On 13/09/2016 19:07, Violeta Georgieva wrote: > 2016-08-30 15:35 GMT+03:00 Violeta Georgieva <miles...@gmail.com>:
<snip/> >> I introduced CoyoteOutputStream.write(ByteBuffer) it uses new methods >> with ByteBuffer instead of ByteChunk. >> Next step is to replace ByteChunk/CharChunk usage in CoyoteOutputStream >> with ByteBuffer/CharBuffer thus I will switch to the new methods and all >> CoyoteOutputStream.write method will use them. >> > > I would like to back port these changes to Tomcat 8.5 if you do not have > any concerns. +0.5 I'm a little concerned about destabilising 8.5.x but the CI system looks to be doing a good job of catching problems. Generally, I like the direction this is heading in. Reducing copying should improve performance but the impact of that doesn't seem to be noticeable so far. I do wonder why. Time to add some performance testing to my TODO list I think. Currently, there are essentially two code paths for read and write. 1. Original. User facing API uses byte[]. byte[] retained all the way to the SocketWrapper where it is copied to ByteBuffer. 2. New. User facing API uses ByteBuffer. ByteBuffer retained all the way to to the SocketWrapper where it is used if possible or copied to another ByteBuffer if not. I haven't explored this at all so far, but my initial impression is that a lot of duplicated functionality could be removed if the original code path copied to a ByteBuffer earlier. Mark --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org
