> This change is wrong for (at least) the uses of pointer in the > profiling code. The GC_sourceLabel.label field corresponds to a > code label, which (on x86) isn't necessarily 4-byte aligned. > Likewise in GC_handleSigProf, the program counter (on x86) isn't > necessarily 4-byte aligned. Makes sense. The right solution is to use a different type(def) for the two different kinds of pointers, no?