On Tue, Jan 22, 2013 at 4:11 AM, William Jordan <wjordan129@xxxxxxxxx> wrote: > Continuing my discussion about how badly SIP is designed, I'm gonna talk > about the via line. First of all each via line can be expressed as via: OR > v: OR you can have multiple via entries on the same line separated by a > comma where random whitespaces are allowed. Hey, we stole that stuff from HTTP. We're not smart enough to invent something that broken. I think HTTP stole it from RFC 822. > Additionally, I can't > understand why each line is terminated with <CR><LF>, why use two characters > when one will do. Microsoft-OS text editors. Seriously. People wanted to be able to write correct SIP messages using text editors, and there were more Microsoft users than Linux users on the list. And we agreed because that's the way it works in HTTP, which got it from RFC 822 (there's a reference in the spec, so we KNOW where that came from!) > I could also go into a design argument about how the same > specification is used for udp as for tcp and why the idea for a stateless > SIP proxy wasn't thrown out completely when writing the draft, but I'll wait > on those. Yeah, keeping both UDP and TCP was stupid. We should've just specified TLS, but there were all these 8-bit processors out there ... and countries where TLS was illegal. In fact, we couldn't export a TLS stack from the US at the time without special paperwork. So, allowing for both protocols got us "exportable" product, and "passing the IESG security police" at the same time. Stateless proxies were seen as key performance enablers. Our high-end super-performant proxy running on a $40,000 quad-core Sun box could only get like 8 transactions a second, so every CPU optimization was sacred. But hey, live and learn ...