Re: [bug #17900] rm -f should ignore ENOTDIR

[Paul Eggert]

Eric Blake <address@hidden> writes:

> Shouldn't this also include ELOOP?

With ELOOP, we don't know whether the file exists.  It could
merely be a symlink chain that's too long.  So it's safer
for "rm" to print a diagnostic.

This is in contrast with ENOTDIR, where we know for sure
that the file does not exist.

The POSIX spec is a bit weird here; it says to print a
diagnostic if the file "does not exist".  But suppose we
lack permission to search the directory D, and the file D/F
exists.  Should "rm -f D/F" report a diagnostic?  There's no
way for "rm" to tell that D/F exists here, so there's no way
to implement the POSIX spec as written.  In this particular
case GNU rm is cautious and prints a diagnostic; Solaris rm
is lackadaisical and doesn't.

I'd prefer GNU rm to be more cautious in this area, so I
think it should print a diagnostic for ELOOP, as well as for
EACCES.