Re: [Bug-tar] creating identical .tar.gz files for use with version control

[Tim Kientzle]

Sergey Poznyakoff wrote:
> Chris Pickett <address@hidden> ha escrit:
>
> > However, if I repeat the process without changing the underlying
> > files, a.tar.gz is different, and thus it needs to get checked in
> > again.
>
> The tar archive keeps, apart from the files themselves, some metadata,
> e.g. their mtime, so the two archive may differ even though they contain
> exactly the same files.

Perhaps more importantly, the default GNU tar format
stores atime, which could cause differences even
if the files are unchanged.

Try --format=ustar and see if that helps.  ustar
format does not store atime, and mtime should
only change if you actually change the file.

Another problem you may run into:  I don't remember
how GNU tar handles this, but bsdtar stores the
files in the order that the OS returns them from
readdir().  On some newer filesystems, this can
vary essentially at random.  You may be able to
control this by using explicit filenames in the
tar command line, e.g.,
   tar czf output.tgz a/file1 a/file2 ...
or by otherwise forcing the filenames into a
fixed order:
   find a | sort | tar czf output.tgz -T -

Good luck,

Tim