The tests intermittently crash on 2008 with RPC_S_SERVER_TOO_BUSY. I suspect
this is due to the server spawned in run_server having a call limit of 20 but
being expected to service 32 threads. Use the default parameter in an attempt
to prevent this failure.
Signed-off-by: Zebediah Figura <z.figura12@gmail.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Fixes a leak in typelib marshaller.
Based on patch by Kevin Puetz.
Signed-off-by: Jacek Caban <jacek@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Fixes regression caused by typelib marshaling rewrite.
Spotted by Kevin Puetz.
Signed-off-by: Jacek Caban <jacek@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
The problem with NtFsControlFile(FSCTL_PIPE_PEEK) is that it happily returns
STATUS_SUCCESS with 0 bytes of data available. That makes the client think
that the server already sent a reply and try to read the reply data, which
hangs. A solution to this is call NtReadFile() with zero sized buffer, which
has correct behaviour: just waits for incoming data to appear without mangling
the data in the pipe.
Signed-off-by: Dmitry Timoshkov <dmitry@baikal.ru>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
This is required to correctly support out reference parameters,
otherwise server method receives not initialized pointers during
the CALLSERVER phase and crashes.
Signed-off-by: Dmitry Timoshkov <dmitry@baikal.ru>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
RPC_CONTEXT_HANDLE_FLAGS contain the values RPC_CONTEXT_HANDLE_SERIALIZE
or RPC_CONTEXT_HANDLE_DONT_SERIALIZE, however all the calls to underlying
binding->Assoc are guarded by a critical section, so it's redundant to
take any action about passed in Flags parameter.
Signed-off-by: Dmitry Timoshkov <dmitry@baikal.ru>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Based on NdrStubCall2 and NdrAsyncClientCall implementations.
In order to test asynchronous RPC I used dlls/rpcrt4/tests/server.c as a base,
converted int_return() and sum() to use asynchronous RPC on both client and
server sides (server.idl doesn't need any changes), and added server.acf:
interface IServer
{
[async] int_return();
[async] sum();
}
With this implementation these tests pass under Wine. Since widl doesn't
support asynchronous RPC it's impossible to add appropriate Wine tests.
Signed-off-by: Dmitry Timoshkov <dmitry@baikal.ru>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Otherwise the caller may decide to wait for the call termination which
will never happen.
Signed-off-by: Dmitry Timoshkov <dmitry@baikal.ru>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
NdrClientCall already uses TRY/EXCEPT block if the procedure has
Oi_HAS_COMM_OR_FAULT in its Oi_flags, I've added similar block
to NdrAsyncClientCall but without mapping the fault code using
CommFaultOffsets because NdrAsyncClientCall doesn't have this
implemented.
An installer that I have here calls NdrAsyncClientCall in order
to query status of a not yet installed service, and that leads
to a crash due to RaiseException from inside of NdrAsyncClientCall.
With this patch it no longer crashes on an unhandled exception,
and then installs the service.
Signed-off-by: Dmitry Timoshkov <dmitry@baikal.ru>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Most notably in NdrProxyInitialize().
This patch and the next bring the scope of exception handling roughly in line
with -Os stubs.
Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=36981
Signed-off-by: Zebediah Figura <z.figura12@gmail.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Jacek Caban