Problem with ITypeInfo

[jack]

Hi.

I've tryed to run game Project Entropia and I got segmentation fault 
because of a bug in OLE.
I've hacked code and I found two bugs:

1. ITypeInfo has back pointer to ITypeLib, but it's possible (and 
happened) that ITypeLib is
destroyed and later ITypeInfo calls its function. I've solve this 
problem by call ITypeLib_AddRef
while creating ITypeInfo (and ITypeLib_Release while destroying ITypeInfo).

2. ITypeInfo_Invoke doesn't support optional parameters. I've added 
support for cParamsOpt>0.
MSDN says that optional parameter is when VARTYPE equals VT_EMPTY, but 
in my case
it was a VT_BYREF | VT_VARIANT type but the pointer was NULL. When I 
treated it as
VT_EMPTY, it was OK.

I enclose my patch. My knownlage about OLE is sparse so I'm not able to 
test this.

Thanks,
Jacek

[Dmitry Timoshkov]

"jack"  wrote:
> @@ -3265,6 +3266,7 @@
>   return NULL;
>        }
>        *ppTypeInfoImpl = (ITypeInfoImpl*)ITypeInfo_Constructor();
> -
>  pTypeLibImpl->ref++;

Are you sure you need this piece of the patch? It causes the following
test program to fail:

#define COBJMACROS

#include <stdarg.h>
#include <stdio.h>
#include <assert.h>

#include "windef.h"
#include "winbase.h"
#include "oleauto.h"

static void load_type_lib(LPCWSTR type_lib)
{
    ITypeLib *iface;
    TLIBATTR *pTLibAttr;
    HRESULT hResult;
    ULONG ref_count;
//    DWORD dwWritten;

printf("Loading: "%ws"\n", type_lib);

hResult = LoadTypeLib(type_lib, &iface);
    printf("LoadTypeLib() = %08lx\n", hResult);
    assert(hResult == S_OK);

hResult = ITypeLib2_GetLibAttr(iface, &pTLibAttr);
    printf("GetLibAttr() = %08lx\n", hResult);
    assert(hResult == S_OK);

//    fflush(stdout);
//    WriteFile(GetStdHandle(STD_OUTPUT_HANDLE), pTLibAttr, sizeof(*pTLibAttr), &dwWritten, NULL);
//    printf("\n\n");

ITypeLib2_ReleaseTLibAttr(iface, pTLibAttr);

ref_count = ITypeLib2_Release(iface);
    printf("ITypeLib2_Release() = %lu\n", ref_count);
    assert(ref_count == 0);
}

int main(void)
{
//    load_type_lib(L"olepro32.dll");
//    load_type_lib(L"msvbvm60.dll");
    load_type_lib(L"stdole32.tlb");
    return 0;
}

[Jacek Caban]

Dmitry Timoshkov wrote:
> "jack"  wrote:
> > @@ -3265,6 +3266,7 @@
> >  return NULL;
> >       }
> >       *ppTypeInfoImpl = (ITypeInfoImpl*)ITypeInfo_Constructor();
> > -
> >  pTypeLibImpl->ref++;
>
> Are you sure you need this piece of the patch? It causes the following
> test program to fail:

No, I don't. In my case I need only this:

/* this is where we are coming from */
+    ITypeLib2_AddRef((ITypeLib*)pLibInfo);
     ptiRet->pTypeLib = pLibInfo;
     ptiRet->index=count;

but I call ITypeLib_Release in ITypeInfo_Release so I thought I should 
increment ref here too.

Thanks,
Jacek

[Dmitry Timoshkov]

"Jacek Caban"  wrote:
> No, I don't. In my case I need only this:
>
> /* this is where we are coming from */
> - ITypeLib2_AddRef((ITypeLib*)pLibInfo);
> ptiRet->pTypeLib = pLibInfo;
> ptiRet->index=count;
>
> but I call ITypeLib_Release in ITypeInfo_Release so I thought I should 
> increment ref here too.

Then use my test program and possibly add some other test for
typelib reference counting until you get it right. Then submit
a tested patch to wine-patches.

[Jacek Caban]

Dmitry Timoshkov wrote:
> "Jacek Caban"  wrote:
> > No, I don't. In my case I need only this:
> >
> > /* this is where we are coming from */
> > - ITypeLib2_AddRef((ITypeLib*)pLibInfo);
> > ptiRet->pTypeLib = pLibInfo;
> > ptiRet->index=count;
> >
> > but I call ITypeLib_Release in ITypeInfo_Release so I thought I should 
> > increment ref here too.
>
> Then use my test program and possibly add some other test for
> typelib reference counting until you get it right. Then submit
> a tested patch to wine-patches.

Thanks for help. I found my bug and I'll send patch soon.

Jacek