Interesting sleep() side-effect

In testing Thies' connection abort detection code, completely unrelated to his stuff, I noticed an interesting side-effect of calling sleep() in a PHP script. I have tried this on both Linux and Solaris and get the same results. Could you guys with access to other OS'es try this one? <? set_time_limit(5); register_shutdown_function("done"); function done() { global $i; $fp = fopen("/tmp/done","w"); fputs($fp,"i reached $i\n"); fclose($fp); } for($i=0; $i<30000; $i++) { for($j=0;$j<1000;$j++) $a=sqrt($j); /* sleep(1); */ echo "$i ................................................................................<br>"; flush(); } ?> This script should run for 5 seconds, unless of course you are on a massively fast machine that can do 30000 sqrt()'s in 5 seconds. However, when you uncomment the sleep(1) call there, it will happily run way past 5 seconds. Looks to me like the sleep() system call is interfering with our itimer. The man pages state that sleep() may be implemented as a SIGALRM, but we are using a SIGPROF itimer, so I thought we were actually safe here. It doesn't worry me too much. It is more an interesting side effect, but something we might want to document as a caveat in the sleep() documentation. I would be intersted in hearing if this behaviour is consistent across all the operating systems. The "/tmp/done" file is not related to this particular message. That's just how I am testing to see if PHP catches when the client closes the socket. -Rasmus -- PHP Development Mailing List http://www.php.net/ To unsubscribe send an empty message to php-dev-unsubscribe@lists.php.net For help: php-dev-help@lists.php.net