[PATCH] ole32: Implement CreateObjrefMoniker().

[Dmitry Timoshkov]

Nikolay Sivov <nsivov at codeweavers.com> wrote:

> > +static const IMonikerVtbl VT_ObjrefMonikerImpl =
> > +{
> > +    ObjrefMonikerImpl_QueryInterface,
> > +    PointerMonikerImpl_AddRef,
> > +    PointerMonikerImpl_Release,
> > +    ObjrefMonikerImpl_GetClassID,
> > +    PointerMonikerImpl_IsDirty,
> > +    PointerMonikerImpl_Load,
> > +    ObjrefMonikerImpl_Save,
> > +    PointerMonikerImpl_GetSizeMax,
> > +    PointerMonikerImpl_BindToObject,
> > +    PointerMonikerImpl_BindToStorage,
> > +    PointerMonikerImpl_Reduce,
> > +    PointerMonikerImpl_ComposeWith,
> > +    ObjrefMonikerImpl_Enum,
> > +    PointerMonikerImpl_IsEqual,
> > +    PointerMonikerImpl_Hash,
> > +    PointerMonikerImpl_IsRunning,
> > +    ObjrefMonikerImpl_GetTimeOfLastChange,
> > +    PointerMonikerImpl_Inverse,
> > +    PointerMonikerImpl_CommonPrefixWith,
> > +    PointerMonikerImpl_RelativePathTo,
> > +    ObjrefMonikerImpl_GetDisplayName,
> > +    PointerMonikerImpl_ParseDisplayName,
> > +    ObjrefMonikerImpl_IsSystemMoniker
> > +};
> > +
> I think separate implementation would be better. According to docs more
> methods are supposed to differ, like running state handling, or
> Load()/GetSizeMax().

Initially I created a fully separate implementation, however after looking
at the test results I moved to a less intrusive version. I think that once
a method that differs in behaviour is found it's fairly easy to add another
standalone implementation.

> It's worth checking if objref moniker keeps a reference to passed
> pointer at all. Since its purpose is to identify running object RPC-way,
> I suspect it's not a simple pointer wrapper.
> 
> > +static PointerMonikerImpl *unsafe_impl_from_IMoniker(IMoniker *iface)
> > +{
> > +    if (iface->lpVtbl != &VT_PointerMonikerImpl && iface->lpVtbl != &VT_ObjrefMonikerImpl)
> > +        return NULL;
> > +    return CONTAINING_RECORD(iface, PointerMonikerImpl, IMoniker_iface);
> > +}
> This implies that IsEqual() could return S_OK for mismatching moniker
> types, which is not backed by tests, or docs.

I guess that it could be a follow up patch. I should probably mention that
this implementation works perfectly for a very large .Net based application.

-- 
Dmitry.