2015-10-26 20:50 GMT+01:00 Mark Thomas <ma...@apache.org>: > On 26/10/2015 08:10, r...@apache.org wrote: > > Author: remm > > Date: Mon Oct 26 15:10:53 2015 > > New Revision: 1710618 > > > > URL: http://svn.apache.org/viewvc?rev=1710618&view=rev > > Log: > > - Fix capacity check algorithm (overflow isn't the right result in that > case). > > - Unwrap first in NIO2 and wait for an explicit underflow to read on the > socket (I'll test adding a flag to optimize this since this is likely a bit > expensive, but commit it for now since everything's now working). > > Just to clarify, do you mean that all the various issues you were > observing with NIO and NIO2 with the OpenSSL engine with and without h2 > are now resolved? If so, WOOT! > > It looks good for me with Chrome and HTTP/2. I can't say if everything is fine yet, but it's likely. The behavior you found is listed as a known bug in OpenSSL's API, unfortunately the native code didn't work around it. So at least the engine has to do it since the behavior is obviously not consistent with JSSE's engine.
NIO and its non blocking reads behaved better in most cases, it would read 0 and proceed with unwrapping (again), while NIO2 would not unwrap again until it got more data [which isn't bad by itself, but this doesn't work due to the OpenSSL's bug]. Rémy
