Decouple SDP parser from SIP parser

[negbie]

I think it would make sense to decouple the SDP parser from the SIP parser. @moggle-mog already saw it here
#9 (comment)

@jart , @moggle-mog what do you think?

[nextpkg]

Just Do IT.

[jart]

Reviewing the most recent code, the SIP->SDP relationship looks clean. We can absolutely improve it though. One possible idea is using the Registration Pattern. The SIP package could have a list of payload parsers. We'd need to define a new API for that. Then we could have the SDP package call sip.RegisterPayloadParser() in its init() function. I like having registration be a side-effect of the linkage choices each user makes. I also like the idea of enabling third party projects to register custom parsers.

Apropos #9 please understand that IETF documents are really well-written recommendations. Open source projects are usually rewarded for following them. However there's no punishment for not following them. I like to think of it as a challenge, to figure out a way to do what I want to do, within the arbitrary constraints defined by the Internet Engineering Task Force. If you look at it the way I do, then you'll find that implementing protocols becomes a whole lot more fun!

[negbie]

I like to think of it as a challenge, to figure out a way to do what I want to do, within the arbitrary constraints defined by the Internet Engineering Task Force. If you look at it the way I do, then you'll find that implementing protocols becomes a whole lot more fun!

That's a great mindset!

Reviewing the most recent code, the SIP->SDP relationship looks clean. We can absolutely improve it though. One possible idea is using the Registration Pattern. The SIP package could have a list of payload parsers.

That sound's good and Payload is already an Interface!

[nextpkg]

Reviewing the most recent code, the SIP->SDP relationship looks clean. We can absolutely improve it though. One possible idea is using the Registration Pattern. The SIP package could have a list of payload parsers. We'd need to define a new API for that. Then we could have the SDP package call sip.RegisterPayloadParser() in its init() function. I like having registration be a side-effect of the linkage choices each user makes. I also like the idea of enabling third party projects to register custom parsers.

Apropos #9 please understand that IETF documents are really well-written recommendations. Open source projects are usually rewarded for following them. However there's no punishment for not following them. I like to think of it as a challenge, to figure out a way to do what I want to do, within the arbitrary constraints defined by the Internet Engineering Task Force. If you look at it the way I do, then you'll find that implementing protocols becomes a whole lot more fun!

It sound's good, when we init SIP, we can also init the payload parser.However, can we put the payload into generic struct?So user can decide how to use it.
For example in msg_parser.rl,

if ctype == sdp.ContentType {
	ms, err := sdp.Parse(string(data[p:len(data)]))
	if err != nil {
		return nil, err
	}
	msg.Payload = ms
} else {
	msg.Payload = &MiscPayload{T: ctype, D: data[p:len(data)]}
}

Another way is,

msg.Payload = &MiscPayload{T: ctype, D: data[p:len(data)]}

User receive this message and use Content-Type to choose the better parser like this,

tp, err := sip.NewTransport(....)
if err != nil {
	panic(err)
}

....

msg := <-tp.C
switch  msg.Payload.ContentType() {
case sdp.ContentType:
        // It is a good idea?
	sp := sdp.Parse(string(msg.Payload.Data()))
	sp...
	....
}
....

Of course, this will increase the complexity of the user code, I don't know whether it will make us crazy or not.

[jart]

One of the things that makes gosip appealing is it empowers the user to implement a command line telephone in a few hundred lines with complete control over the main event loop. We can introduce complexity. Just want to understand benefits.

SIP+SDP are independent designs in the sense that HTTP+HTML are independent. I've never seen a SIP message carry a payload type that isn't SDP. That's why the assumption is baked into the design. Could you cite any specific publicly documented examples of existence of non SDP payloads?

[jart]

By the way! I just realized that the G.729 patents finally expired a couple years ago! It'd be great if we could implement that in the fastest way possible. It's an alternative RTP payload that reduces the bandwidth consumption of each telephone call from 64kb to 8kb, with quality that's somehow better than than the GSM audio codec that needs 32kb! Now that G.729 no longer costs $10 per channel, maybe that's a good area of focus for offering alternative choices.

[negbie]

Mby we could use this https://github.com/BelledonneCommunications/bcg729 but it wouldn't be the fastest way possible.

[negbie]

If you want reduced bandwidth usage and good audio quality opus is quite nice! https://github.com/hraban/opus

[negbie]

Downside of opus is that not many carrier support it.

[jart]

It's unfortunate but all the commercial systems I've seen, tend to usually support a choice of either G.711 (which has been standard since T1 in 1950's) and G.729. The BCG code looks pretty clean. It looks like the kind of thing that would be fun to profile and make go even faster. For example, I'm noticing it's using Q15 arithmetic. Little known optimization secret: the PMULHRSW instruction can make that go wicked fast, on most CPUs.

[negbie]

the PMULHRSW instruction can make that go wicked fast, on most CPUs.

Yip, part of SSSE3 and could be even used by Atom's.

[nextpkg]

One of the things that makes gosip appealing is it empowers the user to implement a command line telephone in a few hundred lines with complete control over the main event loop. We can introduce complexity. Just want to understand benefits.

SIP+SDP are independent designs in the sense that HTTP+HTML are independent. I've never seen a SIP message carry a payload type that isn't SDP. That's why the assumption is baked into the design. Could you cite any specific publicly documented examples of existence of non SDP payloads?

Such as Content-Type: multipart/mixed;boundary="boundary1" from https://tools.ietf.org/html/rfc5621 page 4 and
message/cpim of IM.

[razor-1]

SIPREC is a surprisingly common use case for a different content-type as well. RFC7866 is the spec and RFC8068 has a number of sample call flows.