Yet another answer to the IRC meeting It was me that proposed read I/O buffers is created by the I/O layer who feeds it to the upper layers. There is several reasons to this
- Amount needed not known until available, other than an upper bound. - It's uncertain when the data will be available, and it's undesireable to hold on to buffers until needed. - Makes it easier to layer things needing transforms. Consider for example SSL or chunked. Regarding SSL & chunked, my long term goal is a design where these are just layers in a chain using the same primary interface for data access. There really shouldn't be any difference reading a HTTP body, disk file, raw network socket or whatever stream of data. Or writing for that matter. Regards Henrik
