Re: server: Fix for possible desktop variable being NULL when set_process_default_desktop is called

[Robert Shearman]

Alexandre Julliard wrote:
> Robert Shearman  writes:
> > /* check for an inherited winstation handle (don't ask...) */
> >   if ((handle = find_inherited_handle( process, &winstation_ops )))
> >   {
> >       winstation = (struct winstation *)get_handle_obj( process,
> > handle, 0, &winstation_ops );
> >   }
> >   else if (parent && parent->process->winstation)
> >   {
> >       handle = duplicate_handle( parent->process,
> > parent->process->winstation,
> >                                  process, 0, 0, DUP_HANDLE_SAME_ACCESS );
> >       winstation = (struct winstation *)get_handle_obj( process,
> > handle, 0, &winstation_ops );
> >   }
> >   if (!winstation) goto done;
> >   process->winstation = handle;
> >
> > Handle is guaranteed to be non-NULL here, since winstation and handle
> > are the same object (well, except if duplicate_handle fails).
>
> Yes, it's non-NULL here, but that's irrelevant since it's reset on the
> next line.

Sorry, you are correct. It does indeed look like a flaw in the scanner:

At conditional (5): "handle = find_inherited_handle != 0" taking false path

298          if ((handle = find_inherited_handle( process, &desktop_ops )))
...
Event var_deref_model: Variable "desktop" tracked as NULL was passed to 
a function that dereferences it. [model]
Also see events: [assign_zero]
At conditional (8): "handle != 0" taking true path
311          if (handle) set_process_default_desktop( process, desktop, 
handle );

It doesn't seem to have tracked that handle was assigned zero on line 298.