On 08/12/2014 22:43, Rémy Maucherat wrote: > 2014-12-08 21:59 GMT+01:00 Mark Thomas <ma...@apache.org>: > >> What triggers the exception. That is the bit I don't understand. >> > The test is throwing an exception, it goes to onError which writes a > message, but the writer empty message gets sent first. From what I remember. > > To go further in this discussion, please provide information on where the > specification states empty messages are sent when using a writer or output > stream.
Zero length messages are valid as per RFC 6455 and getSendWriter() is for sending messages. There is no exclusion that I can see for zero length messages. So, I'd turn you question around and ask you to provide a reference that requires special handling in a writer for a zero length message. The TCK appears to be expecting that Session.getBasicRemote().getSendWriter().close() is a NO-OP whereas I would expect that to send a zero length message. I'd also point out that (at least in Tomcat's implementation) Session.getBasicRemote().getSendWriter(); Session.close(); does not send an empty message before the close message. With your change currently in place it is now impossible to send a zero length message using a Writer. I'd prefer to see this commit reverted but, given your requirement for an implementation that can pass the TCK - even if the TCK is wrong, then I'd accept this behaviour being disabled by default and enabled by yet another system property. Mark --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org
