[Stackless] Stackless crash

[Jeff Senn]

Richard Tew wrote:
> On 6/19/06, Jeff Senn <senn at maya.com> wrote:
>> This (tiny bit of) code causes stackless crash on OS-X.
...

> Or a channel being garbage collected.  Looks like a straightforward
> fix, the channel needs to hold a reference to itself while there are 
> tasklets
> blocked on it.
> 
> Care to provide a patch for this one? :)

Hm... I see... seems a little "wrong" for the channel to hold a reference
to itself (but I really unsure how to the do the opposite: have the tasklet
hold the reference to the channel) and other things sometimes *seem* wrong
about stackless too :-) :-)

The patch is pretty easy (included below), but I'm not *certain* it doesn't
change some basic refcnt assumption somewhere...

-Jas


Index: Stackless/module/channelobject.c
===================================================================
--- Stackless/module/channelobject.c    (revision 47111)
+++ Stackless/module/channelobject.c    (working copy)
@@ -64,6 +64,7 @@
  slp_channel_insert(PyChannelObject *channel, PyTaskletObject *task, int dir)
  {
         SLP_HEADCHAIN_INSERT(PyTaskletObject, channel, task, next, prev);
+       if(!channel->balance) Py_INCREF(channel); /*add a reference to myself 
while there is a balance*/
         channel->balance += dir;
         task->flags.blocked = dir;
  }
@@ -76,6 +77,7 @@
         assert(PyTasklet_Check(ret));

          channel->balance -= dir;
+       if(!channel->balance) Py_DECREF(channel); /*remove the extra self 
reference when balance falls to 0*/
         SLP_HEADCHAIN_REMOVE(ret, next, prev);
         ret->flags.blocked = 0;
          return ret;
@@ -91,6 +93,7 @@

         assert(PyTasklet_Check(task));
         channel->balance -= dir;
+       if(!channel->balance) Py_DECREF(channel); /*remove the extra self 
reference when balance falls to 0*/
         SLP_HEADCHAIN_REMOVE(task, next, prev);
         task->flags.blocked = 0;
         return task;

_______________________________________________
Stackless mailing list
Stackless at stackless.com
http://www.stackless.com/mailman/listinfo/stackless