Re: What would you like to see in WebRTC next? A low-level API?

On 24/01/2018 10:59, Philipp Hancke wrote:
> Am 24.01.2018 um 00:37 schrieb Bernard Aboba:
>> I'll provide some more detailed comments later, but would like to 
>> provide a few high-level thoughts (with my Chair hat off).
>>
>> Overall, my experience with developers is that they care most about 
>> stability and functionality.
>>
>> If there is a way to make something work, and if it is stable enough 
>> to deploy in production, they will incorporate it into their 
>> applications, even if many of would consider it a "hideous hack".
>>
>> So enabling something new, useful and solid is a good way to gain 
>> developer mindset.
>>
>> Doing the same thing in a more elegant way can be intellectually 
>> satisfying, but can be hard to convince developers to utilize if 
>> their existing code can do the same thing, albeit somewhat more 
>> clumsily.
>>
>> All this to say that if the goal is to create things that developers 
>> will use, it is often best to start from problems: things developers 
>> want to do, but have not been able to do so far.
>
> Having ported a rather complex app to use addTrack&friends instead of 
> the "legacy" addStream one I can say that:
> 1) I haven't found much that I could not do with addStream
> 2) it takes a lot of time and has close to zero business value
> 3) you pay an extra price for using the latest and greatest. When your 
> CI dashboard goes red because you happen to be using Chrome's native 
> addTrack because its available and not quite ready yet...
> 4) I mainly did this to ensure the WG doesn't specify things that will 
> make my life harder in the distant future
> 5) if I had to start from scratch I would use the "legacy" APIs.
>
> There is also a great disconnect between what the WG is doing and what 
> (web) application developers need, judging by the (lack of) 
> involvement of that group.
>
> I am still happy with Edge's take on ORTC even though given the lack 
> of support in other browsers means that doing something as crazy as 
> implementing RTCPeerConnection ontop of it is viable.
>
+1 I think we have been focusing so much on interoperability with 
inexistent endpoints (ice and dtls but not bundle, for example), 
covering edge cases and trying to map legacy technology to webrtc that 
we have make the simple cases extremely difficult to understand and use 
for the normal use cases.

Whatever we decide to go for next chapter (I have some proposals I will 
send later), IMHO we should change our mindset to "API first".

Best regards

Sergio

Received on Wednesday, 24 January 2018 10:50:37 UTC