Re: [PATCH 1/2] Add C99-style constructor macros for specific-sized integers

[Al Viro <viro@ZenIV.linux.org.uk>]

On Sun, Mar 02, 2008 at 10:12:37PM -0800, H. Peter Anvin wrote:
> Al Viro wrote:
> >On Sun, Mar 02, 2008 at 08:43:22PM -0800, H. Peter Anvin wrote:
> >>>>+#define S64_C(x) x ## LL
> >>>>+#define U64_C(x) x ## ULL
> >>>but can't we use the latter for all arches?
> >>>
> >>You really don't want to, as the whole point is that you want it to 
> >>match what u64/s64 is defined as, in order to avoid spurious warnings.
> >
> >Excuse me, in which contexts?  Where does replacement of explicitly long
> >constant with explicitly long long one generate warnings on a host that has
> >range(long) equal to range(long long)?
> 
> printf, for example.

Oh, for...  It's not as if we _had_ anything better for u64 than %ull and
convert the argument to unsigned long long for that family.

Unless you are seriously suggesting the use of vomit-inducing atrocity of
PRIu64 and constant concatenation in there, that is.  I would rather move
typechecking for printk-style functions to sparse and tell gcc to STFU
completely on that class of warnings.  Making it extensible, while we are
at it - i.e. telling sparse that this conversion is to be used for dma_addr_t,
etc.