WIP: Move the creation of AudioCodecFactory into PeerConnectionFactory.

webrtc/modules/audio_coding/neteq/neteq.cc

Index: webrtc/modules/audio_coding/neteq/neteq.cc
diff --git a/webrtc/modules/audio_coding/neteq/neteq.cc b/webrtc/modules/audio_coding/neteq/neteq.cc
index edc29da4357ab0546275ebb249293b90096798b9..02b674fbb80629e07289752e8acb3fabd4e03061 100644
--- a/webrtc/modules/audio_coding/neteq/neteq.cc
+++ b/webrtc/modules/audio_coding/neteq/neteq.cc
@@ -32,8 +32,12 @@ std::string NetEq::Config::ToString() const {
 
 // Creates all classes needed and inject them into a new NetEqImpl object.
 // Return the new object.
-NetEq* NetEq::Create(const NetEq::Config& config) {
-  return new NetEqImpl(config, NetEqImpl::Dependencies(config));
+NetEq* NetEq::Create(
+    const NetEq::Config& config,
+    std::shared_ptr<AudioDecoderFactory> decoder_factory) {

[kwiberg-webrtc, 2016/05/17 13:33:48]

Why not put the decoder factory in the config?

[ossu, 2016/05/17 14:25:35]

My local git log says this:
    Moved injection of AudioDecoderFactory from NetEq::Config to a separate
    argument in NetEq::Create. This makes the direct usage of NetEqImpl in
    tests unambiguous, since it takes a Config and a DecoderDatabase
    directly (among other things). It also makes sure an
    AudioDecoderFactory is always included - since the signature has
    changed, the compiler will scream at you, rather than just silently
    accepting a nullptr in the Config.

This was before the change to using a Dependencies object, but I believe the
logic goes unchanged: Dependencies contains a DecoderDatabase and is sent in
alongside the Config, so if the factory was in the config, it would be in two
places, or worse, there would be two different factories in different places.

+  return new NetEqImpl(
+      config,
+      NetEqImpl::Dependencies(config, std::move(decoder_factory)));
 }
 
 }  // namespace webrtc