Re: mv symlink-to-dir/, debian bug feed

[Jim Meyering]

Eric Blake <address@hidden> wrote:
> I've noticed that Jim just committed a patch based on a bug originally
> reported through the debian tracker without any additional mention here:
>
> http://git.sv.gnu.org/gitweb/?p=coreutils.git;a=commit;h=c0c8685
> http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=343652

Thanks.

> But regarding the patch, I'm wondering if an Austin Group interpretation
> is needed here.  The next draft for POSIX has already tightened the

IMHO, it is needed.
If you're willing to broach the subject, that'd be great.

> wording to make it clear that rename("dir", "newdir/") must fail with
> ENOTDIR if newdir is not already a directory.  Likewise, it clarifies that
> 'mv dir newdir/' must fail (oops - coreutils 6.10 doesn't do that).
> However, I still don't see any clarification on whether the Linux behavior
> of rename("symlink-to-dir/", "newname") failing with ENOTDIR is valid.

As you probably realize, that is behavior straight from glibc
and the kernel rename syscall.

> Meanwhile, would it be worth subscribing bug-coreutils to the debian bug
> feed list?  That way, this list would see bugs as they are reported, and
> others besides Jim will be able to chime in with advice.

I know Bob Proulx is already subscribed there.
Some of the traffic would not be interesting, i.e., a message
announcing that a bug is closed, or tagged -- but those are easy to skip.

Overall, I think it would be better for both Debian and GNU.
We can always try, and if it doesn't work out, remove it later.
Bob, Michael, what do you think?