Hi Andras, Petter, >>If sizeof(size_t) is bigger then sizeof(u_int32_t), than the call to >>sanei_scsi_cmd() will write outside the space allocated for >>ms->n_control_bytes. If it is smaller, the value in >>ms->n_control_bytes will be wrong. It only work as it should if >>size_t and u_int32_t is the same type. This is not the case on >>64-bit machines. > > > This is definitely a bug in that backend, but your solution doesn't > solve anything. The only thing it might do is create a compiler > warning which is not much use to the end user anyway. (In fact, it > will certainly cause the compiler to warn about an unused variable.) > > This needs fixing, I agree, and it should be done before 1.0.9. The > solution should be rather easy, I guess the types involved should > either be all size_t or all u_int32_t, but I can't tell (I haven't got > the time to look at the code right now).
Using size_t for ms->n_control_bytes is the way to go. It's ok for ia32 since sizeof(size_t) == sizeof(int). It needs some testing of wider platforms, though, but that's a good start. > BTW, is there any specified lower limit for sizeof(size_t), i.e. can > we safely assume that size_t can always take 32-bit values? Probably That's a safe assumption, at least for the platforms SANE runs on. > not a problem in this case, but it could have interesting implications > now that open-source software on embedded devices, PDAs etc. is taking > off. Frank.
