Le 30/08/2010 18:47, Detlev Zundel a écrit :
Hi Reinhard,
Detlev Zundel schrieb:
Detlev, regarding the discussion I would only point out that we have to be sure that such kind of patch will be merged in the current release. It would be a real pity if a new official realease is published and then even a simple "md" command does not work on ARM.
I don't see a problem here. All proposed patches (with/without attribute and union) surely fix a bug, so they will go into mainline when consent is reached on which one to use. This should well happen before the pending release on September 12th[1].
Am I misunderstanding anything here?
No... but I would require that the "union" approach would be wanted, BEFORE I put effort into doing it.
I'd very much appreciate your effort as I want the solution now that you did whet my appetite.
Besides, re: 'fixing with the side-effect of a different thing': I think the alignment caused by using an union is not actually a side effect of it but an intended effect of it, as the compiler must ensure correct alignment of each union member -- on architectures where alignment of 32-bit ints is unnecessary, the union will not cause undue alignment, whereas the __aligned__ attribute would.
Amicalement,