Re: Data API Proposal aligned to WebSocket

On Feb 23, 2012, at 1:39 PM, Adam Bergkvist wrote:

> Hi Michael
> 
> Find my replies inline.
> 
> On 02/23/2012 12:41 PM, Michael Tuexen wrote:
>>> var chan = peerConn.createDataChannel("mylabel");
>> What is the usage of 'mylabel'. It doesn't seem to be used locally or remotely.
>> I assume that you want to map channels to bi-directional streams, is this right?
> 
> You're right that it's not used in either of the examples given here. It's a pure API-related property to help the developer identify a particular DataChannel object if several objects (with the same transport porperties) are used. E.g. you might have a DataChannel object with label "chat" and another with the label "game-data" for different purposes.
OK. So if I understand this correctly, it is just a local thing and it is not expected to be on the wire, right?
> 
> I'm not even sure if it's really needed since I think we could specify that two consecutive calls to createDataChannel(), which creates two DataChannel objects A-side, will generate two "datachannel" events on the B-side in the same order. If the order of the events on the B-side would be undefined, we would need the label property for the web app to know which DataChannel on the B-side that correspons to a particular object on the A-side. Similar to the label property on MediaStream objects.
> 
>>> *** DataChannel ***
>>> 
>>> interface DataChannel {
>>>    readonly attribute DOMString label;
>> What is the semantic of this?
> 
> See above.
> 
>>> 
>>>    readonly attribute boolean reliable;
>> This is a property of a user message, not of a stream.
> 
> That might be the case for SCTP, but it's perhaps not the best model for a Web API. I would really prefer to differentiate reliable and unreliable on DataChannel level in the API. It would make the API simpler to use and more aligned with existing Web APIs IMO.
That's fine. I just wanted to make sure that it is done intentionally.
> 
>>>    readonly attribute long priority;
>> This is a property of a stream.
> 
> Yes, the intention is to let a DataChannel model one stream in each direction.
> 
>>>    // ... extend here ...
>>> };
>> What about unordered? When do you expect a message to be abondoned when treated unreliable?
> 
> I don't really have an opinion on when to abandon messages in unreliable mode.
OK. But it needs to be specified.

Best regards
Michael
> 
> /Adam
> 
> 
> 
>

Received on Thursday, 23 February 2012 13:09:18 UTC