Re: [Pnet-developers] A question about verifier code

[Rhys Weatherley]

On Sunday 25 May 2003 06:42 pm, Gopal V wrote:
> Hi,
>
> case IL_OP_LDELEMA:
> {
>       /* Load the address of an array element */
>       if(STK_BINARY_1 == ILEngineType_O &&
>          (STK_BINARY_2 == ILEngineType_I4 || STK_BINARY_2 == ILEngineType_I) 
> &&
>          (elemType = ArrayElementType(stack[stackSize - 2].typeInfo)) != 0)
>       {
>               classType = GetTypeToken(method, pc);
>               if(classType &&
>                  (elemType == ILType_Void || ILTypeIdentical(elemType, 
> classType)))
>               {
>                       ILCoderArrayAccess(coder, opcode, STK_BINARY_2, 
> classType);
>
> So why are we outputting the type of the elemType ?.. Shouldn't we be like
> outputting &elemType ? ... That seems to be what rectangular arrays are
> spitting out ..

The next two lines take care of creating "&elemType" in the stack:

            stack[stackSize - 2].engineType = ILEngineType_M;
            stack[stackSize - 2].typeInfo = classType;

A type "M" stack item is a managed pointer to "classType".  The 
ILCoderArrayAccess function is passed the LDELEMA opcode, which tells it to 
generate an addressof operation rather than a value fetch.

> Which means that I ran into quite a few problems handling both cases..
> Any quick advice ?.

I'm not really sure that I understand what the problem is.  If rank == 1, use 
LDELEMA, otherwise use "::Address".  The code looks fine to me.  Do you have 
a specific test case that demonstrates whatever problem it is that you're 
seeing?

Cheers,

Rhys.