[Bug 17195] NamedPipe datagrams need to be _really_ datagrams

http://bugs.winehq.org/show_bug.cgi?id=17195

--- Comment #17 from Juan Lang   2009-02-07 19:23:53 ---
(In reply to comment #16)
> - start off in non-message-mode and switch to message-mode with data still
>
> unread

I believe this isn't allowed.  See the existing tests,
dlls/kernel32/tests/pipe.c, lines 213-214:
    /* trying to change the client end of a byte pipe to message mode should
fail */
    ok(!SetNamedPipeHandleState(hFile, &lpmode, NULL, NULL), "Change mode\n");

On the other hand, it's possible that these tests only test the client. 
Perhaps the server can change modes?  I suspect not, but you're right that a
test is needed.
> - start off in message-mode and switch to non-message-mode with data
>
> outstanding and not outstanding, as two separate tests.

I believe this is possible, at least for the client, in contrast to test 2).  I
don't believe there are tests for this though.