Greg Stark <stark@mit.edu> writes:
> Sorry -- with the obvious error fixed:
You didn't show -E output from this version, but the other one had
> __attribute__((packed))
> __attribute__((aligned(2)))
so it appears that clang 4.0 does accept these attributes but then
produces the warning anyway. I suggest filing this as a bug in clang 4.0,
and marking it as a regression from older versions which did not produce
such a warning.
If you get pushback claiming it's intentional, I'd be inclined to hack
our macro definitions so that we don't believe clang understands
attribute(aligned), because it evidently doesn't. But let's see
their response first.
regards, tom lane