Note: this seems to be the same problem as #14222, but I can't seem to
append stuff to someone elses bug so...

This is something I first noticed when I upgraded from XP RC1 to WinXP
Final, when using PHP 4.0.6.  I upgraded to 4.1.0 today, and it doesn't
solve the problem.

The php apache module doesn't seem able to handle outputting moderate to
large sized pages.

I have been able to reliably reproduce the problem with the following
script:

<?PHP
set_time_limit(900);
for ($i = 0; $i<1000000; $i++){
print "This is line number " . ($i+1) . "<br>";
}
?>

Viewing this page with internet explorer causes it to go into an almost
endless loop of relaoding the page.  It will display the first few 'this
line is number...' and then will reload the page over and over.  On some
sessions IE will eventually show it's 'The page cannot be displayed'
page, and sometimes it will just reload indefinitly.

Viewing this page with Mozilla/Netscape doesn't have the same effect. 
Mozilla doesn't go through the reload loop, but will not show the page
as it should either.  There will be unexplained (no errors) jumps in
numbers/missing output such as:

This is line number 2368
This is line number 2369This is line number 2517
This is line number 2518

But the errors in output don't occur in the same spot each time.  And
evenutally (well short of 1,000,000 lines) the output will just stop
with no error,  often in mid line.

I have found that adding 'flush();' just after the 'print "This is
line...' seems to fix the problem.

This problem reveals a memory limit bailout problem.
Even if PHP exhsusted memory, script does not exit.

PHP logs following logs many times. (Linux/PHP4.1.0, without output
buffering)

[13-Dec-2001 10:27:52] PHP Fatal error:  Allowed memory size of 8388608
bytes exhausted (tried to allocate 10240 bytes) in
/home/yohgaki/public_html/bugs/14474/bug.php on line 4

Type is changed to Scripting Engine Problem.

I think this is critical

I don't think that our current analysis is correct.  Take a look at the
access log - I'm pretty sure you'd see that the page is being repeatedly
requested by IE, and not requested only once.

Something about the way the server disconnects may cause IE to think
that the page was not properly fetched, and make it try to reload it. 
That would be my guess...

Probably, Zeev is right about regarding 
[2001-12-12 20:33:45] yohgaki@php.net
update.  Since I didn't see active httpd process.

httpd should close connection when PHP cannot execute script, anyway ;)
(I suppose httpd is not closing connection. With my IE under w2k,
networks may become ususable. Mozilla under linux halts. This is
critical :)

More detailed analysys is required. Any volanteers?

I've tried the script on PHP 4.0.6, Windows 2000 Pro, Apache 1.3.22, php
installed as a module.
If I use a download manager to access the script it works.
I use Mass Downloader 2.2 and to get the file I write something like
this to the url address: http://localhost/bug.php.
The program starts downloading the file and after downloading 28.55 Mb
stops, but it doesn't say that the download finished and after 5 sec. it
starts downloading the file again. If I set the for loop only to 100 the
result is the same, after the program transfers 2.5 Kb it waits for 5
sec. after he tryes again to download it.
But if I try this with IE (when the loop is set to 100) everything it's
ok.
So, after running this tests, I think that the problem is not that PHP
does not support large outputs, but the browser. (28.55 Mb it's a huge
html file).
I've tested IE with the loop set to 100000 and it works.
If I set the loop to 1000000 then I ran aout of memory, because my page
file is only of 380 Mb. But maybe if I would have more memory everything
would be OK.

I've used a packet logger and have found that PHP seems to halt
transmission around a php instruction.
The socket is actually closed, I'm not sure if there needs to be an
'END' string at the end of an http session, but there are no CR's or
anything like that on the end.

I've used a packet logger and have found that PHP seems to halt
transmission around a php instruction.
The socket is actually closed, I'm not sure if there needs to be an
'END' string at the end of an http session, but there are no CR's or
anything like that on the end.

Adding to my previous comment, I saw no:

<CR>
0
<CR>
<Disconnect>

at the end of my http session.
OK - The HTTP Request is not closed properly. Only got:

blahblah blah blah<br>
<CR>
blafjaskasdjadsjf<font size="1"$sample_write[$i]
<Disconnect>

There must be an <CR>0<CR> at the end of the reply.

Seems like a bug in IE to Markus, Jani, Tal and me.
closing