rsync - Feb 2024 - rsync replacing symlinks without warning (resend)

Hello,

when trying to rsync files between hosts, I ran into a surprising case
in which rsync replaces a symlink with a directory, with no indication
of any kind.

In the following reproducer, rsync is called as follows:

  rsync --verbose --recursive --relative --delete a/./lib/modules b/

Directory b contains a 'lib' symlink pointing to 'usr/lib', and
rsync
removes that and replaces it with a directory.

In my real-world use case, this caused '/lib' -> '/usr/lib'
symlinks
to be replaced with '/lib' directories, which left the receiving test
machines in a fairly sad state.

I have since figured out that I can get rsync to behave as expected by
adding the --keep-dirlinks option, but ...

it's very unfortunate that when rsync does that kind of thing, it
leaves no indication in the 'rsync --dry-run' and 'rsync -v'
output.
Could that please be fixed?

Thanks,
Andreas


#! /bin/sh

tmp=$(mktemp -dt ${0##*/}.XXXXXXXXXX)
trap 'cd /; rm -rf $tmp' EXIT
cd "$tmp"

umask 022

mkdir -p a/lib/modules
echo foo > a/lib/modules/foo

mkdir -p b/usr/lib/modules
ln -s usr/lib b/lib

show() {
    find "$@" | xargs stat -c "%F %N" | sort -k2
}

echo "from:"
show a

echo
echo "to:"
show b

echo
echo "rsync:"
rsync \
    --verbose \
    --recursive \
    --relative \
    --delete \
    a/./lib/modules \
    b/

echo
echo "to:"
show b

# SCRIPT OUTPUT with rsync 3.2.7:
# =============================# from:
# directory 'a'
# directory 'a/lib'
# directory 'a/lib/modules'
# regular file 'a/lib/modules/foo'
#
# to:
# directory 'b'
# directory 'b/usr'
# directory 'b/usr/lib'
# directory 'b/usr/lib/modules'
# symbolic link 'b/lib' -> 'usr/lib'
#
# rsync:
# sending incremental file list
# lib/
# lib/modules/
# lib/modules/foo
#
# sent 160 bytes  received 47 bytes  414.00 bytes/sec
# total size is 4  speedup is 0.02
#
# to:
# directory 'b'
# directory 'b/lib'
# directory 'b/lib/modules'
# directory 'b/usr'
# directory 'b/usr/lib'
# directory 'b/usr/lib/modules'
# regular file 'b/lib/modules/foo'

rsync's -v is fairly useless.  Learn to use -i instead or in addition to.

On 2/4/24 12:58, Andreas Gruenbacher via rsync wrote:
> Hello,
> 
> when trying to rsync files between hosts, I ran into a surprising case
> in which rsync replaces a symlink with a directory, with no indication
> of any kind.
> 
> In the following reproducer, rsync is called as follows:
> 
>    rsync --verbose --recursive --relative --delete a/./lib/modules b/
> 
> Directory b contains a 'lib' symlink pointing to 'usr/lib',
and rsync
> removes that and replaces it with a directory.
> 
> In my real-world use case, this caused '/lib' ->
'/usr/lib' symlinks
> to be replaced with '/lib' directories, which left the receiving
test
> machines in a fairly sad state.
> 
> I have since figured out that I can get rsync to behave as expected by
> adding the --keep-dirlinks option, but ...
> 
> it's very unfortunate that when rsync does that kind of thing, it
> leaves no indication in the 'rsync --dry-run' and 'rsync
-v' output.
> Could that please be fixed?
> 
> Thanks,
> Andreas
> 
> 
> #! /bin/sh
> 
> tmp=$(mktemp -dt ${0##*/}.XXXXXXXXXX)
> trap 'cd /; rm -rf $tmp' EXIT
> cd "$tmp"
> 
> umask 022
> 
> mkdir -p a/lib/modules
> echo foo > a/lib/modules/foo
> 
> mkdir -p b/usr/lib/modules
> ln -s usr/lib b/lib
> 
> show() {
>      find "$@" | xargs stat -c "%F %N" | sort -k2
> }
> 
> echo "from:"
> show a
> 
> echo
> echo "to:"
> show b
> 
> echo
> echo "rsync:"
> rsync \
>      --verbose \
>      --recursive \
>      --relative \
>      --delete \
>      a/./lib/modules \
>      b/
> 
> echo
> echo "to:"
> show b
> 
> # SCRIPT OUTPUT with rsync 3.2.7:
> # =============================> > # from:
> # directory 'a'
> # directory 'a/lib'
> # directory 'a/lib/modules'
> # regular file 'a/lib/modules/foo'
> #
> # to:
> # directory 'b'
> # directory 'b/usr'
> # directory 'b/usr/lib'
> # directory 'b/usr/lib/modules'
> # symbolic link 'b/lib' -> 'usr/lib'
> #
> # rsync:
> # sending incremental file list
> # lib/
> # lib/modules/
> # lib/modules/foo
> #
> # sent 160 bytes  received 47 bytes  414.00 bytes/sec
> # total size is 4  speedup is 0.02
> #
> # to:
> # directory 'b'
> # directory 'b/lib'
> # directory 'b/lib/modules'
> # directory 'b/usr'
> # directory 'b/usr/lib'
> # directory 'b/usr/lib/modules'
> # regular file 'b/lib/modules/foo'
> 
>