Re: [PATCH] ppc64 utmp changes

[Jakub Jelinek <jakub at redhat dot com>]

On Wed, Oct 02, 2002 at 09:58:40AM -0500, Steve Munroe wrote:
> 
> Jakub Jelinek writes:
> 
> > Attached are two version of the updated patch.
> > The first one is how it should look like at the end (well,
> > with Roland's Implies change bits/wordsize.h headers could stay where
> > they are and sparc wordsize.h headers could move up one level),
> > the second one is alternative patch which doesn't change anything for x86
> > and other unaffected architectures (with the exception of utmpdump.c
> change).
> > Note that struct lastlog hasn't been changed in the original patch but
> > it has the same problem as struct utmp and struct utmpx.
> > I haven't changed s390x, but IMHO it is better to recompile the affected
> > apps now in the /usr/lib -> /usr/lib64 transition which changes most
> > of the things anyway, than to suffer forever.
> 
> Jakub I don't understand why you split powerpc/bits/wordsize.h into
> powerpc32/bits and powerpc64/bits. Especially since you left the
> "#if defined __powerpc64__" conditional in both versions. Either leave
> wordsize.h where it was and use the "#if defined" or move them and get
> rid of the "#if defines". I prefer wordsize.h where it was :)

ATM it doesn't work though.
The problem is that say powerpc/powerpc32/Implies contains wordsize-32
and thus the search order is powerpc/powerpc32 wordsize-32 powerpc
and bits/wordsize.h gets picked up from wordsize-32. Works fine when you
build glibc, but you cannot share /usr/include between 32-bit and 64-bit
apps.
Roland mailed a few days ago a patch which made it possible to specify
something like wordsize-32+ in the Implies file which would put it after all
arch sysdeps, not immediately after current one. With that patch this won't
be necessary.

	Jakub