It might work fine, but we already have a MODE_EATCRLF call which ought to do the job (down to the core_input_filter anyway) as long as we remember what it returned. Are you suggesting that we replace that? Would that solve some data-stashed-in-connection-filter cases?
Yes as I don't think EATCRLF is the right thing to do here. I'd suggest replacing the EATCRLF with the non-blocking speculative read and seeing how that works. The side-effect of eating CRLFs is unnecessary. It should do the same thing and would be a more general solution as well as it can be called several times and should return the same thing each time. -- justin
