Early initialize recording on the ADM from WebRtcVoiceMediaChannel.

webrtc/media/engine/webrtcvoiceengine.cc

Index: webrtc/media/engine/webrtcvoiceengine.cc
diff --git a/webrtc/media/engine/webrtcvoiceengine.cc b/webrtc/media/engine/webrtcvoiceengine.cc
index 21094fde4d45af8baed7c5228237a5caba363e04..949838062829d7275215c3d5ae7f3fbe8a99a646 100644
--- a/webrtc/media/engine/webrtcvoiceengine.cc
+++ b/webrtc/media/engine/webrtcvoiceengine.cc
@@ -1078,6 +1078,12 @@ int WebRtcVoiceEngine::CreateVoEChannel() {
   return voe_wrapper_->base()->CreateChannel(voe_config_);
 }
 
+webrtc::AudioDeviceModule* WebRtcVoiceEngine::adm() {
+  RTC_DCHECK(worker_thread_checker_.CalledOnValidThread());
+  RTC_DCHECK(adm_);
+  return adm_;
+}
+
 class WebRtcVoiceMediaChannel::WebRtcAudioSendStream
     : public AudioSource::Sink {
  public:
@@ -1799,18 +1805,19 @@ void WebRtcVoiceMediaChannel::SetSend(bool send) {
   if (send_ == send) {
     return;
   }
+  send_ = send;
 
-  // Apply channel specific options when channel is enabled for sending.
-  if (send) {
+  // Apply channel specific options.
+  if (send_) {
     engine()->ApplyOptions(options_);
   }
 
+  SetupRecording();

[juberti2, 2016/03/31 18:57:38]

Won't this happen anyway as a result of SetSend?

[the sun, 2016/04/01 09:07:46]

Well, it used to, but in the patch set you comment on here, AudioSendStream had
been changed to call directly onto voe::Channel (instead of going through
VoEBase), meaning only the channel state  would be altered and the ADM
untouched, hence the explicit call to SetupRecording() here.

+
   // Change the settings on each send channel.
   for (auto& kv : send_streams_) {
     kv.second->SetSend(send);
   }
-
-  send_ = send;
 }
 
 bool WebRtcVoiceMediaChannel::SetAudioSend(uint32_t ssrc,
@@ -1908,6 +1915,8 @@ bool WebRtcVoiceMediaChannel::AddSendStream(const StreamParams& sp) {
     }
   }
 
+  SetupRecording();
+
   send_streams_[ssrc]->SetSend(send_);
   return true;
 }
@@ -2496,6 +2505,22 @@ bool WebRtcVoiceMediaChannel::SetPlayout(int channel, bool playout) {
   }
   return true;
 }
+
+void WebRtcVoiceMediaChannel::SetupRecording() {
+  if (send_ && !engine()->adm()->Recording()) {
+    // InitRecording() is called as soon as we know we will be sending, as it
+    // may take time on some platforms (e.g. Android).
+    if (engine()->adm()->InitRecording() != 0) {
+      LOG(LS_WARNING) << "Failed to initialize recording";

[juberti2, 2016/03/31 18:57:38]

Shouldn't we propagate this error upwards instead of silently failing?

[the sun, 2016/04/01 09:07:46]

We used to propagate it up, but eventually, in pc/channel.cc, the return value
would be ignored. See:
https://codereview.webrtc.org/1741933002/diff/280001/webrtc/pc/channel.cc
(line 1457)
The decision was made to simplify the logic and maintain current semantics.

+      return;
+    }
+    // StartRecording() is only called once we actually have a stream to send.
+    if (!send_streams_.empty() && engine()->adm()->StartRecording() != 0) {
+      LOG(LS_WARNING) << "Failed to start recording";
+      return;
+    }
+  }
+}
 }  // namespace cricket
 
 #endif  // HAVE_WEBRTC_VOICE