Re: HTTP/2.0 Magic

On Jan 31, 2013, at 11:46 PM, David Morris wrote:
> It seems to me that including CRLF is more likely to result in the string
> being ignored then if it is an unterminated sequence which would get
> catenated with the payload by a parser which didn't know about it.

The goal is to have a fast response, and most decently written
text-based protocol servers won't process the input (and see
that it is invalid) until a line is received, a timeout occurs,
or the receive buffer is full.  That is doubly so when the kernel
is asked to do line buffering before accept.

I think it is fair to say we could continue bike-shedding this
for a long time and end up with a fair number of polarized yet
irrelevant designs.  I suggest folks just code it up and see
which ideas break the best on the most networks.

....Roy

Received on Friday, 1 February 2013 09:30:13 UTC