SocketConnection pattern guidance

[mgravell]

SocketConnection pattern guidance

[mgravell]

Looks broadly fine. I'd probably try to avoid dealing in lengths all the time, because in some edge cases this can cause catastrophic performance (usually when dealing with massively recursive models) - if you can deal in positions rather than lengths it may work better, but: in most cases (the ones not I volving massively recursive models) it won't matter. I can't see the parser code - note that for optimal performance you want to use a ref-counting "buffer reader" that tracks a span, rather than constantly accessing spans from the buffer. SE.Redis and protobuf-net both have examples of this that you may find useful.

[Wraith2]

Ok, thanks, that's good. I was worried about the looping with tasks. I'd normally approach this with a dedicated thread and events but that doesn't scale.

The parser implementation contains protocol details which are proprietary so I can't post that unfortunately. The protocol is pretty simple flat utf8 encoded with no nesting. The parser takes the sequence and attempts to pull apart each slice, I can show that part, I grab each segment span and attempt to consume it as far as possible skipping any known prefix so I avoid re-running a complex parse if possible.

I'm not sure what you mean with the lengths. The Message object returned by the parser will always start at offset 0 so length is both a length and a relative position, Which part would you change and what approach would be better?

I'll find and take a look at the SE.Redis code, that's probably closest to what I'm doing.