> JEP-0124 requires a connection manager to handle multiple simultaneous > HTTP connections from the same client. I can't recall the rationale for > this design decision, but I've received feedback off-list that it makes > the connection manager fairly complex to code (and more complex than it > needs to be unless there is some compelling reason to support multiple > connections). Have any implementors run into issues with this?
Didn't seem hard to me when I implemented it. It's there for the simulated pipe stuff, specifically. If hold=1, the connection manager is supporting two requests at at time. Whenever a new one comes in, the old one gets flushed. Whenever a connection gets flushed early, the client should poll and start a new one. I'm not sure there's much pratical use for more than 2, since most browsers seems to be limited to this number anyway. But for instance, if there were hold=4, then you can just do the same thing in a FILO order. When one comes in, flush the oldest request. jack.
