Marshalling of remotable handles

[Michael Karcher]

Hello Wine developers,

while trying to add proxy/stub code for shell32, I stumbled across the
definition of remotable handles in IDL.

We have

- a RemotableHandle union.
 - the IDL typedefs for the wire representation of handles:
|    typedef [unique] RemotableHandle *wireHXXX;
 - the IDL declaration for the handle types itself:
|    typedef [wire_marshal(wireHXXX)] void*HXXX;

This makes widl generate null pointer checks for HWND parameters, as
HXXX is a void pointer without the unique attribute. My question is:
Should widl really add this check? As the type is marshalled as
wireHXXX, the null value is no problem for marshalling, so it could be
safely left out.

What I did in my repository was adding the [unique] attribute to the IDL
typedef of the handle type:
| typedef [unique, wire_marshal(wireHXXX)] void*HXXX;
Is this the right approach to generally get rid of the NULL pointer
checks for handles, or is widl to be fixed instead?

Best regards and thanks for advice,
  Michael Karcher

[Rob Shearman]

2008/12/21 Michael Karcher :
> Hello Wine developers,
>
> while trying to add proxy/stub code for shell32, I stumbled across the
> definition of remotable handles in IDL.
>
> We have
> - a RemotableHandle union.
> - the IDL typedefs for the wire representation of handles:
>
> |    typedef [unique] RemotableHandle *wireHXXX;
> - the IDL declaration for the handle types itself:
>
> |    typedef [wire_marshal(wireHXXX)] void*HXXX;
>
> This makes widl generate null pointer checks for HWND parameters, as
> HXXX is a void pointer without the unique attribute. My question is:
> Should widl really add this check? As the type is marshalled as
> wireHXXX, the null value is no problem for marshalling, so it could be
> safely left out.

Yes, that's a widl bug - it shouldn't output NULL ref pointer checks
for user-marshaled types.
> What I did in my repository was adding the [unique] attribute to the IDL
> typedef of the handle type:
> | typedef [unique, wire_marshal(wireHXXX)] void*HXXX;
> Is this the right approach to generally get rid of the NULL pointer
> checks for handles, or is widl to be fixed instead?

widl should be fixed, since MIDL correctly doesn't output NULL ref
pointer checks when using the same source IDL files.

[Michael Karcher]

Am Freitag, den 26.12.2008, 17:20 +0000 schrieb Rob Shearman:
> 2008/12/21 Michael Karcher :
> > This makes widl generate null pointer checks for HWND parameters, as
> > HXXX is a void pointer without the unique attribute. My question is:
> > Should widl really add this check? As the type is marshalled as
> > wireHXXX, the null value is no problem for marshalling, so it could be
> > safely left out.
>
> Yes, that's a widl bug - it shouldn't output NULL ref pointer checks
> for user-marshaled types.

OK, thanks for your info. I will try to send a patch.

Best regards,
  Michael Karcher