Re: [GIT PULL -tip] fix 22 make headers_check - 200901

["H. Peter Anvin" <hpa@zytor.com>]

Arnd Bergmann wrote:
> On Wednesday 04 February 2009, H. Peter Anvin wrote:
>> Actually, if anything we should move the *non* __KERNEL_STRICT_NAMES out
>> of <linux/types.h> into something else, or completely deep-six them.  I
>> don't know of any libc which wants these anymore, and I think they're
>> just residual libc5 cruft.
>>
>> However, if we want <linux/extra_types.h> that's fine with me; but
>> <linux/types.h> really should be clean, which means doing what
>> __KERNEL_STRICT_NAMES does now.
> 
> Right now, we have 15 exported headers [1] that use the non-strict 
> posix types (pid_t, off_t, clock_t, ...) and a set of 106 (!)
> files [2] using non-strict integer types (u_int32_t, uint32_t, u32, ...),
> 76 of those alone in netfilter.
> 

Geez.  The integer types is just a pattern replacement, so those we can
just fix.  The 15 exported headers that use other types may very well be
real bugs -- we have had a fair share of broken ioctl signatures due to
exactly this problem.

> Do you think we should fix up all of them before 2.6.29? I'm worried
> that we might introduce more regressions in the process.
> Also, should we leave netfilter alone, in order to reduce the changes?
> I'm also unsure whether a hack in headers_install would be better than
> changing the headers in the source tree.

I have been advocating for hacking headers_install for a while.  That
takes care of the 106.  The 15 *need* to be audited immediately, because
that is even likely to be actual manifest bugs.

	-hpa
-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.