oleaut32/tests: initialize pointer before sending it over the wire??

[Rob Shearman]

2008/6/21 Dan Kegel <dank at kegel.com>:
> While looking at the valgrind warning in
> http://kegel.com/wine/valgrind/logs-2008-06-20/vg-oleaut32_tmarshal.txt
>
>  Conditional jump or move depends on uninitialised value(s)
>    at  serialize_param (tmarshal.c:736)
>    by  serialize_param (tmarshal.c:744)
>    by  xCall (tmarshal.c:1414)
>    by  ???
>    by  func_tmarshal (tmarshal.c:1179)
>    by  run_test (test.h:449)
>    by  main (test.h:498)
>  Uninitialised value was created by a stack allocation
>    at  test_typelibmarshal (tmarshal.c:762)
>
> The problem happens during a call to this method
> where widget is a pointer to an uninitialized pointer
> which will receive the pointer to the widget:
>
>    interface IKindaEnumWidget : IUnknown
>    {
>        HRESULT Next(
>                     [out] IWidget **widget);
>
> I discovered that the attached patch prevented the problem.
> I don't quite understand why; at first glance,
> widget is an out parameter from the function,
> why would it be dereferenced while serializing
> the call?

It's a bug in the typelib marshaller. It doesn't check whether a
VT_PTR type is actually an interface pointer and not access it on
input when the parameter is an [out] parameter. Note that because of
the memory re-use semantics it is legal to access memory passed in to
a remote function, even when the parameter is [out].

I think it's getting close to the time to reimplement the typelib
marshaller on top of NDR functions so that we don't have to implement
these subtleties twice, would improve performance and would reduce the
amount of code.

-- 
Rob Shearman