Re^11: Proper undefine queue with multithreads

It's not locking long enough. There's a step by step demonstration of the race condition in the linked post.

Maybe there is a chance of this occuring with sleep 0.01 but BrowserUk originally had sleeps of 1 -- I doubt this race condition could ever manifest with more than one worker .... esp when you're not using pending (like originally)... but even with the ops changed code and pending

Isn't it simpler to just

for( 1..10 ){ ## make sure we're not between DQ and $busy++
    sleep 1;
    sleep 1 while $q->pending || $busy;
}
[download]

I can't quantify the odds exactly other than to say probably never

Comment on Re^11: Proper undefine queue with multithreads Download Code

Replies are listed 'Best First'.
Re^12: Proper undefine queue with multithreads by ikegami (Patriarch) on Jun 07, 2014 at 13:27 UTC
I doubt this race condition could ever manifest with more than one worker It will either affect all workers (race condition occurred) or none (race condition didn't occur). It can't affect just one (unless there's a second race condition I didn't see).	[reply]