Re: Typelib marshalling BSTRs

[Robert Shearman]

> Subject: Re: Typelib marshalling BSTRs
> From: Mike Hearn 
> To: "Gregory M. Turner" 
> Cc: [e-mail]
> Date: 23 Jul 2003 16:28:58 +0100
>
> So, moving on from that problem, I think my woes (currently) are caused
> by the NDR engine not handling VT_DISPATCH variants.
>
> in wire_size():
>   case VT_DISPATCH:
>     FIXME("wire-size interfaces\n");
>
> that causes VARIANT_UserMarshal to skip doing anything special for
> VT_DISPATCH variants, when I think in reality, it should call
> NdrInterfacePointerMarshall (btw, what's up with two spellings of
> "marshal"?) to get the IDispatch* into the stream.

I don't know about NdrInterfacePointerMarshall, but I don't think it should
get there to start with (see below). BTW, I think the two spellings depends
on whether the word is used as a verb or as a noun.
> While rather roundabout, I think it seems clear that this is what's
> needed. The problem then becomes one of NdrInterfacePointerMarshall()
> needing information that isn't available from inside
> VARIANT_UserMarshal, like the MIDL_STUB_BUFFER pointer.
>
> Greg, as you are the relevant Guru here and Ove is probably a bit pissed
> off with all my questions, what should we be doing here? If we are meant
> to use NdrInterfacePointerMarshall to handle VT_DISPATCH variants, how
> do we get the args it needs?

Ok, my name's not Greg or Ove, but I'll have a go anyway. I think the code
should not try to marshal/unmarshal the data in rpcrt4, but should use the
PSOAInterface code in oleaut32 (according to the registry entry for
Interfaces[CLSID of IWebBrowser2]\ProxyStubClsid32). It would probably be
best to find out why this is not happening (maybe by moving some oleaut32
code to rpcrt4?).

Rob

Typelib marshalling BSTRs

[Mike Hearn]

On Wed, 2003-07-23 at 23:04, Robert Shearman wrote:
> I don't know about NdrInterfacePointerMarshall, but I don't think it should
> get there to start with (see below). BTW, I think the two spellings depends
> on whether the word is used as a verb or as a noun.

Yeah, I considered that, but MSDN seems to use marshal as a verb as
well. *shrug*
> I think the code
> should not try to marshal/unmarshal the data in rpcrt4, but should use the
> PSOAInterface code in oleaut32 (according to the registry entry for
> Interfaces[CLSID of IWebBrowser2]\ProxyStubClsid32). It would probably be
> best to find out why this is not happening (maybe by moving some oleaut32
> code to rpcrt4?).

Well, the code doesn't actually marshal IWebBrowser2, it operates
entirely in terms of IDispatch interfaces. Once the dispatch interface
for the WebBrowser control is marshalled into another thread, the code
starts poking it via Invoke, in this case retrieving the Document
property which also returns a dispatch interface.

I'm pretty sure just marshalling IWebBrowser2 would work, but
unfortunately the nature of the Java-COM bridge we use (neva coroutine)
means that it all goes via IDispatch. That, in turn, means we have to
support marshalling of VT_DISPATCH variants.

thanks -mike