[rtcweb] Another reason not to use SDP (was: Draft agenda for IETF 87)

[Martin Thomson <[e-mail]>]

On 12 July 2013 10:19, Peter Thatcher <[e-mail]> wrote:
> I think advanced JS apps are going to use every control knob they can get
> to, whether it's anticipated and well-supported or not.    I think there's a
> good chance that a a popular WebRTC web app will use some SDP mangling that
> wasn't anticipated, but happened to work, but then the browser can't remove
> it because it will break certain websites.  It could get even worse if
> someone writes a popular WebRTC wrapper library that uses tricky SDP
> mangling that is then used by lots of websites.  Certain SDP mangling
> techniques might end up becoming a defacto standard API that can't be
> removed, even if it was originally a bug.  Or worse, one browser will have
> to implement the SDP mangling or even the bugs of another, because WebRTC
> apps have come to rely on them.
>
> In fact, it's already the case that Chrome and Firefox support far different
> SDP manglings.  I don't think any web apps rely on that yet, but it's only a
> matter of time before someone figures out "hey, if I mangle the SDP on this
> browser this way and on that browser that way, I can do things I couldn't do
> otherwise".  Or worse, an advanced web app developer says "hey, I can make
> this work well on browser X via SDP mangling, but not on browser Y, so I'll
> put a 'best used with Browser X' icon on my website". Then that someone
> writes an abstraction on top of that, and then maybe shares that with
> others, and it goes from there.
>
> I think we're in a race with web developers to see if they'll figure out SDP
> mangling before we provide a way to avoid SDP mangling.   Who do you think
> moves faster?

This is one of the arguments we've made against the use of SDP.  There
is a remedy for this (aside from comment 22 ;), but it's a fair amount
of work:

 - provide very explicit and detailed instructions on what SDP to
produce under all starting conditions, and

 - provide very explicit and detailed instructions on what to do with
every single bit of SDP that is provided.

That is what I believe to be necessary if this API is to have any hope
of real success.

Unfortunately, when I sat down to do this, I barely made it to t=
sections.  I simply cannot justify spending the time required for that
sort of chore.