Re: Test failure on FreeBSD (permissible errno values)

[Ed Maste]

On 12 September 2012 10:44, Eric Blake <address@hidden> wrote:
> On 09/12/2012 08:38 AM, Paul Eggert wrote:
>> On 09/12/2012 06:38 AM, Ed Maste wrote:
>>> Should ENOTDIR be a permissible errno for this test?
>>
>> I don't see why.  It's not called out in the GNU documentation
>> for canonicalize_file_name, and the POSIX documentation for
>> realpath does not seem to allow ENOTDIR for the similar
>> situation with realpath.  Why is the code returning ENOTDIR
>> here?
>
> The POSIX wording is:
>
> [ENOENT] A component of file_name does not name an existing file or
> file_name points to
>               an empty string.
> [ENOTDIR] A component of the path prefix is not a directory, or the
> file_name argument
>          contains at least one non-<slash> character and ends with one
> or more trailing
>         <slash> characters and the last pathname component names an
> existing file
>        that is neither a directory nor a symbolic link to a directory.
>
> Since '/path/to/zzz' does not exist, this should be ENOENT; ENOTDIR
> would only be valid in the case where /path/to/zzz exists but is not a
> directory.  It sounds like a bug in the FreeBSD kernel.
>
> That said, are we already replacing realpath() on FreeBSD?  If so, then
> we should fix our replacement; if not, then I'm probably okay with
> relaxing the test to allow the alternate error code.

After a little investigation, I see that the errno comes from
FreeBSD's realpath.  The behaviour was introduced by the fix for
http://www.freebsd.org/cgi/query-pr.cgi?pr=128933 ("realpath(3) does
not follow SUS specification for ENOENT / ENOTDIR conditions"), which
has a followup with the interpretation that POSIX demands ENOTDIR and
provides the following example:

    # realpath /bin/non_existent/cp
    realpath: /bin/non_existent/cp: No such file or directory
    (should be: "Not a directory", as early as non_existent checked)

I don't necessarily agree with this interpretation, but I do think
there's ambiguity in the POSIX wording.  The gnulib test is equivalent
to realpath("/path/to/zzz/.."), where it's arguable that a component
of the path prefix (specifically, the zzz) is not a directory.

-Ed