Posted by Rasmus Lerdorf on 09/02/05 22:38

Rasmus Lerdorf wrote:
> Chris Shiflett wrote:
>
>>Rasmus Lerdorf wrote:
>>
>>
>>>Why modify my test?
>>
>>
>>Because it has less delay. Thus, it's more difficult to tell if the
>>browser is requesting the new URL before or after receiving the entire
>>response. My script is essentially the same thing, but the script takes
>>30 seconds to execute. It makes the distinction very clear.
>>
>>The addition of flush() forces the chunked response. You can remove that
>>if your server uses chunked transfer encoding without it.
>>
>>
>>>What did you see in log.txt from my version and on your screen?
>>
>>
>>You would see the output "Count0\nCount1\n..." in the log and the PHP
>>web site in the browser. I'm not sure how that's relevant. No browser is
>>going to render content from a 302 response, but that doesn't prove that
>>it won't wait for it.
>
>
> Yes it does. The last number in log.txt tells you exactly when the
> browser stopped listening to the response and closed the socket because
> PHP will abort the script at that point. If what you are saying is
> true, how do you explain the fact that you don't see a count all the way
> up to 999,999 in my test in the log.txt file?

Also, just add a single line to your own test script and make it look
like this:

header('Location: http://www.php.net/');
$fp = fopen('/tmp/log.txt', 'w');
for ($i = 0; $i < 30; $i++)
{
$str = "Count $i\n";
echo str_repeat($str,1000);
fputs($fp, $str);
sleep(1);
flush();
}

Then time how long it takes for the redirect to happen. Is it still
taking 30 seconds? If not, why not?

-Rasmus

[Back to original message]