[PATCH libXi] 1UL may be 4 bytes, force to 1ULL

[Peter Hutterer]

On Sun, Oct 02, 2011 at 10:06:55PM -0700, Jeremy Huddleston wrote:
> 
> On Oct 2, 2011, at 16:30, Peter Hutterer wrote:
> 
> > On Sun, Oct 02, 2011 at 10:52:03AM -0700, Jamey Sharp wrote:
> >> On 10/2/11, Daniel Stone <daniel at fooishbar.org> wrote:
> >>> On 2 October 2011 09:19, Jeremy Huddleston <jeremyhu at apple.com> wrote:
> >>>> Peter, any thoughts on this?
> >>> 
> >>> I think using * (1 << 16) * (1 << 16) is a better idea than (1ULL << 32).
> >> 
> >> And I think ldexp is a much better idea than either. :-) It's more
> >> clear, and as an added bonus, probably faster. (You can multiply a
> >> double by a power of two just by adding the exponents. Division by an
> >> arbitrary constant is stupidly slow by comparison.)
> >> 
> >> To me, the interesting part of Jeremy's question was whether helper
> >> functions are worthwhile for these conversions, 
> > 
> > yes.
> > 
> >> and if so, where they should live.
> > 
> > preferably not in in the protocol. it exposes us to more bugs in the
> > protocol headers (which in theory should be static). And since they're
> > built-in updating them in case a bug is found is harder, you'll need to
> > rebuild anything that uses the headers, simply updating the component isn't
> > enough (that's from a distro POV).
> > 
> > tbh, that's one case where I'd accept duplicate implementations in the
> > server and libX*.
> 
> Yeah, with my port maintainer hat on, a bug in a proto header like that
> would be *incredibly* annoying to push out.  I like the idea of them
> living as static functions in libXi and the server.

I'd vote for putting them into the server with a few tests to make sure they
work mostly as requested. Once that's in, we can add them to the library,
hoping that there aren't any big bugs left.

Cheers,
  Peter