lchown/chown semantic in libc/libtricks (was Re: Bug#29454: libtricks does not support lchown+alpha fix)

[Loic Prylli <Loic.Prylli@ens-lyon.fr>]

[with CC: to the libc6 maintainer because the P.S. is relevant to libc]

Mikolaj J. Habryn writes:
 > >>>>> "LP" == Loic Prylli <Loic.Prylli@ens-lyon.fr> writes:
 > 
 >     LP> Would Is is possible at this point to make the chown behaviour
 >     LP> depend on: "#ifdef __alpha__".
 > 
 >   I don't know exactly what the issue is that you are trying to
 > resolve here, but I suspect that the best answer (if possible), is to
 > have configure check for the existence of lchown, and use it if it is
 > there. The current chown/lchown mixup is a transient problem; I humbly 
 > suggest that it would be best to do the ideologically correct thing
 > and use the calls as POSIX intended.

You may have missed the fact that I was adressing libtricks, and not
libc, neither a normal programs that just needs (l)chown for its own
use.


I do not care whether we choose to be POSIX or not. As a matter of
fact, we actually are on Debian/Alpha for libc, that's fine, and I
asked that these semantics be respected when using fakeroot or
libtricks (so packages can be effectively build under fakeroot, as
well as under root, which is something important).

So POSIX or not POSIX is not the issue for libtricks:

*******************************************************************
For LIBTRICKS: the _right_ _thing_ is to do what the libc do (its
POSIX on alpha, be posix, its not posix on intel, do not be posix).
******************************************************************

Regards,

Loic

P.S.: For what semantic to put into libc, you mileage may vary.  You
can argue that for i386 also, it is better to respect POSIX so that
every debian user can be confronted to the great lchown/chown semantic
debate :-), and be given as well a great opportunity to learn the
internal format of dpkg packages, and how to use ar :-), but this was
not the issue in my mail, which was specific to _libtricks_ and not to
libc, nor to any program using (l)chown.

P.S.2: I have yet to be shown one program in the world who would need
the right chown POSIX semantic, it has not been available for about 7
years under Linux, was not missed, so I would say it can wait a few
months for glibc2.1 for i386, but this is just my humble opinion.