[Box Backup] Box Backup 0.11rc2 on OS X 10.5

Achim boxbackup@boxbackup.org
Wed, 18 Mar 2009 14:44:06 +0100


Hello Chris:

On Tue, 17 Mar 2009 20:51:16 +0000 (GMT), Chris Wilson <chris@qwirx.com>
wrote:
>>> (re: directory timestamps and hard links)
> 
> So far nobody has come to us and said that they need those features. A
few 
> people have asked about directory timestamps, but I think it's more 
> because they were surprised that the directory listing in bbackupquery 
> show 1970-01-01 rather than an actual time for directories.

I found some more information on this topic, in another interesting project
called xar [1] which "aims to provide an easily extensible archive format.
Important design decisions include an easily extensible XML table of
contents for random access to archived files, storing the toc at the
beginning of the archive to allow for efficient handling of streamed
archives, the ability to handle files of arbitrarily large sizes, the
ability to choose independent encodings for individual files in the
archive, the ability to store checksums for individual files in both
compressed and uncompressed form, and the ability to query the table of
content's rich meta-data."

One of the bugs reads [2]: "The unxar'd subdirectories should have the same
timestamps as the original subdirectories. Instead, they have the current
timestamp applied to them. This is likely because OS X modifies the
timestamp on the parent directory when xar creates the files within the
directories." 

The fix [3] was to save directories until the end when extracting. This
lets timestamps get set correctly on directories and addresses.

> OS semantics for directory timestamps on Unix are a bit weird. They
change 
> themselves when some properties of files in the directory are changed,
but 
> don't reflect the latest date of any file in the directory. I can't think

> of a program that uses them. The only application that I can think of
that 
> uses hardlines is Dirvish, another backup application.

What about non-Unix systems, e.g. Windows? Creation time is quite important
for folders, for instance when sorting projects (stored in folders!) by
creation date. Sometimes we advise clients to sort their projects and
related data into current and inactive projects, and creation dates and
modification dates are an important indicator, also on folder level.

Basically, I am convinced that any backup solution (and in particularly one
that is so mature as Box Backup) should preserve as much existing meta
information as possible. The information is available, it does not cost
much to include it, and finally you never know what programs rely on it, so
why toss it away?

I don't want to drag out this discussion unnecessarily: I hope I convinced
you that this is an important (although not critical) issue. If you still
feel that directory timestamps are not relevant, then there is no need to
keep talking about it.

Best regards, Achim

[1] <http://code.google.com/p/xar/>
[2] <http://code.google.com/p/xar/issues/detail?id=58>
[3]
<http://code.google.com/p/xar/source/diff?spec=svn222&r=222&format=side&path=/trunk/xar/src/xar.c>