On Apr 1, 2005, at 5:14 PM, Bob Hutchison wrote:

Hi,

I've been looking into that unbound memory consumption situation when directly connecting to TBNL (that doesn't happen when going through mod_lisp or Araneida and only seems to happen on OS/X).

I put some debugging statements into KMRCL and have (finally) noticed something that was blindingly obvious before.

There is a KMRCL function lw-worker in listener.lisp that is called for every request made to TBNL directly. It is called only once (or, at least, a very few times) if TBNL is invoked indirectly (i.e. via mod_lisp or Araneida). The call to this function is through a lambda that is defined in make-socket-server and passed to comm:start-up-server.

A bit more specific information. In the worker initializer there is a lambda being assigned to the thread-fun slot. In the case of mod_lisp->TBNL this function very rarely returns (but it does). In the case of browser->TBNL this function always returns. So it turns out that the tbnl::listen-for-request function is returning. Don't know why yet. How could this be chewing up memory? Where is the bug, surely returning from the listener-function should cause things to be cleaned up?

I am assuming at this point that this has something to do with the problem.

Now, I don't really understand what is going on here. I will continue to look into this, but if anyone happens to already know what is going on here I'd love to hear about it.

Cheers, Bob

---- Bob Hutchison -- blogs at <http://www.recursive.ca/hutch/> Recursive Design Inc. -- <http://www.recursive.ca/>