Richard Frith-Macdonald <[EMAIL PROTECTED]> writes:
> I've committed a change which moves the call to -gcFinalise into -
> dealloc and replaces the implementation of -release with:
>
> {
> M_LOCK(connection_table_gate);
> if (NSDecrementExtraRefCountWasZero(self))
> {
> NSHashRemove(connection_table, self);
> M_UNLOCK(connection_table_gate);
> [self dealloc];
> }
> else
> {
> M_UNLOCK(connection_table_gate);
> }
> }
I have tested this fix and it seems to work well.
I also understand now why my quick fix did not work.
The reason is that I introduced, naively, a new lock to guard the whole
release/dealloc
sequence. This will not work.
It should happen as above, the decrementing and testing of the refcount and
removal
from the hashtable should happen together inside a region locked by the
connection_table_gate.
Oh and the lock should be recursive, as I mentioned before. Because there are
calls to the
delegate in protected regions, and preventing release being called in the
delegate
(even on another connection) is a little bit harsh.
Next week, debugging the next annoying multithreading bug :-(
Wim Oudshoorn
_______________________________________________
Bug-gnustep mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/bug-gnustep
