Just tried with php5-200509081630:

PHP 5.1.0-dev (cgi-fcgi) (built: Sep  8 2005 15:09:53)

No luck either. Output still comes at once.

Tried with apache2 + php-4.3.11 - the problem is present there too.

And strace of php process looks like:

read(4, "<?php\n\tfor ($i=0; $i<4; $i++){\n\t"..., 8192) = 65
read(4, "", 4096)                       = 0
read(4, "", 8192)                       = 0
close(4)                                = 0
munmap(0x4056b000, 4096)                = 0
rt_sigprocmask(SIG_BLOCK, [CHLD], [], 8) = 0
rt_sigaction(SIGCHLD, NULL, {SIG_DFL}, 8) = 0
rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
nanosleep({1, 0}, {1, 0})               = 0
rt_sigprocmask(SIG_BLOCK, [CHLD], [], 8) = 0
rt_sigaction(SIGCHLD, NULL, {SIG_DFL}, 8) = 0
rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
nanosleep({1, 0}, {1, 0})               = 0
rt_sigprocmask(SIG_BLOCK, [CHLD], [], 8) = 0
rt_sigaction(SIGCHLD, NULL, {SIG_DFL}, 8) = 0
rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
nanosleep({1, 0}, {1, 0})               = 0
rt_sigprocmask(SIG_BLOCK, [CHLD], [], 8) = 0
rt_sigaction(SIGCHLD, NULL, {SIG_DFL}, 8) = 0
rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
nanosleep({1, 0}, {1, 0})               = 0
setitimer(ITIMER_PROF, {it_interval={0, 0}, it_value={0, 0}}, NULL) = 0
write(3, "\1\6\0\1\0@\0\0Content-type: text/html\r"..., 96) = 96

I.e. there are 4 calls to sleep and the script output is writen at once. Any
pointers to documentation or further tests that I should perform are welcome.

How would it be useful to be able to turn it off?

This is how you could write a chat client for example. Connection is kept open for the duration of chat session and php script is looping over the client and server messages.

Writing such things using PHP is pretty useless IMO, but if it works with other SAPIs..

It could be also used for writing progress bars. On a random shared hosting server:

locate .php |wc -l
  25147
locate .php |xargs -n1 -i grep -H -e "flush *(" {}
    641

So people are using it for some reason... I can see flush() used in Moveable Type, Word Press, MediaWiki and so on.

This would also be useful for Ajax (using, for example, the RicoAjaxEngine.) If you are unable to disable buffering, there is no way to stream XML for progress updates through Ajax.

I have the same problem.. What I did was to send a lot of blank space to fill up the buffer:

add_message($id, str_pad('',4096));

That fixed the problem but increases the bandwidth.

I've the same problem... At the newest php, on Debian and FreeBSD, doesn't matter, using fastcgi compiled in... Any suggestions about this problem?!

It is a very important bug for me, because I'd like to give webpages to users by parts, compiling them on fly!-)

So, here is a trivial patch, that works... Sent it to [email protected]
Hope, this will close this bug.

? khext.patch
Index: sapi/cgi/cgi_main.c
===================================================================
RCS file: /repository/php-src/sapi/cgi/cgi_main.c,v
retrieving revision 1.271
diff -u -r1.271 cgi_main.c
--- sapi/cgi/cgi_main.c 3 Dec 2005 01:27:37 -0000       1.271
+++ sapi/cgi/cgi_main.c 6 Dec 2005 21:58:18 -0000
@@ -264,13 +264,15 @@

 static void sapi_cgibin_flush(void *server_context)
 {
-       if (!FCGX_IsCGI()) {
+       int flush_result;
+    if (!FCGX_IsCGI()) {
                FCGX_Request *request = (FCGX_Request *) server_context;
-               if (
+               flush_result = FCGX_FFlush(request->out);
+        if (
 #ifndef PHP_WIN32
                !parent &&
 #endif
-               (!request || FCGX_FFlush(request->out) == -1)) {
+               (!request || flush_result == -1)) {
                        php_handle_aborted_connection();
                }
                return;

Fixed in CVS HEAD, PHP_5_1 and PHP_5_0.

I've just tested it and can confirm that flush() now works as expected. Thank you.

This problem exist on php v5.3.13 or php v5.3.2 stable.
Please fix it.