On Nov 4, 2015, at 2:12 PM, Martin Buchholz <marti...@google.com> wrote:
> At Google, we've also noticed crashes in Bits.c with recent gcc. > Adding nio-dev, since this is not really a build problem. Agreed. The posted patch is not intended to be a final solution. > It appears that Bits.c is resorting to undefined behavior, even though there > is a long tradition of unaligned access on x86 being permitted by the > hardware. > > My colleague Alexander Smundak has also been looking at this, and is trying > to actually fix Bits.c. Any suggestions here would be welcome. > I've also been thinking about this a little. When compiling on x86 with gcc, > we might want to use attribute(aligned(...)) to keep gcc from making > unwarranted assumptions about alignment. > > I think Andrew Haley has also been thinking about efficient array/buffer > operations?