Re: coreutils-6.4: mv(1), rename(2) and 9p

[sqweek]

On 1/6/07, Paul Eggert <address@hidden> wrote:
> sqweek <address@hidden> writes:
>
> > Quoting from copy.c (just above the patch):
>
> Sure, but that quote talks about another problem: cross-device
> renames that POSIX says should fail with errno == EXDEV but
> instead fail with errno having some other value.

Yeah alright, I'll pay that. I'm sure I came across some comment that
directly addressed filesystems where rename is not implemented, I'll
see if I can track it down...
Ah, it's in do_move() from mv.c, and I guess it's just giving some
specifics on the semantics expected of copy().

> is obviously too broad, since it will cause 'mv' to mishandle ordinary
> cases of EPERM.

EPERM did strike me as a bit of a weird return code, even after I
found "; or the filesystem containing pathname does not support
renaming of the type requested." tacked on to the end of the "EPERM or
EACCES" section in rename(2). Have to wonder where that came from...

> This problem will affect lots of applications, so to some extent
> you're swimming upstream trying to patch coreutils to work around the
> problem.

Figured as much after your first reply, so I'm looking into a fix on
the 9p side.
Thanks for the prompt response.