2015-05-27 8:44 GMT+02:00 Mark Thomas <ma...@apache.org>: > On 26/05/2015 19:43, Rémy Maucherat wrote: > > 2015-05-26 20:37 GMT+02:00 Mark Thomas <ma...@apache.org>: > > > >> On 26/05/2015 19:33, Rémy Maucherat wrote: > >>> 2015-05-26 20:17 GMT+02:00 Mark Thomas <ma...@apache.org>: > >>> > >>>> It was a left over from the attempt to make BIO support more > connections > >>>> than it had threads by not doing an immediate read on a keep-alive > >>>> connection. It was no longer required. There is another "return false" > >>>> in the correct place in the next if block. > >>>> > >>>> The testsuite still fails with NIO2, so something has to be wrong :) > >> > >> It hasn't had a chance to run with my fix yet. > >> > > That is correct, I thought the previous one had it already. > > Now I do see some test failures with my fix in place. I'll take a look > at these today. > > You're running into my little tricks which aimed at avoiding that a processor with a pending read would get disassociated from its socket, and thus it wasn't a good idea to try a non blocking read with NIO2 since it causes side effects. Since you moved the IO code into the socket, this is no longer a problem in theory.
Rémy