invalid ice state transition from checking to disconnected?

[fippo]

the (non-normative) diagram for ice state transitions shows a transition from 'checking' to 'disconnected' (also in the opposite direction) which is not shown in in the equivalent peerconnection diagram.

That looks like a bug, I can't think of a way how the checking->disconnected transition would happen. And disconnected to checking in the case of an ice-restart would happen via new.

(note: even though both diagrams look quite different otherwise that seems ok since in the ORTC version there is an annotation that 'new' is reachable from most other states, reducing the visual clutter)

[aboba]

The transition from 'disconnected' to 'checking' can occur if a new interface comes up, thereby providing new candidates (and pairs) to check.

As for the transition from 'checking' to 'disconnected', in WebRTC 1.0, it can only occur via the 'connected' or 'failed' states, not directly. Thinking....

[pthatcherg]

If it can go from "disconnected" to "checking" it can go back to
"disconnected" from "checking", can't it? For example:

• Wifi goes away => disconnected
• 3G candidates gathered => checking
• 3G candidates gathered so far all fail => disconnected (still gathering,
  so not failed)

On Thu, Jan 7, 2016 at 3:52 PM, aboba notifications@github.com wrote:

The transition from 'disconnected' to 'checking' can occur if a new
interface comes up, thereby providing new candidates (and pairs) to check.

As for the transition from 'checking' to 'disconnected', in WebRTC 1.0, it
can only occur via the 'connected' or 'failed' states, not directly.
Thinking....

—
Reply to this email directly or view it on GitHub
#332 (comment).

[aboba]

Yes. So shouldn't the WebRTC 1.0 and ORTC ICE state diagrams be the same?

[aboba]

Submitted a PR to provide examples of RTCIceTransportState transitions.

[aboba]

Discussion on the same issue in WEBRTC WG: w3c/webrtc-pc#457

[aboba]

Slides prepared for the January 14, 2016 virtual interim of the WEBRTC WG on this topic:
https://www.w3.org/2011/04/webrtc/wiki/images/f/f4/WebRTCWG-2016-01-14.pdf

Unfortunately we ran out of time before we got to those slides :(

[aboba]

Some progress made at the February 25 WEBRTC WG interim.
Slides:
https://www.w3.org/2011/04/webrtc/wiki/images/5/50/WebRTCWG-2016-02-25.pdf
Minutes:
https://www.w3.org/2016/02/25-webrtc-minutes

[ibc]

PSD page 21:

• checking: ICE agent is checking candidate pairs, and has never been connected.
• connected: ICE agent is connected (may or may not be checking other pairs).
• disconnected: ICE agent is not currently connected, and was either previously connected or is not
  checking.

+1000 for this

[robin-raymond]

This is my state table I am using:
ice transport state table

[aboba]

Now that RTCIceTransportState also exists in WebRTC 1.0, and RTCIceConnectionState has been redefined based on that, the ICE transport state definitions (and transitions) in WebRTC 1.0 and ORTC are in the process of converging. The major outstanding issue appears to be the effect of ICE consent check failures on the states and transitions.

WebRTC 1.0 Issue 517 relates to state transitions due to ICE consent check failures:
w3c/webrtc-pc#517
The following PR has been submitted to address it:
w3c/webrtc-pc#611

Based on the resolution proposed in that PR, we are making progress on a proposed WebRTC 1.0 ICE transport state transition diagram:
w3c/webrtc-pc#457

Diagram:
https://cloud.githubusercontent.com/assets/808811/15027597/786a296a-11f9-11e6-8094-39ffb7cf49ed.png