Tssk.  Be /very/ careful when assuming bugs in gcc: I find it very

satisfactory, and most bugs aren't in gcc but in the person who thinks

they are.  (Well, there is one known bug in the linux-gcc, but that's
No, 12(%esp) is correct.  You are forgetting the return address: there

are three longs on the stack before the push: %eax, %ebx and the return

address.  Thus the offset 12 is correct.  The fact that the resulting

code actually works should have made you cautious.  Not using the frame

pointer makes for code that is a bit harder to follow, but on a 386

where the registers are few anyway, I hate to leave one register

practically unused for no good reason. 
                Linus