Re: [Bug-tar] [PATCH] two patches for --atime-preserve races and other problems

[Eric Blake]

On 09/16/2010 04:49 PM, Paul Eggert wrote:
> [Adding bug-gnulib.]
>
> On 09/16/10 11:17, Eric Blake wrote:
>
> > Why not just use the gnulib module fdutimensat instead of reimplementing
> > this yourself as fd_utimensat?
>
> Mainly because fdutimensat does not let 'tar' pass AT_SYMLINK_NOFOLLOW to the
> underlying system call.  Sometimes that flag is needed, sometimes not,
> but fdutimensat hardwires a '0'.

Makes sense, and easy enough to fix.  It's a gnulib API, so with proper 
notice in NEWS, we can add a parameter and adjust all callers.

> If fdutimensat were modified to have an additional "atflag" argument, that
> would overcome the main objection.
>
> By the way, in a future version of tar I will also have a need for similar
> functions fdchmodat, fdchownat, fdfstatat, but I don't know whether they're
> worth gnulibifying.  (It's not entirely clear that fdutimensat is worth
> gnulibifying either, to be honest.  :-)

Well, now that both coreutils and tar want to use it (modulo the changes 
to coreutils to accept the new parameter), it makes sense to me to keep 
it in gnulib.  The other functions, like fdchmodat, may also make sense 
to move to gnulib once coreutils' copy.c file is rewritten to use fts 
and the *at functions rather than hand-recursing, but we can cross that 
bridge when we get there.

--
Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org