- - Added the --only-write-batch=FILE option that may be used (instead
- of --write-batch=FILE) to create a batch file without doing any
- actual updating of the destination. This allows you to divert all
- the file-updating data away from a slow data link (as long as you
- are pushing the data to the remote server when creating the batch).
+ - A new incremental-recursion algorithm is now used when rsync is talking
+ to another 3.0.0 version. This starts the transfer going more quickly
+ (before all the files have been found), and requires much less memory.
+ See the --recursive option in the manpage for some restrictions.
+
+ - The default --delete algorithm is now --delete-during when talking to a
+ 3.x rsync. This is a faster scan than using --delete-before (which is
+ the default when talking to older rsync versions), and is compatible
+ with the new incremental recursion mode.
+
+ - Added the --delete-delay option, which is a more efficient way to
+ delete files at the end of the transfer without needing a separate
+ delete pass.
+
+ - You may specify --max-delete=0 to a 3.0.0 client as long as the
+ receiving side is at least version 3.0.0. This means that you
+ can pull from an older rsync with this option, but pushing to an
+ older rsync will generate an error. *Be sure to never specify a 0
+ value to an older rsync client, or it will be silently ignored.*
+
+ - The --hard-link option now uses less memory on both the sending and
+ receiving side for all protocol versions. For protocol 30, the use
+ of a hashtable on the sending side allows us to more efficiently
+ convey to the receiver what files are linked together. This reduces
+ the amount of data sent over the socket by a considerable margin
+ (rather than adding more data), and limits the in-memory storage of
+ the device+inode information to just the sending side for the new
+ protocol 30, or to the receiving side when speaking an older protocol
+ (note that older rsync versions kept the device+inode information on
+ both sides).