This is off the top of my head, but I'd say that this is reasonable place to put a refcount if *all* instances will need to have it. If you had an optimization in place for "objects with a refcount of anything but one" or some other thing, then you might consider using the syncblock record (which is where the monitor and the hash value hang out, lazily). The syncblock index is found at a negative offset from the pointer to the MethodTable - this accounts for the second DWORD that you are seeing.
You *are* going to have to grind through a lot of asserts, I'd bet. Remember that there will be multiple places in the runtime (such as RuntimeHelpers.cs in System.Runtime.CompilerServices, where OffsetToStringData will need to be changed to return the correct size). The GC also depends upon the size of the object header when it walks the heap. There may also be stubs or places in the JIT's codegen where assumptions about size have leaked in... The right way to look at this particular project: an adventure! (I hope :) -- David Stutz -----Original Message----- From: Chris Tavares [mailto:[EMAIL PROTECTED]] Sent: Thursday, July 18, 2002 2:11 PM To: [EMAIL PROTECTED] Subject: [DOTNET-ROTOR] Hacking ObjHeader, hitting a wall Hi all. I was hoping I could get some guidance from the folks here that have been playing with the Rotor code a bit longer than I have. All this is with the latest Rotor drop on WinXP. As part of an experiement in adding refcounting to Rotor, the first step, of course, is adding the refcount field itself. It seems that the ObjHeader structure is the appropriate place to put these precious four bytes. So I went ahead and added my DWORD, just before the m_SyncBlockValue field. I also updated the SIZEOF_OBJHEADER macro to 8 just above the ObjHeader class definition. (BTW - is there a reason for SIZEOF_OBJHEADER to exist? What's wrong with plain old sizeof()?) I got an assertion error about MIN_OBJECT_SIZE (defined in object.h) not being the right size anymore. I looked at the definition, and it looked like there were four bytes of padding in there anyway, so I took out the "2 * " from MIN_OBJECT_SIZE, and fixed that assertion. Now, I'm hitting an assertion in binder.cpp, line 636. It looks like there's a table of field offsets, and it's comparing the actual offset to the ones in the table. I'm kinda stuck here - I'm not sure why they don't match. So, my questions are: 1) Was the change I make to the MIN_OBJECT_SIZE ok? 2) What's the deal with the offset tables in the Binder class? How do I update those tables so things line up properly again? 3) Is there a better place to put the ref count? Thanks, -Chris Tavares
