>IIRC it was something to do with data aborts on a 4K page boundary. Phil, >didn't you work out some scheme so that the compiler could try and make sure >this didn't happen? Of course that means fixing the compiler and re-compiling >everything with this option, which isn't very practical, but would provide a >solution. Or did you decide that in fact it wasn't going to work?
Well, yes, a number of people have proposed methods of avoiding the problem. But all of these would involve applying gruesome changes to the compiler and/ or linker, and recompiling all the affected binaries, plus they would result in worse code for the majority of users who weren't affected by the original bug. Because of that I took the view that it wasn't worth trying to work around this particular bug; environmental considerations aside, we'd be better off just taking all the affected machines out and burning them. p.
