data type bug or not....

I've noticed that the following definitions generate different data
results. Perhaps, I'm overlooking something.

#define TEMP (*(unsigned char xdata *) 0x10F0))

unsigned char xdata sutton2, *sutton;
unsigned char xdata * xdata sutton3;

sutton = &TEMP;
sutton3 = &TEMP;
sutton2 = TEMP;

If XDATA location 0x10F0 has 0x35, then the following compilation
occurs:

'sutton' has the value X:10F0, which is correct.
'sutton2' has the value 0x35, which is correct.
'sutton3' has the value D:10F0, which is not correct.

To complicate things, if I swap the 'sutton2' and 'sutton' declarations to:

unsigned char xdata *sutton, sutton2;

then 'sutton' has the value D:10F0, which is not correct.

There is no DATA location 10F0, therefore garbage is the result.

The source code generated for *sutton being first in the declaration
list is a 2 byte XDATA pointer, which points to the DATA area.

If *sutton is not the first declaration in the list, it is a 3 byte pointer
(which works correctly) that points to XDATA.

Anyone know what the problem is and why the sutton3 declaration
listed above does not force the compiler to generate a pointer that
points to XDATA?

Thanks

Sutton Mehaffey
Lookout Portable Security
sutton@lookoutportablesecurity.com