[Bug-tar] need help with tar output stability problem

[Joey Hess]

Hi, I maintain pristine-tar[1], which relies on GNU tar outputting a
bit-identical tarball, given identical inputs. That had held true since 
at least tar 1.20. But between tar 1.23 and 1.24, tar began to
output smaller tarballs for some inputs than it had before, which 
while surely a good thing in general unfortunatly breaks pristine-tar.

-rw-r--r-- 1 joey joey 60753920 Nov 10 03:03 1.23/recreatetarball
-rw-r--r-- 1 joey joey 60856320 Nov 10 17:38 1.24/recreatetarball

For other inputs, such as tar's source tree, both versions of tar still 
create identical tarballs. My only hint is that larger tarballs seem
to trigger the difference.

I'd like to have bisected the cause, but I cannot get old versions of
tar to build from git -- it seems that bootstrap assumes that the
current version of gnulib it downloads will work, and old tar needs old
gnulib, which seems to make bisection very hard.

So instead, I though I'd just ask, can someone tell why the output
changed? And is there any way to get the old output, without me having
to include a copy of tar 1.23 in pristine-tar?


The two files above can be found here:

http://kitenet.net/~joey/tmp/1.23/recreatetarball
http://kitenet.net/~joey/tmp/1.24/recreatetarball

This contains the input used to produce both, and a Makefile
that runs tar with options intended to ensure stable output, and
checks if the size of the result matches the 1.23 version.

http://kitenet.net/~joey/tmp/tar-testcase.tgz

-- 
see shy jo

[1] http://kitenet.net/~joey/code/pristine-tar/

signature.asc
Description: Digital signature