Ticket #188 (closed enhancement)
Not a ticket for this component, but the request enhancement for the web server
|Reported by:||anonymous||Owned by:||jaymccarthy|
|Component:||untyped/dispatch.plt||Keywords:||web server response incremental comet|
Hello! Sorry for this 'transit-request' post.
Please move following enhancement ticket to the web server team.
The current make-response/incremental (in response.ss) has two
disadvantages: 1st can lead to browser-confusing chunked response
technology) programming - there is no way to flush out selected
chunks during paused response/incremental (with sleeps).
Please consider following patch, which should eliminate both issues. It interprets zero-length chunk simply as `flush-output' request for previous chunks. Comet now runs well using
'(do-loop (wait-and-send-next-event) (send/bytes #""))' pattern.
--- collects/web-server/http/response.ss.orig 2009-06-05 10:40:07.747989122 +0400
+++ collects/web-server/http/response.ss 2009-06-05 10:57:55.674988779 +0400
@@ -110,10 +110,13 @@
- (fprintf o-port "~x\r\n"
- (apply + 0 (map bytes-length chunks)))
+ (let ((len (apply + 0 (map bytes-length chunks))))
+ (if (= 0 len) ; Protects against zero-length-chunk bug
+ (flush-output o-port) ; ... and makes Comet happy
+ (fprintf o-port "~x\r\n" len)
(for-each (lambda (chunk) (display chunk o-port)) chunks)
- (fprintf o-port "\r\n")))
+ (fprintf o-port "\r\n"))))))
; one \r\n ends the last (empty) chunk and the second \r\n ends the (non-existant) trailers
(fprintf o-port "0\r\n\r\n")))]))