Sometimes you have to repack the upstream tarball to remove some stuff. There are several workflows for that, like manually editing the tarball, the git-buildpackage dfsg_clean branch, cleanup scripts ...
The best practice at the moment is to use the Files-Excluded field in debian/copyright. Here is how you do that.
1. Choose a repack suffix
The version string has this format: [epoch:]upstream_version[-debian_revision]. Repacked tarballs are denoted by adding a repack suffix to the upstream_version part of the version string. The repack suffix has this format: [repack_separator]repack_reason[repack_version]. For example +dfsg, where + is repack_separator and dfsg is repack_reason and repack_version is not specified. So a package name like node-parse-glob_3.0.4-1 becomes node-parse-glob_3.0.4+dfsg-1.
- The common repack reasons are:
ds ("Debian Source"), used when embedded 3rd-party libraries or the upstream debian directory are removed.
The repack_version is a small integer, used to version the repacks as in dfsg, dfsg1, dfsg2 ...
The repack_separator is optional and only serves to improve legibility: some people like to append repack_reason[repack_version] without separator. Based on the admissible non-alphanumerics characters in the upstream_version, and excluding . - : (full stop, hyphen and colon) because they have other meanings, possible separators are + ~ (plus and tilde).
The most commonly used separator is +. If you separate the repack suffix with +, it means that you believe that whatever upstream does with the tarball for this upstream version, you'll always have to repack. You can decide to stop repacking only when upstream releases a new tarball with a new, higher upstream version (but you can always fix it with the epoch).
The ~ is special because it sorts before anything, and is commonly used for pre-releases. So the consequences of using ~ as repack_separator are:
When a dependency on this package is specified, it has to be >= x.y.z~ as opposed to commonly used >= x.y.z.
If upstream re-releases the same version but with the DFSG violation fixed, or if we change the rules and the offending file was now "free", then you can just drop the ~dfsg and release it again with the same upstream_version: the new version will be "higher".
In the following, let's assume the chosen repack suffix is +dfsg.
2. Mark the files to exclude
Append to the first section of debian/copyright (no newline inbetween) a single Files-Excluded field with the list of all excluded files:
Files-Excluded: non_dfsg_compliant_file.extension Comment: reason for exclusion
The Comment field with the reason for exclusion is mandatory as per Debian Policy 12.5.
In the Files-Excluded field you can use the same simplified shell glob syntax of the Files field (see Machine-readable debian/copyright file specification), list files etc.
If more than one file/globbing pattern is required, you can use:
Files-Excluded: file_a file_b file_c dist ... Comment: The following files were removed because file_a: Binary with no source file_b: non-free license file_c: just cruft dist: only generated files in this directory ...
3. Do the actual repack
Add the chosen repack suffix to debian/watch params, for example:
Note: xz offers better compression. Seean example watch file.
Remove the packagename_X.Y.Z.orig.tar.gz tarball and do
Thanks to the magic of uscan, this command will download the upstream tarball, remove the files are save it as packagename_x.y.x+dfsg.orig.tar.gz. No need to edit the ...orig.tar, remove the file and save it manually as ...+dfsg.orig.tar.gz.
Now set aside the uncommitted changes to the git repo (git stash) and import the repacked tarball:
gbp import-orig --pristine-tar ../packagename_X.Y.Z+dfsg.orig.tar.gz
Now restore the changes to debian/copyight and debian/watch (git stash apply) and edit the changelog - the version in the 1st line should read:
packagename (X.Y.Z+dfsg-1) unstable; urgency=low
Finally commit your changes with:
git commit -am 'repack to ...'
- Based on a tip from Jérémy Lal, with contributions by Pirate Praveen, Ross Gammon and Jonas Smedegaard.