inotify patch

[Joakim Tjernlund]

> 
> On Wednesday 21 June 2006 02:46, Joakim Tjernlund wrote:
> > > -----Original Message-----
> > > From: uclibc-bounces at uclibc.org
> > > [mailto:uclibc-bounces at uclibc.org] On Behalf Of Mike Frysinger
> > > Sent: 21 June 2006 01:52
> > > To: uclibc at uclibc.org
> > > Cc: Peter S. Mazinger
> > > Subject: Re: inotify patch
> > >
> > > On Tuesday 20 June 2006 13:49, Peter S. Mazinger wrote:
> > > > - uClibc normally does not provide dummies for the case a
> > >
> > > syscall is not
> > >
> > > >     present
> > >
> > > i was thinking about this ... it's a good idea to always
> > > provide the symbol
> > > even if it isnt supported, otherwise you have an inconsistent
> > > ABI and that's
> > > bad mmmkay
> > >
> > > glibc does this by just setting errno to -ENOSYS and
> > > returning an error
> > >
> > > to save on space, what if we do something like:
> > >  - define two internal functions like
> > > __uclibc_not_implemented_return_{0,1}
> > > the functions would set errno to -ENOSYS and return the
> > > respective value
> > >  - create some macro's in libc-symbols.h to alias to these stubs
> > > #define libc_enosys_stub_ret1(name) \
> > >     weak_alias(name, __uclibc_not_implemented_return_1)
> >
> > This makes the configure step for cross compiling 
> difficult. How should
> > one test for a feature without running a test program?
> 
> glibc faces the same issue ... if you want a stable ABI then 
> you have no 
> choice ...

Why do we need a stable ABI as above? This just makes things harder for
xcompile and I don't see any real advantage.

 Jocke