I'm resending this patch because my reply to Henri's concern came too late. 
Henri noted that I am enabling lights that do not exist. Existing tests show 
that if no light is assigned to the index, LightEnable creates a light with a 
set of default parameters, so the tests should be fine.

From 9ee4c61805b50886f79e87d744b52f27b7b00b4e Mon Sep 17 00:00:00 2001
From: Stefan Doesinger <stefan@codeweavers.com>
Date: Thu, 29 Nov 2007 13:22:47 +0100
Subject: [PATCH] WineD3D: Enabling too many lights is silently ignored

This patch adds tests for all d3d versions that show that Windows
pretends that enabling more lights than supported succeeds. D3D_OK is
returned, and the light is reported as enabled.

What is not tested in this patch is the rendering output of this
situation, thus the FIXME is still written.

wined3d: Fix handling of a special case in IWineD3DImpl_FillGLCaps() and adjust type of loop variable.

d3dx8: Initialise temp in D3DXQuaternionInverse to avoid a uninitialised variable warning with some versions of gcc.

In one condition (fMustAlloc == TRUE), base_ptr_val may be uninitialised
and so cause a valgrind warning. Reversing the check is harmless and
doesn't result in a performance decrease.

rpcrt4: Initialise the memory in test_simple_struct_marshal before passing it to NdrSimpleStructUnmarshall.

Otherwise, the unmarshalling code will read from unintialised memory
when trying to reuse the embedded pointers in the memory block.

Also fix a couple of memory leaks.

include: Fix the HRESULT_FROM_WIN32 macro to not change the HRESULT if the error is already an HRESULT.

This is so that we don't try to host it.

rpcrt4: Improve PointerUnmarshall to cope with keeping the non-NULL source pointer of a client unmarshall.

This will enable various types with embedded pointers to handle [in,out] 
parameters correctly by keeping the existing memory.

This will enable them to be individually fixed up to use memory if it 
was provided by the caller.

rpcrt4: EmbeddedPointerUnmarshall doesn't need to change the address of the allocated memory, so reduce the level of indirection of the memory parameter by one.

rpcrt4: Errors in sending or receiving packets should result in RPC_S_CALL_FAILED being returned, not RPC_S_PROTOCOL_ERROR.