OLEAUT32: bug 4502 - request for comment before trying to fix

[Alex Villací­s Lasso]

Alex Villací­s Lasso wrote:
> I would like to draw attention to bug 4502. This bug can cause any VB 
> application to crash when the DatePicker control is used, and an 
> attempt is made to assign a date value in VB code. Even though I found 
> this bug while testing an application I wrote, this might not be 
> specific to this application. It is only recently that I have some 
> time to dedicate to this bug, but I would like somebody to comment on 
> this.
>
> When the sample application is run with WINEDEBUG=ole,variant, the 
> failing invocation looks like this:
>
> trace:ole:ITypeInfo_fnInvoke 
> (0x7fdd5440)(0x7fd9f61c,id=20,flags=0x00000004,0x7fa3ee8c,(nil),0x7fa3ee6c,0x7fa3ee9c) 
>
> trace:ole:dump_DispParms args=1 named args=1
> trace:ole:dump_Variant 0x7fa3eef4->{VT_DATE(38788,0x7fa3ec7c)
> trace:variant:VarUdateFromDate (38788,0x00000000,0x7fa3ec36)
> trace:ole:dump_Variant ,2006/03/12 00:00:00}
> trace:ole:ITypeInfo_fnInvoke invoking:
> L"Value"(1)
>    parm0: L"Value"
> memid is 00000014
> Param 0:
>        tdesc.vartype 12 (VT_VARIANT)
>        u.paramdesc.wParamFlags PARAMFLAG_FIN
>        u.paramdesc.lpex (nil)
>    funckind: 1 (pure virtual)
>    invkind: 4 (property put)
>    callconv: 4 (stdcall)
>    oVft: 224
>    cParamsOpt: 0
>    wFlags: 3c
>    elemdescFunc (return value type):
>        tdesc.vartype 25 (VT_HRESULT)
>        u.paramdesc.wParamFlags PARAMFLAGS_NONE
>        u.paramdesc.lpex (nil)
>    helpstring: L"Returns/sets the current date."
>    entry: (null)
> trace:ole:ITypeInfo_fnInvoke changing args
> trace:ole:dump_Variant 0x7fa3eef4->{VT_DATE(38788,0x7fa3ec7c)
> trace:variant:VarUdateFromDate (38788,0x00000000,0x7fa3ec36)
> trace:ole:dump_Variant ,2006/03/12 00:00:00}
> trace:variant:VariantChangeTypeEx 
> (0x7fdd5520->(VT_DATE),0x7fa3eef4->(VT_DATE),0x00000400,0x0000,VT_VARIANT) 
>
> trace:variant:VariantClear (0x7fa3ec40->(VT_EMPTY))
> trace:variant:VariantClear (0x7fa3ec30->(VT_EMPTY))
> trace:variant:VariantCopyInd 
> (0x7fa3ec30->(VT_EMPTY),0x7fa3eef4->(VT_DATE))
> trace:variant:VariantCopy (0x7fa3ec30->(VT_EMPTY),0x7fa3eef4->(VT_DATE))
> trace:variant:VariantClear (0x7fa3ec30->(VT_EMPTY))
> trace:variant:VARIANT_Coerce 
> (0x7fa3ec40->(VT_EMPTY),0x00000400,0x0000,0x7fa3ec30->(VT_DATE),VT_VARIANT) 
>
> trace:variant:VariantClear (0x7fa3ec40->(VT_EMPTY))
> trace:variant:VariantClear (0x7fa3ec30->(VT_DATE))
> trace:variant:VariantChangeTypeEx returning 0x80020005, 
> 0x7fdd5520->(VT_DATE)
> err:ole:ITypeInfo_fnInvoke failed to convert param 0 to VT_VARIANT 
> from VT_DATE
> trace:ole:ITypeInfo_fnInvoke -- 0x80020005
>
> The Property Put seems to be declared as taking one VT_VARIANT 
> parameter, and the VB code supplies a VT_DATE value. Therefore, an 
> attempt is made to coerce VT_DATE --> VT_VARIANT. However, 
> VariantChangeTypeEx() does not support conversion of *any* kind of 
> variant into VT_VARIANT, so the call fails, and this brings down the 
> application. I was on half a mind to add the VT_VARIANT conversion to 
> VariantChangeTypeEx() myself, but then I checked the tests for this 
> operation, and it seems that the absence of support for VT_VARIANT is 
> actually expected (unless I am misreading the code). Therefore, one of 
> the following two things is happening:
>
> 1) There is some error in the reading of the typelib, since a Property 
> Put should not have a VT_VARIANT parameter in the first place.
> 2) ITypeInfo::Invoke should do the VT_VARIANT conversion itself, 
> precisely because VariantChangeTypeEx() won't do it.
>
> I am inclined to the second option, but I want to hear some comment 
> before attempting to create a patch.
>
> Alex Villacís Lasso
>
After some digging, I found that bugs 4370 and 4573 seem to expose the 
same problem. The root cause (in bug 4370) seems to be a patch by Robert 
Shearman that rewrites ITypeInfo::Invoke to use VariantChangeTypeEx 
(commit at 
http://source.winehq.org/git/?p=wine.git;a=commit;h=8cf5373f60aca5345a0c55df2711a9f0b73b8133). 
Saved me the trouble of the regression testing.

Alex Villacís Lasso