> Event is at the mercy of perl's reference counting mechanism. Event
> must delay deallocation of a cancelled watcher until perl releases all
> references to it.
Ah, I see.
> Events & ioevents are never deallocated. If you think this is a
> problem then I can provide a function to deallocate everything on the
> freelist. However, I doubt that this is worthwhile.
Hm, what does it mean? If it does not result in an ongoing memory allocation while
thousands and thousands of IO watchers are installed and cancelled - as I wrote
before, the servers and clients should run for very long periods -, this will be no
problem from my point of view. But IF this means that more and more memory is
allocated ... hm, then I think there should be a release method. On the other hand,
I suppose there was a reason for the current design ...
Just to understand it better - Events and ioevents are collections of all (specific)
events detected while a loop is running?
> > Nevertheless, the patched module is now installed and the server already runs
> > on its base. Hopefully it no longer grows ...
>
> Fantastic! I'll upload a new release of Event to CPAN soon...
I think it is real improvement, but a few minutes ago I had to recognize that the
server grows again (after a period of constant memory allocation) ... I wished it
would not, really. I will implement the checks. But, of course, my be there's a
fault in my own code.
Jochen
