doc/COROUTINE_MANAGEMENT

Wed, 15 Jul 2009 10:39:10 -0400

author
Thomas Harning Jr <harningt@gmail.com>
date
Wed, 15 Jul 2009 10:39:10 -0400
changeset 64
dd47dbe85bec
parent 39
a0a7f7ebca3f
permissions
-rw-r--r--

Merge branch 'v0.2'

Due to the issue w/ self-resuming threads and crashing out threads, 
a management system needs to be in place.

Example thread system:

MAIN
EVENT_LOOP --------running---
WAITING ON READ
WAITING ON WRITE
WAITING ON CONNECT


Since main and the other 'waiting' threads are yielded, it is unsafe to call things arbitrarily on them 
or resume them from themselves...
However the EVENT_LOOP one is running and thus can execute the callbacks (which can resume the threads)
Each of the 'waiting' events are attached to an event and contain a pointer, this pointer can be setup to point
to a per event_base item which will be updated w/ the lua_State of whatever calls EVENT_LOOP...
this will guarantee that the thread will be resumed from the currently running EVENT_LOOP


Other system that's more complicated and less likely:

MAIN
EVENT_LOOP a -----running---

WAITING ON READ a
WAITING ON WRITE a

EVENT_LOOP b ----yielded
WAITING ON READ b


Since there can only be one event_loop running per event_base, you do not have to worry about 
cross-pollination of the different waits...

NOTES:
If the event_loop thread goes away... then the waiting coroutines will have no way to get back...
though in this case, they are dead in the water anyways.. until a new event_loop starts...
in which case the lua_State references has been updated...

mercurial