2016-10-27 0:44 GMT+03:00 Rémy Maucherat <r...@apache.org>: > > 2016-10-26 20:45 GMT+02:00 Violeta Georgieva <miles...@gmail.com>: > > > Hi Remy, > > > > 2016-10-25 1:49 GMT+03:00 Rémy Maucherat <r...@apache.org>: > > > > > > 2016-10-24 20:18 GMT+02:00 Violeta Georgieva <miles...@gmail.com>: > > > > > > > There are no issues reported for the refactored functionality available > > in > > > > 9.0.0.M11. > > > > I think it is safe to back port the changes for the next 8.5 release. > > > > Do you see any issues? > > > > > > > > Ok, so it probably works. I suppose it's up to me, at some point, to > > try > > > to improve whatever I would like to improve (for example, it lost API > > > symmetry - normally, read/write APIs are the same, not the case here -, > > and > > > it's not very GC friendly). > > > > Regarding API symmetry: for the read API I cannot use ByteBuffer directly > > as parameter so the option is to change the write API to use > > ApplicationBufferHandler. > > This change can be done easily but the drawback is that one additional > > wrapper object will be created for each Response object. > > If you are OK with this I can make it. > > > > No. Since my own async scatter/gather didn't provide any benefits, unless > something unexpected occurs I'll give up on that and could try to rework > this at some point.
So are you OK to have in Tomcat 8.5 the read/write with ByteBuffer in CoyoteInputStream/CoyoteOutputStream? The main purpose for this refactoring was to introduce them without regression in the performance. I would like to have these APIs in Tomcat 8.5. Thanks, Violeta > > > > Let me know if you have other requirements for this change so that I can > > implement them and after that backport the change in Tomcat 8.5. > > > > I will double check my tests again for GC issues. Do you have some > > observations or concrete concerns so I can focus on them? > > > > It does duplicate ByteBuffers, that creates some objects. it doesn't have > any measurable impact on performance however since I didn't notice anything. > > Rémy
