summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authornanbor <nanbor@ae88bc3d-4319-0410-8dbf-d08b4c9d3795>1999-07-21 10:05:59 +0000
committernanbor <nanbor@ae88bc3d-4319-0410-8dbf-d08b4c9d3795>1999-07-21 10:05:59 +0000
commit6d1c286aa59a8665c741c89b3b8760327763238e (patch)
treeee52f1537291ad63a2a1697fb103f7d0395cebe8
parentbec226a7461df45ba15da87c1e3e1553c7940ffb (diff)
downloadATCD-6d1c286aa59a8665c741c89b3b8760327763238e.tar.gz
*** empty log message ***
-rw-r--r--ChangeLog-99b84
1 files changed, 84 insertions, 0 deletions
diff --git a/ChangeLog-99b b/ChangeLog-99b
index f47efa5bea0..be5c7e36461 100644
--- a/ChangeLog-99b
+++ b/ChangeLog-99b
@@ -1,3 +1,87 @@
+Wed Jul 21 04:32:20 1999 Nanbor Wang <nanbor@cs.wustl.edu>
+
+ * ace/Reactor.cpp:
+ * ace/Reactor.h:
+ * ace/Reactor.i:
+ * ace/Reactor_Impl.h:
+ * ace/Select_Reactor_T.cpp:
+ * ace/Select_Reactor_T.h:
+ * ace/Select_Reactor_T.i:
+ * ace/TP_Reactor.cpp:
+ * ace/WFMO_Reactor.cpp:
+ * ace/WFMO_Reactor.h:
+ * ace/WFMO_Reactor.i: The change was motivated by removing a race
+ condition in TP_Reactor during shutdown. There was a small time
+ gap between a thread running TP_Reactor event loop checking its
+ end_event_loop flag to the thread actually grabing the token in
+ TP_Reactor. Since the token return to its "normal" state after
+ broadcasting, this thread has no chance to notice that the
+ reactor has already been shutdowned and thus a deadlock occured.
+ In order to fix the problem, we must keep a state indicating
+ whether the reactor has been shutdowned or not in the reactor.
+
+ An added benefit of adding this state is that every reactor now
+ has its own run_reactor_event_loop which mimic the behavior of
+ the static ACE_Reactor::run_event_loop but instead of working on
+ the singleton reactor, the added function work on an instance of
+ reactor object. In fact, all event_loop control functions have
+ been added. They are named *_reactor_event_loop to
+ differentiate their behavior. These newly added function use
+ the added state in the reactor to control the looping of event
+ handling. All run_reactor_event_loop methods take an extra
+ argument of a function pointer. The function will be called in
+ every iteration if it is set.
+
+ Notice that the singleton event loop control functions are still
+ available. Although these function merely forward the call to
+ the singleton reactor's *_reactor_event_loop functions, you can
+ still use them to control the "main" event loop. In fact, it is
+ required that you use these functions to run the main event loop
+ because it facilitates Service_Configurator's "reconfigure"
+ feature.
+
+ All reactors now also have two extra functions namely
+ "deactivated" which is used to query whether the reactor will
+ handle more incoming event or not, and "deactivate" which is
+ used to control this new feature.
+
+ An side effect of the change is that, once a reactor get
+ deactivated, calls to handle_events will return -1 immediately.
+ You can use the "deactivated" method to check if there's
+ actually an error in the handle_events method or it simply
+ returned because the reactor has been deactivated. This is also
+ how the various "run_reactor_event_loop" handle the situation.
+ Notice that this should not cause any backward compatibility
+ problem becuase if a user is using his own end_event_loop flag,
+ he will not (and should not) invoke any more handle_events after
+ the flag has been set. Also, if he is using the global
+ run_event_loop methods (for controling singleton reactor,) they
+ should still behave the same.
+
+ To summarize the change:
+
+ 1. The run_event_loop control logic has been "pushed" down into
+ each individual reactor. The methods supporting the logic are
+ defined in ACE_Reactor and are:
+
+ run_reactor_event_loop ()
+ run_alertable_reactor_event_loop ()
+ end_reactor_event_loop ()
+ reactor_event_loop_done ()
+ reset_reactor_event_loop ()
+
+ 2. The origianl run_event_loop methods in ACE_Reactor are still
+ working but they now use the run_reactor_event_loop methods in
+ the implementation of the reactor.
+
+ 3. Reactors now keep a state to indicate handle_events should be
+ called or not. They can be controlled thru:
+
+ deactivated ()
+ deactivate ()
+
+ methods.
+
Tue Jul 20 21:31:50 1999 Jim Rogers jrogers@viasoft.com
* ace/config-mvs.h: added ACE_HAS_EXCEPTIONS.