Software update: rsync 2.6.7pre1

The first preview of rsync 2.6.7 has appeared on the Samba.org website. The program provides a quick way to sync changes from one file to another location. It is therefore widely used in all kinds of synchronization programs, but also, for example, on file servers to provide snapshot functionality. If you want to know more about this, the documentation which also lists a number of tutorials. The announcement is as follows:

Rsync version 2.6.7pre1 is now available for release testing. Please give it a try and send email to the regular mailing list with any questions, comments, bug reports, etc.

output changes:

• The itemized output now uses ‘S’ for a special file instead of clumping them together with the ‘D’ for devices.
• The way rsync escapes unreadable characters has changed. First, rsync now has support for recognizing valid multibyte character sequences in your current locale, allowing it to escape fewer characters than before for a locale such as UTF-8. Second, it now uses an escape idiom of “\#123”, which is the literal string “\#” followed by exactly 3 octal digits. Rsync no longer doubles a backslash character in a filename (eg it used to output “foo\\bar” when copying “foo\bar”) — now it only escapes a backslash that is followed by a hash-sign and 3 digits ( 0-9) (eg it will output “foo\#134#789” when copying “foo\#789”). See also the –8-bit-output (-8) option, mentioned below.
  Script writers: the local rsync is the one that outputs escaped names, so if you need to support unescaping of filenames for older rsyncs, I’d suggest that you parse the output of “rsync –version” and only use the old unescaping rules for 2.6.5 and 2.6.6.

bug fixes:

• Fixed a really old bug that caused –checksum (-c) to checksum all the files encountered during the delete scan (ouch).
• Fixed a potential hang in a remote generator: when the receiver gets a read-error on the socket, it now signals the generator about this so that the generator does not try to send any of the terminating error messages to the client (avoiding a potential hang in some settings).
• Made hard links work with symlinks and devices again.
• If the sender gets an early EOF reading a source file, we propagate this error to the receiver so that it can discard the file and try requesting it again (which is the existing behavior for other kinds of read errors).
• If a device-file/special-file changes permissions, rsync now updates the permissions without recreating the file.
• If the user specifies a remote-host for both the source and destination, we now output a syntax error rather than trying to open the destination hostspec as a filename.
• When –inplace creates a new destination file, rsync now creates it with permissions 0600 instead of 0000 — this makes restarting possible when the transfer gets interrupted in the middle of sending a new file.
• Reject the combination of –inplace and –sparse since the sparse-output algorithm doesn’t work when overwriting existing data.
• Fixed the directory name in the error that is output when pop_dir() fails.
• Really fixed the parsing of a “!” entry in .cvsignore files this time.
• If the generator gets a stat() error on a file, output it (this used to require at least -vv for the error to be seen).
• If waitpid() fails or the child rsync didn’t exit cleanly, we now handle the exit status properly and generate a better error.
• Fixed some glitches in the double-verbose output when using –copy-dest, –link-dest, or –compare-dest.
• Fixed the matching of the dont-compress items (eg *.gz) against files that have a path component containing a slash.
• If code reading a filter/exclude file an EINTR error, rsync now clears the error flag on the file handle so it can keep on reading.
• If –relative is active, the sending side cleans up trailing “/” or “/.” suffixes to avoid triggering a bug in older rsync versions. Also, we now reject a “..” dir if it would be sent as a relative dir.
• If a non-directory is in the way of a directory and rsync is run with –dry-run and –delete, rsync no longer complains about not being able to opendir() the not-yet present directory.
• Got rid of the need for –force to be used in some circumstances with –delete-after (making it consistent with –delete-before/-during).

Enhancements:

• Added the –specials option to tell rsync to copy special files (and does not require root). The –devices option now affects just character and block devices (which now matches the documentation). The -D option still requests both –devices and –specials, and -a still implies -D.
• Added the –append option that makes rsync append data onto files that are longer on the source than the destination (this includes new files).
• Added the –min-size=SIZE option to exclude small files from the transfer.
• Added the –compress-level option to allow you to set how aggressive rsync’s compression should be (this option implies –compress).
• Enhanced the parsing of the SIZE value for –min-size and –max-size to allow easy entry of multiples of 1000 (instead of just multiples of 1024) and off-by-one values ​​too (eg –max-size =8mb-1).
• Added the –8-bit-output (-8) option, which tells rsync to avoid escaping high-bit characters that it thinks are unreadable in the current locale.
• The new option –human-readable (-h) changes the output of –progress, –stats, and the end-of-run summary to be easier to read. If repeated, the units become powers of 1024 instead of powers of 1000.
• If lutimes() and/or lchmod() are around, use them to allow the preservation of attributes on symlinks.
• The –link-dest option now affects symlinks and devices (when possible).
• Improved the output of hard-linked and copied files when using –link-dest, –copy-dest, or –compare-dest.
• Added two config items to the rsyncd.conf parsing: “pre-xfer exec” and “post-xfer exec”. These allow a command to be specified on a per-module basis that will be run before and/or after a daemon-mode transfer. (See the manpage for a list of the environment variables that are set with information about the transfer.)
• When using the –relative option, you can now insert a dot dir in the source path to indicate where the replication of the source dirs should start. For example, if you specify a source path of rsync://host/module/foo/bar/./baz/dir with -R, rsync will now only replicate the “baz/dir” part of the source path (note: a trailing dot dir is unaffected unless it also has a trailing slash).
• Added some new –no-FOO options that make it easier to override unwanted implied or default options. For example, “-a –no-o” (aka “–archive –no-owner”) can be used to turn off the preservation of file ownership that is implied by -a.
• Added the –chmod=MODE option that allows the destination permissions to be changed from the source permissions. Eg –chmod=g+w,o-rwx
• Added the “incoming chmod” and “outgoing chmod” daemon options that allow a module to specify what permissions changes should be applied to all files copied to and from the daemon.
• Allow the –temp-dir option to be specified when starting a daemon, which sets the default temporary directory for incoming files.
• If –delete is combined with –dirs without –recursive, rsync will now delete in any directory whose content is being synchronized.
• If –backup is combined with –delete without –backup-dir (and without –delete-excluded), we add a “protect” filter-rule to ensure that files with the backup suffix are not deleted.
• The file-count stats that are output by –progress were improved to better indicate what the numbers mean. For instance, the output: “(xfer#5, to-check=8383/9999)” indicates that this was the fifth file to be transferred, and we still need to check 8383 more files out of a total of 9999.
• The include/exclude code now allows a dir/*** directive (with 3 trailing stars) to match both the dir itself as well as all the content below the dir (dir/** would not match the dir).
• Added the –prune-empty-dirs (-m) option that makes the receiving rsync discard empty chains of directories from the file-list. This makes it easier to selectively copy files from a source hierarchy and end up with just the directories needed to hold the resulting files.
• If the –itemize-changes (-i) option is repeated, rsync now includes unchanged files in the itemized output (similar to -vv, but without all the other verbose messages that can get in the way). Of course, the client must be version 2.6.7 for this to work, but the remote rsync only needs to be 2.6.7 if you’re pushing files.
• Added the –super option to make the receiver always attempt super-user activities. This is useful for systems that allow things such as devices to be created or ownership to be set without being UID 0, and is also useful for someone who wants to ensure that errors will be output if the receiving rsync isn’t being run as root .
• Added the –sockopts option for those few who want to customize the TCP options used to contact a daemon rsync.
• Added a way for the –temp-dir option to be combined with a partial-dir setting that lets rsync avoid non-atomic updates (for those times when –temp-dir is not being used because space is tight).
• A new support script, files-to-excludes, will transform a list of files into a set of include/exclude directives that will copy those files.
• A new option, –executability (-E) can be used to preserve just the execute bit on files, for those times when using the –perms option is not desired.
• The daemon now logs each module-list request it receives.
• New log-format options: %M (modtime), %U (uid), %G (gid), and %B (permission bits, eg “rwxr-xrwt”).
• The –dry-run option no longer forces the enabling of –verbose.
• Some minor documentation improvements.
• Updated some diffs in the patches dir.

Internal:

• We now use sigaction() and sigprocmask() if possible, and fall back on signal() if not. Using sigprocmask() ensures that rsync enables all the signals that it needs, just in case it was started in a masked state.
• Some buffer sizes were expanded a bit, particularly on systems where MAXPATHLEN is overly small (eg cygwin).
• If io_printf() tries to format more data than fits in the buffer, exit with an error instead of transmitting a truncated buffer.
• If a va_copy macro is defined, lib/snprintf.c will use it when defining the VA_COPY macro.
• Reduced the amount of stack memory needed for each level of directory recursion by nearly MAXPATHLEN bytes.
• The wildmatch function was extended to allow an array of strings to be supplied as the string to match. This allows the exclude code to do less string copying.
• Got rid of the safe_fname() function (and all the myriad calls) and replaced it with a new function in the log.c code that filters all the output going to the terminal.
• Unified the f_name() and the f_name_to() functions.

Developer related:

• The diffs in the patches dir now require “patch -p1
• Several diffs in the patches dir now use the proper –enable-FOO configure option instead of –with-FOO to turn on the inclusion of the newly patched feature.
• There is a new script, “prepare-source” than can be used to update the various generated files (proto.h, configure, etc.) even before configure has created the Makefile (this is mainly useful when patching the source with a patch that doesn’t affect generated files).
• The testsuite now sets HOME so that it won’t be affected by a file such as ~/.popt.