exceptions.CancelError: self.desktop_sharing = False return self.desktop_sharing_data = { "path": str(screenshare_data["node_id"]) } self.do_desktop_switch(desktop_active) def do_desktop_switch(self, desktop_active: bool) -> None: if self.video_muted: # Update the active source state but do not switch self.desktop_sharing = desktop_active return source = "desktop" if desktop_active else "video" self.switch_video_source(source) self.desktop_sharing = desktop_active def switch_video_source(self, source: str|None) -> None: """Activates the specified source while deactivating the others. @param source: 'desktop', 'video', 'muted' or None for muted source. """ if source is None: source = "muted" if source not in ["video", "muted", "desktop"]: raise ValueError( f"Invalid source: {source!r}, use one of {'video', 'muted', 'desktop'}" ) self.pipeline.set_state(Gst.State.PAUSED) # Create a new desktop source if necessary if source == "desktop": self._setup_desktop_source(self.desktop_sharing_data) # Activate the chosen source and deactivate the others for src_name in ["video", "muted", "desktop"]: src_element = self.pipeline.get_by_name(f"{src_name}_src") if src_name == source: if src_element: src_element.set_state(Gst.State.PLAYING) else: if src_element: if src_name == "desktop": self._remove_desktop_source(src_element) else: src_element.set_state(Gst.State.NULL) # Set the video_selector active pad if source == "desktop": if self.desktop_sink_pad: pad = self.desktop_sink_pad else: log.error(f"No desktop pad available") pad = None else: pad_name = f"sink_{['video', 'muted'].index(source)}" pad = self.video_selector.get_static_pad(pad_name) if pad is not None: self.video_selector.props.active_pad = pad self.pipeline.set_state(Gst.State.PLAYING) def _setup_desktop_source(self, properties: dict[str, object]|None) -> None: """Set up a new desktop source. @param properties: The properties to set on the desktop source. """ source_elt = "ximagesrc" if self.desktop_portal is None else "pipewiresrc" desktop_src = Gst.ElementFactory.make(source_elt, "desktop_src") if properties is None: properties = {} for key, value in properties.items(): log.debug(f"setting {source_elt} property: {key!r}={value!r}") desktop_src.set_property(key, value) video_convert = Gst.ElementFactory.make("videoconvert", "desktop_videoconvert") queue = Gst.ElementFactory.make("queue", "desktop_queue") queue.set_property("leaky", "downstream") self.pipeline.add(desktop_src) self.pipeline.add(video_convert) self.pipeline.add(queue) desktop_src.link(video_convert) video_convert.link(queue) sink_pad_template = self.video_selector.get_pad_template("sink_%u") self.desktop_sink_pad = self.video_selector.request_pad(sink_pad_template, None, None) queue_src_pad = queue.get_static_pad("src") queue_src_pad.link(self.desktop_sink_pad) desktop_src.sync_state_with_parent() video_convert.sync_state_with_parent() queue.sync_state_with_parent() def _remove_desktop_source(self, desktop_src: Gst.Element) -> None: """Remove the desktop source from the pipeline. @param desktop_src: The desktop source to remove. """ # Remove elements for the desktop source video_convert = self.pipeline.get_by_name("desktop_videoconvert") queue = self.pipeline.get_by_name("desktop_queue") if video_convert: video_convert.set_state(Gst.State.NULL) desktop_src.unlink(video_convert) self.pipeline.remove(video_convert) if queue: queue.set_state(Gst.State.NULL) self.pipeline.remove(queue) desktop_src.set_state(Gst.State.NULL) self.pipeline.remove(desktop_src) # Release the pad associated with the desktop source if self.desktop_sink_pad: self.video_selector.release_request_pad(self.desktop_sink_pad) self.desktop_sink_pad = None if self.desktop_portal is not None: self.desktop_portal.end_screenshare() async def end_call(self) -> None: """Stop streaming and clean instance""" self.reset_instance()