Re: [PATCH] linux/inotify.h: do not include <linux/fcntl.h> in userspace

[Adrian Bunk <bunk@kernel.org>]

On Wed, Sep 17, 2008 at 01:04:23PM +0300, Kirill A. Shutemov wrote:
> On Wed, Sep 17, 2008 at 12:32:40PM +0300, Kirill A. Shutemov wrote:
> > On Tue, Sep 16, 2008 at 07:09:02PM +0300, Adrian Bunk wrote:
> > > On Tue, Sep 16, 2008 at 07:10:25AM -0700, Ulrich Drepper wrote:
> > > > Kirill A. Shutemov wrote:
> > > > >> What is the error message?
> > > > > 
> > > > > /usr/include/asm-generic/fcntl.h:117: error: redefinition of 'struct
> > > > > flock'
> > > > 
> > > > And?  None of these programs should use <linux/inotify.h>.  There has
> > > > for the longest time been a <sys/inotify.h> header which doesn't need
> > > > any kernel headers.  In fact, <linux/inotify.h> should not be exported.
> > > 
> > > Even if userspace applications shouldn't use it directly this doesn't 
> > > sound right:
> > > 
> > > We shouldn't force all libc implementations to copy the contents into a 
> > > private header.
> > 
> > glibc and dietlibc provide <sys/inotify.h>. newlib and uclibc don't. klibc
> > provides <sys/inotify.h> but it depends on <linux/inotify.h>
> 
> Oh.. Sorry. uclibc provides <sys/inotify.h>.
> 
> So unexporting <linux/inotify.h> breaks only klibc.

klibc is actually doing the right thing.

The userspace kernel headers situation used to be a complete mess, and 
it is therefore understandable that some libc implementations are 
currently not using <linux/inotify.h>, but ideally in the long term all 
libc implementations should use <linux/inotify.h>.

Whether, and if yes when, libc implementations starts using 
<linux/inotify.h> is not our business, but we have to ensure that it 
works for the libc implementations that do use it.

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed