On 13 May 2005, at 15:33, Brian Grossman wrote:

This caused a memory leak, so I've had to go back to one connection per
accept_handler.

If we use check_earlytalker, each new connection goes into EventLoop()
for N seconds (via Client:can_read()), delaying the continuation of the
accept_handler loop.  The accept_handlers stack up? apparently without
bound.

Is there any other way to get the slick accept_handler?

Ah! I was thinking about this when I was looking at the code, but then I forgot about it and went along my merry business...


What we have to do is do all the accept()ing, and *then* call all the connect handlers after.

I'm also wondering if there's a better way to do can_read than going back into the event loop...

Matt.



Reply via email to