Rdiff-backup

Now every so often (default is 20 seconds, the --checkpoint-interval option controls it) rdiff-backup checkpoints by dumping its state to temporary files in the rdiff-backup-data directory.
If rdiff-backup is rerun with the same destination directory, it can either try to resume the previous backup or at least clean things up so the archive is consistent and accurate.

Added new options --resume, --no-resume, and --resume-interval, which control when rdiff-backup tries to resume a previous failed backup.

Fixed a bug with the --exclude-device-files option which caused the option to be ignored when the source directory was remote.

By default, if rdiff-backup encounters a certain kind of IOError (currently types 26 and 5) while trying to access a file, it logs the error, skips the file, and tries to continue.

If settings requiring an integer argument (like -v or --checkpoint-interval) are given a bad (non-integer) argument, fail with better explanation.

Fixed annoying logging bug.
Now no matter which computer a logging message originates on, it should be routed to the process which is writing to the logging file, and written correctly.
However, logging messages about network traffic will not be routed, as this will generate more traffic and lead to an infinite regress.

When calling rdiff, uses popen2.Popen3 and os.spawnvp instead of os.popen and os.system.
This should make rdiff-backup more secure.
Thanks to Jamie Heilman for the suggestion.

Instead of calling the external shell command 'stat', rdiff-backup uses os.lstat().st_rdev to determine a device file's major and minor numbers.
The new method should be more portable.
Thanks to Jamie Heilman for the suggestion.

All the file operations were examined and tweaked to try to minimize/eliminate the chance of leaving the backup directory in an inconsistent state.

Upon catchable kinds of errors, try to checkpoint before exiting so later rdiff-backup processes have more information to work with.

At the suggestion of Jason Piterak, added a --windows-time-format option so rdiff-backup will (perhaps) work under MS windows NT.

Applied Berkan Eskikaya's "xmas patch" (I was travelling and didn't have a chance on Christmas).
He fixed important bugs in the --terminal-verbosity and --remove-older-than options.

Added an --exclude-device-files option, which makes rdiff-backup skip any device files in the same way it skips files selected with the --exclude option.

Plugged another memory hole.
At first I thought it might have been python's fault, but it was all me.
If rdiff-backup uses more than a few megabytes of memory, tell me because it is probably another memory hole..

rdiff-backup is now a bit more careful about deleting temporary files it creates when it is done with them.

Changed the rpm spec a little.
The enclosed man page is gzipped and the package file is GPG signed (it can be checked with, for example, "rpm --checksig -v rdiff-backup-0.4.3-1.noarch.rpm").

rdiff-backup no longer checks the mtimes or atimes of device files.
Use of these times was inconsistent (sometimes writing to device files updates their times, sometimes not) and leads to unnecessary backing up of files.

Significant speed increases (maybe 20% for local sessions) when dealing with directories that do not need to be updated much.

Fixed memory leak.
rdiff-backup should now run in almost constant memory (about 6MB on my system).

Enabled buffering of object transfers, so remote sessions can be 50-100%+ faster.

rdiff-backup now thinks it is running as root if the destination connection is root.
Thus rdiff-backup will preserve ownership even if it is not running as root on the source end.

If you abort rdiff-backup or it fails for some reason, it is now more robust about recovering the next time it is run (before it could fail in ways which made subsequent sessions fail also).
However, it is still not a good idea to abort, as individual files could be in the process of being written and could get corrupted.

If rdiff-backup encounters an unreadable file (or, if --change-source-perms is given, a file whose permissions it cannot change), it will log a warning, ignore the file, and continue, instead of exiting with an error.

Now either the source, or the target, or both can be remote.
To make this less confusing, now rdiff-backup supports host::file notation.
So it is legal to run:

rdiff-backup billhost1.net::source_file joneshost2.net::target

Also, the test suites have been improved and found a number of bugs (which were then fixed).

Much of the rdiff-backup internals were rewritten.
The result should be better performance when operating remotely over a pipe with significant latency.
Also the code dealing with changing permissions is much cleaner, and should generalize later to similar jobs (for instance preserving atimes.)

Listing and deleting increments and restoring should work remotely now.
In earlier versions a file or directory had to be restored locally and then copied over to its final destination.

At the request of the FSF, a copy of the GPL has been included in the packaged distributions.
It is in the file "COPYING".