The parameter controls whether weak references are ignored, not
whether the last stub_manager_int_release should be called when the
total reference count goes to zero.
This is so that an object will only be released when a weak-reference is
used after the last weak reference has been released (either by
CoReleaseMarshalData or CoUnmarshalInterface + Release).
Copy the OXID_INFO from the stub manager, if available, in
proxy_manager_construct. Otherwise, attempt to resolve it.
Use ipidRemUnknown from OXID_INFO in the proxy manager instead of
magically generating the IPID for RemUnknown.
Create a server channel and store this in the ifstub.
Rename ipid_to_apt_and_stubbuffer to ipid_get_dispatch_params and make
the function return an HRESULT.
Return the channel for the ifstub in ipid_get_dispatch_params.
Change stub manager to track the number of normal marshals instead of
using the state machine so that multiple marshals of the same object
and interface work correctly.
pointing to the completely wrong object when a proxy does a
queryinterface.
- Remove assumption that the stub buffer will handle the lifetime of
the object.
- Move the modal loop called during RPCs into CoWaitForMultipleHandles.
- Use a mutex for long remoting calls to IRemUnknown methods.
- Remove locking in apartment_disconnectproxies as it is not needed.
- Use PostMessage instead of SendMessage so we can run the message
loop or not as appropriate.
- Rename register_ifstub to marshal_object to more accurately describe
what it does.
- Add new function, apartment_getoxid, to prepare for a possible
future patch where remoting is started on demand.
- Change CoLockObjectExternal so that it does the correct action now
and eliminate a fair few lines of now redundant code.
- Rename OLE32_Dll{Register,Unregister}Server to
Dll{Register,Unregister}Server.
- Rework RPC dispatch layer to be simpler and not get confused by
server/client duality.
- Make threads shut down at the right time and not access freed memory
after apartment destruction.
- Rename stub_dispatch_thread to client_dispatch_thread.
- Add some more tracing
- Check return value of WaitNamedPipe.
- Change named pipe timeouts to 0.5s, which should be enough for even the slowest machines.
show that IInternalUnknown isn't exposed.
- Implement IMultiQI on top of IRemUnknown calls.
- Silence some fixmes that occur during tests and don't give us any
useful information.
- Fix typo in class factory proxy that caused us to use the wrong
offset into the CFProxy structure, causing us to not call the
outer_unknown properly.
Francois Gouget