Mike Hearn wrote:
On Mon, 2005-06-06 at 10:26 -0500, Robert Shearman wrote:
switch (type) {
- case RPC_FC_RP: /* ref pointer (always non-null) */
-#if 0 /* this causes problems for InstallShield so is disabled - we
need more tests */
- if (!Pointer)
- RpcRaiseException(RPC_X_NULL_REF_POINTER);
-#endif
+ case RPC_FC_RP: /* ref pointer (always non-null but may point to
null) */
break;
case RPC_FC_UP: /* unique pointer */
case RPC_FC_OP: /* object pointer - same as unique here */
This looks wrong. A ref pointer shouldn't be treated as a unique
pointer in any circumstances AFAIK. I'll add this case to my mini test
suite to confirm or deny this hypothesis.
This is wire-sizing, the full code is:
switch (type) {
case RPC_FC_RP:
case RPC_FC_OP:
case RPC_FC_UP:
pStubMsg->BufferLength += 4;
/* NULL pointer has no further representation */
if (!Pointer)
return;
break;
case RPC_FC_FP:
default:
FIXME("unhandled ptr type=%02x\n", type);
RpcRaiseException(RPC_X_BAD_STUB_DATA);
}
m = NdrBufferSizer[*desc & NDR_TABLE_MASK];
if (m) m(pStubMsg, Pointer, desc);
else FIXME("no buffersizer for data type=%02x\n", *desc);
In this case, we need to reserve space for a refptr on the wire to be
able to tell the difference between NULL and non-NULL. So it reserves 4
bytes in the buffer.
Exactly. A refptr shouldn't have those extra 4 bytes because it should
never be NULL. I would be very surprised if Microsoft have chosen to be
inconsistent here.
--
Rob Shearman