webrtc/talk/media/base/fakemediaengine.h

1046 lines
33 KiB
C
Raw Normal View History

/*
* libjingle
* Copyright 2004 Google Inc.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
* 3. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
* EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
* OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
* OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
* ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef TALK_MEDIA_BASE_FAKEMEDIAENGINE_H_
#define TALK_MEDIA_BASE_FAKEMEDIAENGINE_H_
#include <list>
#include <map>
#include <set>
#include <string>
#include <vector>
#include "talk/base/buffer.h"
#include "talk/base/stringutils.h"
#include "talk/media/base/audiorenderer.h"
#include "talk/media/base/mediaengine.h"
#include "talk/media/base/rtputils.h"
#include "talk/media/base/streamparams.h"
#include "talk/p2p/base/sessiondescription.h"
namespace cricket {
class FakeMediaEngine;
class FakeVideoEngine;
class FakeVoiceEngine;
// A common helper class that handles sending and receiving RTP/RTCP packets.
template <class Base> class RtpHelper : public Base {
public:
RtpHelper()
: sending_(false),
playout_(false),
fail_set_send_codecs_(false),
fail_set_recv_codecs_(false),
send_ssrc_(0),
ready_to_send_(false) {}
const std::vector<RtpHeaderExtension>& recv_extensions() {
return recv_extensions_;
}
const std::vector<RtpHeaderExtension>& send_extensions() {
return send_extensions_;
}
bool sending() const { return sending_; }
bool playout() const { return playout_; }
const std::list<std::string>& rtp_packets() const { return rtp_packets_; }
const std::list<std::string>& rtcp_packets() const { return rtcp_packets_; }
bool SendRtp(const void* data, int len) {
if (!sending_) {
return false;
}
talk_base::Buffer packet(data, len, kMaxRtpPacketLen);
return Base::SendPacket(&packet);
}
bool SendRtcp(const void* data, int len) {
talk_base::Buffer packet(data, len, kMaxRtpPacketLen);
return Base::SendRtcp(&packet);
}
bool CheckRtp(const void* data, int len) {
bool success = !rtp_packets_.empty();
if (success) {
std::string packet = rtp_packets_.front();
rtp_packets_.pop_front();
success = (packet == std::string(static_cast<const char*>(data), len));
}
return success;
}
bool CheckRtcp(const void* data, int len) {
bool success = !rtcp_packets_.empty();
if (success) {
std::string packet = rtcp_packets_.front();
rtcp_packets_.pop_front();
success = (packet == std::string(static_cast<const char*>(data), len));
}
return success;
}
bool CheckNoRtp() { return rtp_packets_.empty(); }
bool CheckNoRtcp() { return rtcp_packets_.empty(); }
virtual bool SetRecvRtpHeaderExtensions(
const std::vector<RtpHeaderExtension>& extensions) {
recv_extensions_ = extensions;
return true;
}
virtual bool SetSendRtpHeaderExtensions(
const std::vector<RtpHeaderExtension>& extensions) {
send_extensions_ = extensions;
return true;
}
void set_fail_set_send_codecs(bool fail) { fail_set_send_codecs_ = fail; }
void set_fail_set_recv_codecs(bool fail) { fail_set_recv_codecs_ = fail; }
virtual bool AddSendStream(const StreamParams& sp) {
if (std::find(send_streams_.begin(), send_streams_.end(), sp) !=
send_streams_.end()) {
return false;
}
send_streams_.push_back(sp);
return true;
}
virtual bool RemoveSendStream(uint32 ssrc) {
return RemoveStreamBySsrc(&send_streams_, ssrc);
}
virtual bool AddRecvStream(const StreamParams& sp) {
if (std::find(receive_streams_.begin(), receive_streams_.end(), sp) !=
receive_streams_.end()) {
return false;
}
receive_streams_.push_back(sp);
return true;
}
virtual bool RemoveRecvStream(uint32 ssrc) {
return RemoveStreamBySsrc(&receive_streams_, ssrc);
}
virtual bool MuteStream(uint32 ssrc, bool on) {
if (!HasSendStream(ssrc) && ssrc != 0)
return false;
if (on)
muted_streams_.insert(ssrc);
else
muted_streams_.erase(ssrc);
return true;
}
bool IsStreamMuted(uint32 ssrc) const {
bool ret = muted_streams_.find(ssrc) != muted_streams_.end();
// If |ssrc = 0| check if the first send stream is muted.
if (!ret && ssrc == 0 && !send_streams_.empty()) {
return muted_streams_.find(send_streams_[0].first_ssrc()) !=
muted_streams_.end();
}
return ret;
}
const std::vector<StreamParams>& send_streams() const {
return send_streams_;
}
const std::vector<StreamParams>& recv_streams() const {
return receive_streams_;
}
bool HasRecvStream(uint32 ssrc) const {
return GetStreamBySsrc(receive_streams_, ssrc, NULL);
}
bool HasSendStream(uint32 ssrc) const {
return GetStreamBySsrc(send_streams_, ssrc, NULL);
}
// TODO(perkj): This is to support legacy unit test that only check one
// sending stream.
const uint32 send_ssrc() {
if (send_streams_.empty())
return 0;
return send_streams_[0].first_ssrc();
}
// TODO(perkj): This is to support legacy unit test that only check one
// sending stream.
const std::string rtcp_cname() {
if (send_streams_.empty())
return "";
return send_streams_[0].cname;
}
bool ready_to_send() const {
return ready_to_send_;
}
protected:
bool set_sending(bool send) {
sending_ = send;
return true;
}
void set_playout(bool playout) { playout_ = playout; }
virtual void OnPacketReceived(talk_base::Buffer* packet,
const talk_base::PacketTime& packet_time) {
rtp_packets_.push_back(std::string(packet->data(), packet->length()));
}
virtual void OnRtcpReceived(talk_base::Buffer* packet,
const talk_base::PacketTime& packet_time) {
rtcp_packets_.push_back(std::string(packet->data(), packet->length()));
}
virtual void OnReadyToSend(bool ready) {
ready_to_send_ = ready;
}
bool fail_set_send_codecs() const { return fail_set_send_codecs_; }
bool fail_set_recv_codecs() const { return fail_set_recv_codecs_; }
private:
bool sending_;
bool playout_;
std::vector<RtpHeaderExtension> recv_extensions_;
std::vector<RtpHeaderExtension> send_extensions_;
std::list<std::string> rtp_packets_;
std::list<std::string> rtcp_packets_;
std::vector<StreamParams> send_streams_;
std::vector<StreamParams> receive_streams_;
std::set<uint32> muted_streams_;
bool fail_set_send_codecs_;
bool fail_set_recv_codecs_;
uint32 send_ssrc_;
std::string rtcp_cname_;
bool ready_to_send_;
};
class FakeVoiceMediaChannel : public RtpHelper<VoiceMediaChannel> {
public:
struct DtmfInfo {
DtmfInfo(uint32 ssrc, int event_code, int duration, int flags)
: ssrc(ssrc), event_code(event_code), duration(duration), flags(flags) {
}
uint32 ssrc;
int event_code;
int duration;
int flags;
};
explicit FakeVoiceMediaChannel(FakeVoiceEngine* engine)
: engine_(engine),
fail_set_send_(false),
ringback_tone_ssrc_(0),
ringback_tone_play_(false),
ringback_tone_loop_(false),
time_since_last_typing_(-1) {
output_scalings_[0] = OutputScaling(); // For default channel.
}
~FakeVoiceMediaChannel();
const std::vector<AudioCodec>& recv_codecs() const { return recv_codecs_; }
const std::vector<AudioCodec>& send_codecs() const { return send_codecs_; }
const std::vector<AudioCodec>& codecs() const { return send_codecs(); }
const std::vector<DtmfInfo>& dtmf_info_queue() const {
return dtmf_info_queue_;
}
const AudioOptions& options() const { return options_; }
uint32 ringback_tone_ssrc() const { return ringback_tone_ssrc_; }
bool ringback_tone_play() const { return ringback_tone_play_; }
bool ringback_tone_loop() const { return ringback_tone_loop_; }
virtual bool SetRecvCodecs(const std::vector<AudioCodec>& codecs) {
if (fail_set_recv_codecs()) {
// Fake the failure in SetRecvCodecs.
return false;
}
recv_codecs_ = codecs;
return true;
}
virtual bool SetSendCodecs(const std::vector<AudioCodec>& codecs) {
if (fail_set_send_codecs()) {
// Fake the failure in SetSendCodecs.
return false;
}
send_codecs_ = codecs;
return true;
}
virtual bool SetPlayout(bool playout) {
set_playout(playout);
return true;
}
virtual bool SetSend(SendFlags flag) {
if (fail_set_send_) {
return false;
}
return set_sending(flag != SEND_NOTHING);
}
virtual bool SetStartSendBandwidth(int bps) { return true; }
virtual bool SetMaxSendBandwidth(int bps) { return true; }
virtual bool AddRecvStream(const StreamParams& sp) {
if (!RtpHelper<VoiceMediaChannel>::AddRecvStream(sp))
return false;
output_scalings_[sp.first_ssrc()] = OutputScaling();
return true;
}
virtual bool RemoveRecvStream(uint32 ssrc) {
if (!RtpHelper<VoiceMediaChannel>::RemoveRecvStream(ssrc))
return false;
output_scalings_.erase(ssrc);
return true;
}
virtual bool SetRemoteRenderer(uint32 ssrc, AudioRenderer* renderer) {
std::map<uint32, AudioRenderer*>::iterator it =
remote_renderers_.find(ssrc);
if (renderer) {
if (it != remote_renderers_.end()) {
ASSERT(it->second == renderer);
} else {
remote_renderers_.insert(std::make_pair(ssrc, renderer));
renderer->AddChannel(0);
}
} else {
if (it != remote_renderers_.end()) {
it->second->RemoveChannel(0);
remote_renderers_.erase(it);
} else {
return false;
}
}
return true;
}
virtual bool SetLocalRenderer(uint32 ssrc, AudioRenderer* renderer) {
std::map<uint32, AudioRenderer*>::iterator it = local_renderers_.find(ssrc);
if (renderer) {
if (it != local_renderers_.end()) {
ASSERT(it->second == renderer);
} else {
local_renderers_.insert(std::make_pair(ssrc, renderer));
renderer->AddChannel(0);
}
} else {
if (it != local_renderers_.end()) {
it->second->RemoveChannel(0);
local_renderers_.erase(it);
} else {
return false;
}
}
return true;
}
virtual bool GetActiveStreams(AudioInfo::StreamList* streams) { return true; }
virtual int GetOutputLevel() { return 0; }
void set_time_since_last_typing(int ms) { time_since_last_typing_ = ms; }
virtual int GetTimeSinceLastTyping() { return time_since_last_typing_; }
virtual void SetTypingDetectionParameters(
int time_window, int cost_per_typing, int reporting_threshold,
int penalty_decay, int type_event_delay) {}
virtual bool SetRingbackTone(const char* buf, int len) { return true; }
virtual bool PlayRingbackTone(uint32 ssrc, bool play, bool loop) {
ringback_tone_ssrc_ = ssrc;
ringback_tone_play_ = play;
ringback_tone_loop_ = loop;
return true;
}
virtual bool CanInsertDtmf() {
for (std::vector<AudioCodec>::const_iterator it = send_codecs_.begin();
it != send_codecs_.end(); ++it) {
// Find the DTMF telephone event "codec".
if (_stricmp(it->name.c_str(), "telephone-event") == 0) {
return true;
}
}
return false;
}
virtual bool InsertDtmf(uint32 ssrc, int event_code, int duration,
int flags) {
dtmf_info_queue_.push_back(DtmfInfo(ssrc, event_code, duration, flags));
return true;
}
virtual bool SetOutputScaling(uint32 ssrc, double left, double right) {
if (0 == ssrc) {
std::map<uint32, OutputScaling>::iterator it;
for (it = output_scalings_.begin(); it != output_scalings_.end(); ++it) {
it->second.left = left;
it->second.right = right;
}
return true;
} else if (output_scalings_.find(ssrc) != output_scalings_.end()) {
output_scalings_[ssrc].left = left;
output_scalings_[ssrc].right = right;
return true;
}
return false;
}
virtual bool GetOutputScaling(uint32 ssrc, double* left, double* right) {
if (output_scalings_.find(ssrc) == output_scalings_.end())
return false;
*left = output_scalings_[ssrc].left;
*right = output_scalings_[ssrc].right;
return true;
}
virtual bool GetStats(VoiceMediaInfo* info) { return false; }
virtual void GetLastMediaError(uint32* ssrc,
VoiceMediaChannel::Error* error) {
*ssrc = 0;
*error = fail_set_send_ ? VoiceMediaChannel::ERROR_REC_DEVICE_OPEN_FAILED
: VoiceMediaChannel::ERROR_NONE;
}
void set_fail_set_send(bool fail) { fail_set_send_ = fail; }
void TriggerError(uint32 ssrc, VoiceMediaChannel::Error error) {
VoiceMediaChannel::SignalMediaError(ssrc, error);
}
virtual bool SetOptions(const AudioOptions& options) {
// Does a "merge" of current options and set options.
options_.SetAll(options);
return true;
}
virtual bool GetOptions(AudioOptions* options) const {
*options = options_;
return true;
}
private:
struct OutputScaling {
OutputScaling() : left(1.0), right(1.0) {}
double left, right;
};
FakeVoiceEngine* engine_;
std::vector<AudioCodec> recv_codecs_;
std::vector<AudioCodec> send_codecs_;
std::map<uint32, OutputScaling> output_scalings_;
std::vector<DtmfInfo> dtmf_info_queue_;
bool fail_set_send_;
uint32 ringback_tone_ssrc_;
bool ringback_tone_play_;
bool ringback_tone_loop_;
int time_since_last_typing_;
AudioOptions options_;
std::map<uint32, AudioRenderer*> local_renderers_;
std::map<uint32, AudioRenderer*> remote_renderers_;
};
// A helper function to compare the FakeVoiceMediaChannel::DtmfInfo.
inline bool CompareDtmfInfo(const FakeVoiceMediaChannel::DtmfInfo& info,
uint32 ssrc, int event_code, int duration,
int flags) {
return (info.duration == duration && info.event_code == event_code &&
info.flags == flags && info.ssrc == ssrc);
}
class FakeVideoMediaChannel : public RtpHelper<VideoMediaChannel> {
public:
explicit FakeVideoMediaChannel(FakeVideoEngine* engine)
: engine_(engine),
sent_intra_frame_(false),
requested_intra_frame_(false),
start_bps_(-1),
max_bps_(-1) {}
~FakeVideoMediaChannel();
const std::vector<VideoCodec>& recv_codecs() const { return recv_codecs_; }
const std::vector<VideoCodec>& send_codecs() const { return send_codecs_; }
const std::vector<VideoCodec>& codecs() const { return send_codecs(); }
bool rendering() const { return playout(); }
const VideoOptions& options() const { return options_; }
const std::map<uint32, VideoRenderer*>& renderers() const {
return renderers_;
}
int start_bps() const { return start_bps_; }
int max_bps() const { return max_bps_; }
bool GetSendStreamFormat(uint32 ssrc, VideoFormat* format) {
if (send_formats_.find(ssrc) == send_formats_.end()) {
return false;
}
*format = send_formats_[ssrc];
return true;
}
virtual bool SetSendStreamFormat(uint32 ssrc, const VideoFormat& format) {
if (send_formats_.find(ssrc) == send_formats_.end()) {
return false;
}
send_formats_[ssrc] = format;
return true;
}
virtual bool AddSendStream(const StreamParams& sp) {
if (!RtpHelper<VideoMediaChannel>::AddSendStream(sp)) {
return false;
}
SetSendStreamDefaultFormat(sp.first_ssrc());
return true;
}
virtual bool RemoveSendStream(uint32 ssrc) {
send_formats_.erase(ssrc);
return RtpHelper<VideoMediaChannel>::RemoveSendStream(ssrc);
}
virtual bool SetRecvCodecs(const std::vector<VideoCodec>& codecs) {
if (fail_set_recv_codecs()) {
// Fake the failure in SetRecvCodecs.
return false;
}
recv_codecs_ = codecs;
return true;
}
virtual bool SetSendCodecs(const std::vector<VideoCodec>& codecs) {
if (fail_set_send_codecs()) {
// Fake the failure in SetSendCodecs.
return false;
}
send_codecs_ = codecs;
for (std::vector<StreamParams>::const_iterator it = send_streams().begin();
it != send_streams().end(); ++it) {
SetSendStreamDefaultFormat(it->first_ssrc());
}
return true;
}
virtual bool GetSendCodec(VideoCodec* send_codec) {
if (send_codecs_.empty()) {
return false;
}
*send_codec = send_codecs_[0];
return true;
}
virtual bool SetRender(bool render) {
set_playout(render);
return true;
}
virtual bool SetRenderer(uint32 ssrc, VideoRenderer* r) {
if (ssrc != 0 && renderers_.find(ssrc) == renderers_.end()) {
return false;
}
if (ssrc != 0) {
renderers_[ssrc] = r;
}
return true;
}
virtual bool SetSend(bool send) { return set_sending(send); }
virtual bool SetCapturer(uint32 ssrc, VideoCapturer* capturer) {
capturers_[ssrc] = capturer;
return true;
}
bool HasCapturer(uint32 ssrc) const {
return capturers_.find(ssrc) != capturers_.end();
}
virtual bool SetStartSendBandwidth(int bps) {
start_bps_ = bps;
return true;
}
virtual bool SetMaxSendBandwidth(int bps) {
max_bps_ = bps;
return true;
}
virtual bool AddRecvStream(const StreamParams& sp) {
if (!RtpHelper<VideoMediaChannel>::AddRecvStream(sp))
return false;
renderers_[sp.first_ssrc()] = NULL;
return true;
}
virtual bool RemoveRecvStream(uint32 ssrc) {
if (!RtpHelper<VideoMediaChannel>::RemoveRecvStream(ssrc))
return false;
renderers_.erase(ssrc);
return true;
}
virtual bool GetStats(VideoMediaInfo* info) { return false; }
virtual bool SendIntraFrame() {
sent_intra_frame_ = true;
return true;
}
virtual bool RequestIntraFrame() {
requested_intra_frame_ = true;
return true;
}
virtual bool SetOptions(const VideoOptions& options) {
options_ = options;
return true;
}
virtual bool GetOptions(VideoOptions* options) const {
*options = options_;
return true;
}
virtual void UpdateAspectRatio(int ratio_w, int ratio_h) {}
void set_sent_intra_frame(bool v) { sent_intra_frame_ = v; }
bool sent_intra_frame() const { return sent_intra_frame_; }
void set_requested_intra_frame(bool v) { requested_intra_frame_ = v; }
bool requested_intra_frame() const { return requested_intra_frame_; }
private:
// Be default, each send stream uses the first send codec format.
void SetSendStreamDefaultFormat(uint32 ssrc) {
if (!send_codecs_.empty()) {
send_formats_[ssrc] = VideoFormat(
send_codecs_[0].width, send_codecs_[0].height,
cricket::VideoFormat::FpsToInterval(send_codecs_[0].framerate),
cricket::FOURCC_I420);
}
}
FakeVideoEngine* engine_;
std::vector<VideoCodec> recv_codecs_;
std::vector<VideoCodec> send_codecs_;
std::map<uint32, VideoRenderer*> renderers_;
std::map<uint32, VideoFormat> send_formats_;
std::map<uint32, VideoCapturer*> capturers_;
bool sent_intra_frame_;
bool requested_intra_frame_;
VideoOptions options_;
int start_bps_;
int max_bps_;
};
class FakeSoundclipMedia : public SoundclipMedia {
public:
virtual bool PlaySound(const char* buf, int len, int flags) { return true; }
};
class FakeDataMediaChannel : public RtpHelper<DataMediaChannel> {
public:
explicit FakeDataMediaChannel(void* unused)
: send_blocked_(false), max_bps_(-1) {}
~FakeDataMediaChannel() {}
const std::vector<DataCodec>& recv_codecs() const { return recv_codecs_; }
const std::vector<DataCodec>& send_codecs() const { return send_codecs_; }
const std::vector<DataCodec>& codecs() const { return send_codecs(); }
int max_bps() const { return max_bps_; }
virtual bool SetRecvCodecs(const std::vector<DataCodec>& codecs) {
if (fail_set_recv_codecs()) {
// Fake the failure in SetRecvCodecs.
return false;
}
recv_codecs_ = codecs;
return true;
}
virtual bool SetSendCodecs(const std::vector<DataCodec>& codecs) {
if (fail_set_send_codecs()) {
// Fake the failure in SetSendCodecs.
return false;
}
send_codecs_ = codecs;
return true;
}
virtual bool SetSend(bool send) { return set_sending(send); }
virtual bool SetReceive(bool receive) {
set_playout(receive);
return true;
}
virtual bool SetStartSendBandwidth(int bps) { return true; }
virtual bool SetMaxSendBandwidth(int bps) {
max_bps_ = bps;
return true;
}
virtual bool AddRecvStream(const StreamParams& sp) {
if (!RtpHelper<DataMediaChannel>::AddRecvStream(sp))
return false;
return true;
}
virtual bool RemoveRecvStream(uint32 ssrc) {
if (!RtpHelper<DataMediaChannel>::RemoveRecvStream(ssrc))
return false;
return true;
}
virtual bool SendData(const SendDataParams& params,
const talk_base::Buffer& payload,
SendDataResult* result) {
if (send_blocked_) {
*result = SDR_BLOCK;
return false;
} else {
last_sent_data_params_ = params;
last_sent_data_ = std::string(payload.data(), payload.length());
return true;
}
}
SendDataParams last_sent_data_params() { return last_sent_data_params_; }
std::string last_sent_data() { return last_sent_data_; }
bool is_send_blocked() { return send_blocked_; }
void set_send_blocked(bool blocked) { send_blocked_ = blocked; }
private:
std::vector<DataCodec> recv_codecs_;
std::vector<DataCodec> send_codecs_;
SendDataParams last_sent_data_params_;
std::string last_sent_data_;
bool send_blocked_;
int max_bps_;
};
// A base class for all of the shared parts between FakeVoiceEngine
// and FakeVideoEngine.
class FakeBaseEngine {
public:
FakeBaseEngine()
: loglevel_(-1),
options_changed_(false),
fail_create_channel_(false) {}
bool Init(talk_base::Thread* worker_thread) { return true; }
void Terminate() {}
void SetLogging(int level, const char* filter) {
loglevel_ = level;
logfilter_ = filter;
}
void set_fail_create_channel(bool fail) { fail_create_channel_ = fail; }
const std::vector<RtpHeaderExtension>& rtp_header_extensions() const {
return rtp_header_extensions_;
}
protected:
int loglevel_;
std::string logfilter_;
// Flag used by optionsmessagehandler_unittest for checking whether any
// relevant setting has been updated.
// TODO(thaloun): Replace with explicit checks of before & after values.
bool options_changed_;
bool fail_create_channel_;
std::vector<RtpHeaderExtension> rtp_header_extensions_;
};
class FakeVoiceEngine : public FakeBaseEngine {
public:
FakeVoiceEngine()
: output_volume_(-1),
delay_offset_(0),
rx_processor_(NULL),
tx_processor_(NULL) {
// Add a fake audio codec. Note that the name must not be "" as there are
// sanity checks against that.
codecs_.push_back(AudioCodec(101, "fake_audio_codec", 0, 0, 1, 0));
}
int GetCapabilities() { return AUDIO_SEND | AUDIO_RECV; }
AudioOptions GetAudioOptions() const {
return options_;
}
AudioOptions GetOptions() const {
return options_;
}
bool SetOptions(const AudioOptions& options) {
options_ = options;
options_changed_ = true;
return true;
}
VoiceMediaChannel* CreateChannel() {
if (fail_create_channel_) {
return NULL;
}
FakeVoiceMediaChannel* ch = new FakeVoiceMediaChannel(this);
channels_.push_back(ch);
return ch;
}
FakeVoiceMediaChannel* GetChannel(size_t index) {
return (channels_.size() > index) ? channels_[index] : NULL;
}
void UnregisterChannel(VoiceMediaChannel* channel) {
channels_.erase(std::find(channels_.begin(), channels_.end(), channel));
}
SoundclipMedia* CreateSoundclip() { return new FakeSoundclipMedia(); }
const std::vector<AudioCodec>& codecs() { return codecs_; }
void SetCodecs(const std::vector<AudioCodec> codecs) { codecs_ = codecs; }
bool SetDelayOffset(int offset) {
delay_offset_ = offset;
return true;
}
bool SetDevices(const Device* in_device, const Device* out_device) {
in_device_ = (in_device) ? in_device->name : "";
out_device_ = (out_device) ? out_device->name : "";
options_changed_ = true;
return true;
}
bool GetOutputVolume(int* level) {
*level = output_volume_;
return true;
}
bool SetOutputVolume(int level) {
output_volume_ = level;
options_changed_ = true;
return true;
}
int GetInputLevel() { return 0; }
bool SetLocalMonitor(bool enable) { return true; }
bool StartAecDump(talk_base::PlatformFile file) { return false; }
bool RegisterProcessor(uint32 ssrc, VoiceProcessor* voice_processor,
MediaProcessorDirection direction) {
if (direction == MPD_RX) {
rx_processor_ = voice_processor;
return true;
} else if (direction == MPD_TX) {
tx_processor_ = voice_processor;
return true;
}
return false;
}
bool UnregisterProcessor(uint32 ssrc, VoiceProcessor* voice_processor,
MediaProcessorDirection direction) {
bool unregistered = false;
if (direction & MPD_RX) {
rx_processor_ = NULL;
unregistered = true;
}
if (direction & MPD_TX) {
tx_processor_ = NULL;
unregistered = true;
}
return unregistered;
}
private:
std::vector<FakeVoiceMediaChannel*> channels_;
std::vector<AudioCodec> codecs_;
int output_volume_;
int delay_offset_;
std::string in_device_;
std::string out_device_;
VoiceProcessor* rx_processor_;
VoiceProcessor* tx_processor_;
AudioOptions options_;
friend class FakeMediaEngine;
};
class FakeVideoEngine : public FakeBaseEngine {
public:
FakeVideoEngine() : renderer_(NULL), capture_(false), processor_(NULL) {
// Add a fake video codec. Note that the name must not be "" as there are
// sanity checks against that.
codecs_.push_back(VideoCodec(0, "fake_video_codec", 0, 0, 0, 0));
}
bool GetOptions(VideoOptions* options) const {
*options = options_;
return true;
}
bool SetOptions(const VideoOptions& options) {
options_ = options;
options_changed_ = true;
return true;
}
int GetCapabilities() { return VIDEO_SEND | VIDEO_RECV; }
bool SetDefaultEncoderConfig(const VideoEncoderConfig& config) {
default_encoder_config_ = config;
return true;
}
VideoEncoderConfig GetDefaultEncoderConfig() const {
return default_encoder_config_;
}
const VideoEncoderConfig& default_encoder_config() const {
return default_encoder_config_;
}
VideoMediaChannel* CreateChannel(VoiceMediaChannel* channel) {
if (fail_create_channel_) {
return NULL;
}
FakeVideoMediaChannel* ch = new FakeVideoMediaChannel(this);
channels_.push_back(ch);
return ch;
}
FakeVideoMediaChannel* GetChannel(size_t index) {
return (channels_.size() > index) ? channels_[index] : NULL;
}
void UnregisterChannel(VideoMediaChannel* channel) {
channels_.erase(std::find(channels_.begin(), channels_.end(), channel));
}
const std::vector<VideoCodec>& codecs() const { return codecs_; }
bool FindCodec(const VideoCodec& in) {
for (size_t i = 0; i < codecs_.size(); ++i) {
if (codecs_[i].Matches(in)) {
return true;
}
}
return false;
}
void SetCodecs(const std::vector<VideoCodec> codecs) { codecs_ = codecs; }
bool SetCaptureDevice(const Device* device) {
in_device_ = (device) ? device->name : "";
options_changed_ = true;
return true;
}
bool SetLocalRenderer(VideoRenderer* r) {
renderer_ = r;
return true;
}
bool SetCapture(bool capture) {
capture_ = capture;
return true;
}
VideoFormat GetStartCaptureFormat() const {
return VideoFormat(640, 480, cricket::VideoFormat::FpsToInterval(30),
FOURCC_I420);
}
sigslot::repeater2<VideoCapturer*, CaptureState> SignalCaptureStateChange;
private:
std::vector<FakeVideoMediaChannel*> channels_;
std::vector<VideoCodec> codecs_;
VideoEncoderConfig default_encoder_config_;
std::string in_device_;
VideoRenderer* renderer_;
bool capture_;
VideoProcessor* processor_;
VideoOptions options_;
friend class FakeMediaEngine;
};
class FakeMediaEngine :
public CompositeMediaEngine<FakeVoiceEngine, FakeVideoEngine> {
public:
FakeMediaEngine() {
voice_ = FakeVoiceEngine();
video_ = FakeVideoEngine();
}
virtual ~FakeMediaEngine() {}
virtual void SetAudioCodecs(const std::vector<AudioCodec> codecs) {
voice_.SetCodecs(codecs);
}
virtual void SetVideoCodecs(const std::vector<VideoCodec> codecs) {
video_.SetCodecs(codecs);
}
FakeVoiceMediaChannel* GetVoiceChannel(size_t index) {
return voice_.GetChannel(index);
}
FakeVideoMediaChannel* GetVideoChannel(size_t index) {
return video_.GetChannel(index);
}
AudioOptions audio_options() const { return voice_.options_; }
int audio_delay_offset() const { return voice_.delay_offset_; }
int output_volume() const { return voice_.output_volume_; }
const VideoEncoderConfig& default_video_encoder_config() const {
return video_.default_encoder_config_;
}
const std::string& audio_in_device() const { return voice_.in_device_; }
const std::string& audio_out_device() const { return voice_.out_device_; }
VideoRenderer* local_renderer() { return video_.renderer_; }
int voice_loglevel() const { return voice_.loglevel_; }
const std::string& voice_logfilter() const { return voice_.logfilter_; }
int video_loglevel() const { return video_.loglevel_; }
const std::string& video_logfilter() const { return video_.logfilter_; }
bool capture() const { return video_.capture_; }
bool options_changed() const {
return voice_.options_changed_ || video_.options_changed_;
}
void clear_options_changed() {
video_.options_changed_ = false;
voice_.options_changed_ = false;
}
void set_fail_create_channel(bool fail) {
voice_.set_fail_create_channel(fail);
video_.set_fail_create_channel(fail);
}
bool voice_processor_registered(MediaProcessorDirection direction) const {
if (direction == MPD_RX) {
return voice_.rx_processor_ != NULL;
} else if (direction == MPD_TX) {
return voice_.tx_processor_ != NULL;
}
return false;
}
};
// CompositeMediaEngine with FakeVoiceEngine to expose SetAudioCodecs to
// establish a media connectionwith minimum set of audio codes required
template <class VIDEO>
class CompositeMediaEngineWithFakeVoiceEngine :
public CompositeMediaEngine<FakeVoiceEngine, VIDEO> {
public:
CompositeMediaEngineWithFakeVoiceEngine() {}
virtual ~CompositeMediaEngineWithFakeVoiceEngine() {}
virtual void SetAudioCodecs(const std::vector<AudioCodec>& codecs) {
CompositeMediaEngine<FakeVoiceEngine, VIDEO>::voice_.SetCodecs(codecs);
}
};
// Have to come afterwards due to declaration order
inline FakeVoiceMediaChannel::~FakeVoiceMediaChannel() {
if (engine_) {
engine_->UnregisterChannel(this);
}
}
inline FakeVideoMediaChannel::~FakeVideoMediaChannel() {
if (engine_) {
engine_->UnregisterChannel(this);
}
}
class FakeDataEngine : public DataEngineInterface {
public:
FakeDataEngine() : last_channel_type_(DCT_NONE) {}
virtual DataMediaChannel* CreateChannel(DataChannelType data_channel_type) {
last_channel_type_ = data_channel_type;
FakeDataMediaChannel* ch = new FakeDataMediaChannel(this);
channels_.push_back(ch);
return ch;
}
FakeDataMediaChannel* GetChannel(size_t index) {
return (channels_.size() > index) ? channels_[index] : NULL;
}
void UnregisterChannel(DataMediaChannel* channel) {
channels_.erase(std::find(channels_.begin(), channels_.end(), channel));
}
virtual void SetDataCodecs(const std::vector<DataCodec>& data_codecs) {
data_codecs_ = data_codecs;
}
virtual const std::vector<DataCodec>& data_codecs() { return data_codecs_; }
DataChannelType last_channel_type() const { return last_channel_type_; }
private:
std::vector<FakeDataMediaChannel*> channels_;
std::vector<DataCodec> data_codecs_;
DataChannelType last_channel_type_;
};
} // namespace cricket
#endif // TALK_MEDIA_BASE_FAKEMEDIAENGINE_H_