Splitting a Debian package

Some applications are using tons of data files, which then leads to very large Debian packages that use a lot of disk space. We have identified two ways to reduce the required disk space: resampling files or removing some of them. The second alternatives supposes that we are able to split a package into small sub-packages. As we are about to see, this is not that complicated :).

Get the source

The first step of course is to fetch the source code of the package from Debian repositories. To do this, there is a very convenient command apt-get source [1]. As a Debian source package is made of several source files, it is better to create a directory per application:

$ mkdir <pkgname>
$ cd <pkgname>
$ apt-get source <pkgname>

This will fetch the source package then build the source code of the application patched by Debian developers. The source code directory should be named <pkgname>-<version>, where version is usually x.x or x.x.x, for example 3.2 or 5.2.4.

Tweak the package

Splitting a Debian package is not complicated. First of all, you need to locate the path of package data files. You can either open the binary Debian package (this is a compressed archive file), or build the Debian package from within its source code directory:

$ cd <pkgname>-<version>
$ dpkg-buildpackage

This will create a directory in the directory debian/ which will contains all the files of the binary package. The name of this directory is usually <pkgname>.

Now you can start to split the package. The principle is very simple, in the directory debian/:

• create text files named <pkgname>-<part>.install and a file named <pkgname>.install for the main package.
• edit the file control which controls how many packages will be built
• edit the file changelog in order to tell that this is a modified package

Details are given in the following paragraphs.

The .install files

The files <pkgname>-<part>.install contain the full paths of files or directories to be copied into the sub-package, one per line. For example, the main TBO package contains in its file tbo.install:

/usr/bin
/usr/share/tbo/icons
/usr/share/tbo/ui

while the common data package tbo-common-data.install contains:

/usr/share/applications
/usr/share/pixmaps
/usr/share/locale
/usr/share/tbo/tutorial.pdf
/usr/share/tbo/tut.tbo
/usr/share/tbo/icon.png

The control file

The control file sets package information such as dependencies, target architecture or description. Several information block need to be added to match the .install files (one block per file). Of course dependencies may need to be adjusted differently per sub-package. For example for TBO, the first four package information blocks are:

Package: tbo
Architecture: any
Depends: ${shlibs:Depends}, libgtk2.0-0, ${misc:Depends}, tbo-common-data, tbo-data-doodle-bubble
Recommends: tbo-data-accesories, tbo-data-arcadia, tbo-data-bubble, tbo-data-political, tbo-data-tbo
Suggests: tbo-data-dgm, tbo-data-doodle1, tbo-data-facilware, tbo-data-guadalinex, tbo-data-linuxhispano, tbo data-southpark
Description: Intuitive GNOME comic creator
 TBO is a simple and easy drag and drop comic creator. It uses a libray of SVG draws as input to make the comic frames.
 .
 This package provides the core program without SVG libraries. Libraries are installed by the packages named tbo-data-doodle-*.
 
Package: tbo-data-doodle-all
Architecture: all
Depends: tbo, tbo-data-accesories, tbo-data-arcadia, tbo-data-bubble, tbo-data-political, tbo-data-tbo, tbo-data-dgm, tbo-data-doodle1, tbo-data-facilware, tbo-data-guadalinex, tbo-data-linuxhispano, tbo-data-southpark
Description: Intuitive GNOME comic creator
 TBO is a simple and easy drag and drop comic creator. It uses a libray of SVG draws as input to make the comic frames.
 .
 This package is a meta package that will install all the TBO SVG libraries.

Package: tbo-common-data
Architecture: all
Depends: tbo
Description: Intuitive GNOME comic creator
 TBO is a simple and easy drag and drop comic creator. It uses a libray of SVG draws as input to make the comic frames.
 .
 This package provides the common data of the program, without any SVG libraries. Libraries are installed by the packages named tbo-data-doodle-*.

Package: tbo-data-doodle-accesories
Architecture: all
Depends: tbo
Description: Intuitive GNOME comic creator
 TBO is a simple and easy drag and drop comic creator. It uses a libray of SVG draws as input to make the comic frames.
 .
 This package provides the SVG library subset named accesories.

• Please note that you’ll have to keep the first information block as is, since it targets the source package, not the binary ones. The architecture all leads to a package *_all.deb that is architecture-independent, while the architecture any tells that source code can be compiled onto any architecture. The later case will lead to a package *_i386.deb on i386, *_amd64.deb on amd64, etc.

The changelog file

This file summarizes the changes of the packages and defines its version number. It is better to change the version number as soon as anything is modified. The convention for DoudouLinux is to append doudouN, where N is a number, to the official package version. For example the package plymouth 0.8.3-20 would become plymouth 0.8.3-20doudou1. This is done by adding this kind of information to changelog:

tbo (0.98~git20110321-0doudou2) unstable; urgency=low

  * Split package into core and data sets.

 -- Bob Author <bob.author@doudoulinux.org>  Sat, 19 Nov 2011 17:24:12 +0100

You will have to write the correct summary of your changes, as well as to set your real name and email ;).

Rebuild the package

Rebuilding a Debian binary package is quite simple, the commands were indicated above:

$ cd <pkgname>-<version>
$ dpkg-buildpackage

Note that this will build both source and binary packages. A source package is a file *.dsc which contains references to one or two other files. These files are the original source code of the application, a file *.orig.tar.gz, and the eventual Debian patches that were applied to the original source code.