On Fri, Nov 01, 2019 at 10:56:17AM -0400, Rich Felker wrote:
> The time64 switchover I had been testing and hoped to push yesterday
> turned out to be missing the ioctl and sockopt bits, which I
> eventually remembered I'd omitted in the late-summer i386-only draft
> because making a bunch more archs have arch variants for these did not
> seem fun. So, I've worked out some changes to allow top-level socket.h
> to do a lot more. The msghdr move is pretty much orthogonal, aside
> from interaction with x32.
> 
> It's probably not perfect at this point, but I think it's an
> improvement, and short of errors good enough to commit. It brings
> things closer to a state of only needing arch-specific bits for archs
> that actually do weird things, rather than repeating things that are
> universal to all 32-bit archs or all 64-bit archs on a per-arch basis.
> 
> These patches should apply ok on the last time64 switchover draft
> posted to the list. But I'm posting them more for the sake of a (very
> brief) comment period about what they're doing, and to have a record
> of them as individual patches -- some of them will be squashed since
> they yield ABI-inconsistent revisions if left split.

I left out the first patch of this series, attached. Unfortunately it
had a bug: without changes to the bits headers, it caused
SO_TIMESTAMP* definitions to disappear from archs that define their
own SO_RCVTIMEO (powerpc* and mips*).

It will be fixed by separating these out into 2 groups with separate
#ifdefs.

Rich