TreeTransform assumes whole tree is on one filesystem; gives "invalid cross-device link"

I have a feeling this is a dupe but I can't find it.

As you identified the problem is that bzr assumes it can directly move files from the limbo directory in .bzr into the tree.

As a cheap but slow workaround we could just change the rename to a copy. It would probably be a cheap fix if you'd like to try.

Doing so would somewhat defeat the purpose of writing things into limbo first, and perhaps it would be better to identify a directory next to the source files where they should be written. Doing that would need to take into account that in extreme cases people's actual working tree may span filesystems.

>As a cheap but slow workaround we could just change the rename to a copy. It would probably be a cheap fix if you'd like to try.

Yes. I'd like to have some patch. Is it possible?
I'm now in the middle of update operation which gives just the same error.

About the rest - yes, you caught the idea right. Moving .bzr to /usr partition solves the problem in my example.

2009/8/10 Denis Golovan <email address hidden>:
>>As a cheap but slow workaround we could just change the rename to a
> copy. It would probably be a cheap fix if you'd like to try.
>
> Yes. I'd like to have some patch. Is it possible?

you should probably look at adding (or calling) code in osutils to
move the file and do a copy if necessary.
--
Martin <http://launchpad.net/~mbp/>

2009/9/24 Denis Golovan <email address hidden>:
>
> ** Attachment added: "Patch for 1.17/1.18"
>   http://launchpadlibrarian.net/32330771/cross-partition.patch

Hi, thanks for the patch. Please push up a branch and propose a merge.

--
Martin <http://launchpad.net/~mbp/>

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Martin Pool wrote:
> 2009/9/24 Denis Golovan <email address hidden>:
>> ** Attachment added: "Patch for 1.17/1.18"
>> http://launchpadlibrarian.net/32330771/cross-partition.patch
>
> Hi, thanks for the patch. Please push up a branch and propose a merge.
>
>

I'll just do a quick comment.

Using 'shutil.move' seems nice, except instead of failing if the target
is a directory, it will move into a subdirectory.

This is the shutil code:
    real_dst = dst
    if os.path.isdir(dst):
        real_dst = os.path.join(dst, _basename(src))
        if os.path.exists(real_dst):
            raise Error, "Destination path '%s' already exists" % real_dst

^- we don't want this

v- we do what this

    try:
        os.rename(src, real_dst)
    except OSError:
        if os.path.isdir(src):
            if destinsrc(src, dst):
                raise Error, "Cannot move a directory '%s' into itself
'%s'." % (src, dst)
            copytree(src, real_dst, symlinks=True)
            rmtree(src)
        else:
            copy2(src, real_dst)
            os.unlink(src)

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAkq7nfAACgkQJdeBCYSNAAPb7ACeNvXpaqyJVB0K82pGjW2E34ff
SOEAoI8YFmkTReSWUbpr4hWXi+/02970
=khPk
-----END PGP SIGNATURE-----

On Thu Sep 24 16:27:28 UTC 2009 John A Meinel wrote:
> Using 'shutil.move' seems nice, except instead of failing if the target
> is a directory, it will move into a subdirectory.

I also think this behaviour may even vary by python version.

Thanks,

James

For now, would it be more user-friendly to throw the error right at the start when a "bzr commit" or a "bzr add" is made on a file that resides on another filesystem?

The behavior currently is that the error is only thrown during a "bzr revert".

I would rather just fix it than fail early.

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 8/21/2010 9:02 AM, Rahul Nabar wrote:
> For now, would it be more user-friendly to throw the error right at the
> start when a "bzr commit" or a "bzr add" is made on a file that resides
> on another filesystem?
>
> The behavior currently is that the error is only thrown during a "bzr
> revert".
>

I'll note that it will also fail during 'bzr up' or 'bzr pull' or 'bzr
merge', basically any command that generates new file content. 'bzr
commit' works because it only reads from the disk.

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAkxylPwACgkQJdeBCYSNAAOjggCeMhQLUroiTZe8FkWACqxW+zOy
NDUAnR0wukjhTZ0pmR7TQFVw+ugyRN+D
=njno
-----END PGP SIGNATURE-----