anchoring of patterns for per-directory merge files

[Andy Smith]

Hi,

I'm extremely confused about anchoring of patterns for per-directory
merge files, as what I see seems to not be the behaviour the
documentation suggests.

On the receiver side I have rsync  version 3.0.3  protocol version
30.  On the sender I have rsync  version 2.6.9  protocol version 29.
I'm calling:

rsync --filter='dir-merge /.rsync-filter' -avz --delete \
--numeric-ids --relative --delete-excluded --sparse --hard-links \
--rsh='/usr/bin/ssh -4 -i /root/.ssh/rsync -q' \
root at corona:/data \
/data/backup/.sync/corona/

On the sender, I have in the file "/data/home/andy/.rsync-filter":

- /tmp/

The documentation leads me to believe that this should exclude a
directory called "tmp/" located only in the same directory as the
merge file.  What happens is, nothing gets excluded.

If I use the following content instead:

- tmp/

then every path containing a directory called "tmp/" is excluded.

I found I can also make it more precise by using:

- data/home/andy/tmp/

to exclude just the path I wanted.  It's workable but I would fear
that somehow a path like this might be matched by chance in a big
directory tree.

So why doesn't "- /tmp/" work?

The documentation also says that if I am confused over what to
exclude, I should just use the path as printed by --dry-run -v and
put a "/" on the start. I tried that:

- /data/home/andy/tmp/

but this also does not match.

What am I doing wrong?

Cheers,
Andy