On Tue, Feb 07, 2017 at 02:56:33PM -0300, Lisandro Damián Nicanor Pérez Meyer wrote: > By the description you give here I *think* you are just taking into account > 32/64 bits, but that's not necessarily enough: you might have to do > substitutions on 32 bits archs. Example: qt4's qreal on arm is a float, on > i386 is a double.
That's entirely correct. This approach does not make updating substitutions go away. It only makes those cases go away that really only depend on the number of bits. Your qreal example doesn't depend on the bits, so it doesn't go away. This is not a full solution, but it removes a significant chunk of updating. That may be worth it already. > Using compile-time definitions might not always work except you always build > depend upon whatever defines a type. Example: qt4's qreal, once again. Again, that's entirely correct. Determining the size of qreal requires depending on qt of course. Having to depend on qt will result in bootstrap cycles, so let's not go down that road. My suggestion was simply short sighted. > Am I missing something? No. Thank you for the attention to detail. Given the above, I think that the approach from my patch looks most promising. Do you agree? Helmut
