Re: rename() over NFS

[Jim Meyering]

Bruno Haible wrote:

> A followup to 2010-11-13
> <http://lists.gnu.org/archive/html/bug-gnulib/2010-11/msg00180.html>:
>
> On a Linux 2.6.16.60 machine, I still get failures from 'test-rename'
> and 'test-renameat' over NFS:
>
>   test-rename.h:279: assertion failed
>   FAIL: test-renameat
>
> The reason is that the mkdir() call at line 279 fails with EEXIST
> because the old directory name is still in the stat() cache.
>
> This change makes the tests succeed. OK to commit?
>
>
> 2010-12-25  Bruno Haible  <address@hidden>
>
>       rename, renameat: Avoid test failures at NFS mounted locations.
>       * tests/test-rename.h (assert_nonexistent): Remove the old directory,
>       so that subsequent mkdir calls succeed.
>
> --- tests/test-rename.h.orig  Sat Dec 25 15:20:33 2010
> +++ tests/test-rename.h       Sat Dec 25 15:20:07 2010
> @@ -56,11 +56,16 @@
>    if (stat (filename, &st) == -1)
>      ASSERT (errno == ENOENT);
>    else
> -    /* But after renaming a directory over an empty directory on an 
> NFS-mounted
> -       file system, on Linux 2.6.18, for a period of 30 seconds the old
> -       directory name is "present" according to stat() but "nonexistent"
> -       according to dentry_exists().  */
> -    ASSERT (!dentry_exists (filename));
> +    {
> +      /* But after renaming a directory over an empty directory on an NFS-
> +         mounted file system, on Linux 2.6.18, for a period of 30 seconds the
> +         old directory name is "present" according to stat() but 
> "nonexistent"
> +         according to dentry_exists().  */
> +      ASSERT (!dentry_exists (filename));
> +      /* Remove the old directory name, so that subsequent mkdir calls
> +         succeed.  */
> +      (void) rmdir (filename);
> +    }
>  }

Nice.  Thanks for pursuing that.