Re: [Fwd: [bug #6047] rm -f "" prints diagnostic message under the Hurd]

[Ognyan Kulev]

Here "POSIX" means "The Single UNIX Specification Version 3".

Marcus Brinkmann wrote:
> At Thu, 29 Apr 2004 12:32:35 +0300,
> Ognyan Kulev wrote:
> > dir_link("x", "") returns EINVAL

link(path1, path2) contains the following text:

[ENOENT]
    A component of either path prefix does not exist; the file named by 
path1 does not exist; or path1 or path2 points to an empty string.

So dir_link("x","") should return ENOENT.  (Here is implied that Hurd 
RPCs should follow POSIX semantics.  Another solution is to change glibc 
and not change diskfs.)

> > dir_unlink("") returns ENOENT

unlink(path):

[ENOENT]
    A component of path does not name an existing file or path is an 
empty string.

> > dir_mkdir("") returns EINVAL

mkdir(path):

[ENOENT]
    A component of the path prefix specified by path does not name an 
existing directory or path is an empty string.

> > dir_rmdir("") returns ENOENT

[ENOENT]
    A component of path does not name an existing file, or the path 
argument names a nonexistent directory or points to an empty string.

> > dir_rename("x", "") returns EINVAL
> > dir_rename("", "x") returns ENOENT

rename(old,new):

[ENOENT]
    [CX] The link named by old does not name an existing file, or 
either old or new points to an empty string.


open(path,oflag):

[ENOENT]
    O_CREAT is not set and the named file does not exist; or O_CREAT is 
set and either the path prefix does not exist or the path argument 
points to an empty string.

I think that touch "" returns EINVAL under the Hurd, but I'm lazy to 
check it right now.

> > mv x "" should give EINVAL, I think.  I suppose this is for the 
> > coreutils developers.
>
> You must be careful here.  The main thing that matters for us is the
> POSIX interface here.  The coreutils already behave correctly, unless
> there is a bug that also shows on other POSIX systems.  The dir_* are
> not specified, but we can specify them.  So, we can decide if we want
> to use EINVAL or ENOENT in them.  I think that there is a consistent
> pattern to make them behave like the equivalent POSIX functions, but I
> am not sure it is completely followed.
>
> In any case, how rename(), unlink() etc should behave is defined by
> POSIX.  I only checked unlink(), mkdir() and rename() (not rmdir and
> link), and for those ENOENT is required.  You may want to check the
> others.

From the excerpts above, it looks like POSIX system interfaces always 
return ENOENT when some pathname is empty string.  Here is the 
description of ENOENT in error numbers:

[ENOENT]
    No such file or directory. A component of a specified pathname does 
not exist, or the pathname is an empty string.

Again, the easiest is to make diskfs_lookup return ENOENT when empty 
string is given.  Or we can change dir_* functions, which is not hard. 
glibc can be changed too, but I suppose we want Hurd RPCs to have the 
POSIX behaviour.

Regards,
ogi