> On Tue, 9 Apr 2002, Remy Maucherat wrote: > > > > Can you add a small comment in Response and clearly indicate who should > > > call it - I'm fine with any, but we should do it consistenlty in the 2 > > > adapters and in the 2 protocols. > > > > The class which is implementing the Adapter interface is calling it in the > > TC 4 code. > > Ok. I'll change tomcat3 and revert the change on 4.0 ( and add a comment > on adapter ). But first let's clarify how it's supposed to work, the > current Tomcat4Adapter doesn't seem to call it - it calls > Tomcat4Response.finish() but that didn't call Response.finish() before my > patch. > > > > outputBuffer.close(); > > should COMMIT > > There's no outputBuffer.close in the API - only in 4.0. > > Who should send the commit ?
The OutputBuffer there is the class in the same package, so it's not the one you're thinking about (yes, I know). > > coyoteResponse.finish(); > > should CLOSE > > Yes, but Tomcat4 Response finish() method didn't call > coyoteResponse.finish(). It is called by CoyoteAdapter now (or CoyoteProcessor in the old connector). > > For the HTTP/1.1 protocol: > > - Committing twice is not ok. > > - Closing twice is not ok. > > Of course, same for ajp or any other protocol. We should definitely commit > only once and close only once - the question is who is responsible for > that. It would be reasonable to be more robust. I added a flag for finish. > Coyote Response has a sendHeaders() method, which sets commited to true. > I think it should also call the COMMIT hook - if not the caller of > sendHeaders() ( Tomcat4 or Tomcat3 Response ) must do it. ( before my > fix it wasn't true - the COMMIt was sent by http11 before the first write > ). That one looks ok. It's not mandatory to call sendHeader, though (just writing something will commit and send the headers). > Also, Response has a finish() that calls CLOSE - but finish wasn't called > by tomcat4's Response, neither by the adapter ( which called > response.finishResponse() - but on the tomcat4 response, which didn't > propagate to the coyote.Response ). Ok. However, in the processor, endRequest is called, which does the same as a finish. So that would have done some bad stuff with the filters without the flag I added (not good). Remy -- To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]> For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
