Let's solve the recent getStats() instability

[feross]

I've been playing with the code that decides which version of RTCPeerConnection.prototype.getStats gets called to add support for , and it's unfortunately pretty finicky.

There are at least 3 implementations that I know of:

• Single-parameter callback-based getStats() (non-standard)
  • Used by Chrome, electron-webrtc, node-wrtc
• Promise-based getStats() (standard)
  • Used by Firefox, Chrome (if "Experimental Web Platform Features" is enabled)
• Two-parameter callback-based getStats() (deprecated, former standard)
  • Used by react-native-webrtc

It's not easy to feature detect which version is available, so we have to resort to user agent sniffing.

I propose that we:

• [x] Add an explicit peer._browserDetect() function so all the browser sniffing nastiness is isolated to a single function.
• [x] Re-enable the node.js tests that use node-wrtc, to catch regressions. (Hopefully it's easier to build these days).
• [x] Add node.js tests for electron-webrtc.
• [ ] Consider adding a try-catch around the getStats calls in case detection fails, or a user is using a very old or new version of a browser that has switched implementations (or a Chrome user who turned on "Experimental Web Platform Features")
• [ ] Encourage package maintainers to switch to the standard promise-based APIs by opening issues. Callback methods are non-standard/deprecated.

[feross]

First PR, review please: https://github.com/feross/simple-peer/pull/123

[feross]

Okay, this is all fixed now! Lots of good stuff is in here now, including support for the new spec-compliant getStats() report format.

This new format is available in Chrome today if you enable "Experimental Web Platform Features". It will ship in Chrome 58 and simple-peer will be ready for it. I'm not sure when Firefox is shipping a spec-compliant format, but this should work for that too.

[feross]

6.2.0