Re: [git pull] core kernel updates for v2.6.28

[Linus Torvalds <torvalds@linux-foundation.org>]

On Fri, 17 Oct 2008, Frédéric Weisbecker wrote:
>
> 2008/10/17 Linus Torvalds <torvalds@linux-foundation.org>:
> > The fact is, that second argument was a "ptrdiff_t", which is neither
> > "int" nor "long".  It should be "%td" I think. But the thing is, when you
> > fix a warning, you should damn well know what the hell you're doing, not
> > just shut it up.
> 
> Sorry, I made some falses assumptions about the printed type I guess...

Well, the thing is, on 32-bit x86, ptrdiff_t is "int". And on 64-bit, it's 
"long". And on some (most?) other architectures, it's "long" regardless of 
whether it's 32-bit or 64-bit.

So you fixed a warnign on x86-32, but you introduced it just about 
everywhere else.

And it so happens that the old use of "%ld" was better than "%d", because 
regardless of the exact type of ptrdiff_t, with gcc it is essentially 
always going to be at least the same _size_ as "long". IOW, even when it's 
"int", it will always print out correctly with "%ld", despite the format 
warning. IOW, the type may be "wrong" from a C standards standpoint, but 
it will work in practice.

In contrast, using "%d" can actually print it out wrong, because it will 
be literally the wrong physical size, not just a type issue on a C level. 
So depending on calling conventions, you might end up with the upper bits 
cleared, or even the wrong bits printed out.

Using "%td" is always right, assuming the underlying printing library is 
recent enough to know about it. And the kernel has known about %td for the 
last three years.

			Linus