On 04/12/2014 13:17, Rémy Maucherat wrote: > 2014-12-04 14:11 GMT+01:00 Mark Thomas <ma...@apache.org>: > >> The used flag is only ever set to true. It needs to be reset to false >> after a write to avoid multiple flush() calls one after the other >> triggering empty writes. >> >> Looking at the code, what would be better is if something along the >> lines of using buffer.flip(), buffer.remaining(). >> >> I'm looking at the code now so I'll put a test case and a patch together. >> > I made one further commit on this, but it is lost in limbo at the moment.
Ah. I see that now. My code reviews for the WebSocket changes have fallen too far behind. I'm working on catching up. > I would recommend not touching anything for now. I might still write the test case. Nothing is going to happen until svn returns to read/write anyway. > The purpose of the flag is so that close always writes a last part unless > nothing was ever written. Shouldn't there still be a single empty message in that case? Otherwise, how is the client meant to differentiate between no write and a write of zero length? > I did add a if (buffer.position > 0) in flush to > avoid useless empty writes. I see that now. Mark --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org