Chapitre 5. Système de paquetage, outils et principes fondamentaux

En tant qu'administrateur de système Debian, vous allez régulièrement manipuler des paquets (fichiers .deb) car ils abritent des ensembles fonctionnels cohérents (applications, documentations...) dont ils facilitent l'installation et la maintenance. Mieux vaut donc savoir de quoi ils sont constitués et comment on les utilise.

Vous trouverez ci-après la description des structures et contenus des paquets de type « binaire », puis « source ». Les premiers sont les fichiers directement utilisables par dpkg alors que les seconds contiennent les codes sources des programmes ainsi que les instructions pour créer les paquets binaires.

5.1. Structure d'un paquet binaire

The Debian package format is designed so that its content may be extracted on any Unix system that has the classic commands ar, tar, and xz or sometimes gzip or bzip2. This seemingly trivial property is important for portability and disaster recovery.

Imagine, for example, that you mistakenly deleted the dpkg program, and that you could thus no longer install Debian packages. dpkg being a Debian package itself, it would seem your system would be done for... Fortunately, you know the format of a package and can therefore download the .deb file of the dpkg package and install it manually (see sidebar OUTILS dpkg, APT et ar). If by some misfortune one or more of the programs ar, tar or gzip/xz/bzip2 have disappeared, you will only need to copy the missing program from another system (since each of these operates in a completely autonomous manner, without dependencies, a simple copy will suffice). If your system suffered some even more outrageous misfortune, and even these don't work (maybe the deepest system libraries are missing?), you should try the static version of busybox (provided in the busybox-static package), which is even more self-contained, and provides subcommands such as busybox ar, busybox tar and busybox xz.

In case of a misfortune, you better also have a backup of your system (see Section 9.10, « Sauvegarde »).

OUTILS dpkg, APT et ar

dpkg is the program that handles .deb files (binary packages), notably extracting, analyzing, and unpacking them.

APT (the abbreviation of "Advanced Packaging Tool") is a group of programs that allows the execution of higher-level modifications to the system: installing or removing a package (while keeping dependencies satisfied), updating and upgrading the system, listing the available packages, etc.

As for the ar program, provided by the binutils package, it allows handling files of the same name:

ar t archive displays the list of files contained in such an archive,
ar x archive extracts the files from the archive into the current working directory,
ar d archive file deletes a file from the archive, etc.

Its man page ar(1) documents all its other features. ar is a very rudimentary tool that a Unix administrator would only use on rare occasions, but admins routinely use tar, a more evolved archive and file management program. This is why it is easy to restore dpkg in the event of an erroneous deletion. You would only have to download the Debian package and extract the content from the data.tar.xz archive in the system's root (/):

# ar x dpkg_1.21.22_amd64.deb
# tar -C / -p -xJf data.tar.xz

B.A.-BA Notation des pages de manuel

Il est déroutant, pour les néophytes, de trouver dans la littérature des références à « ar(1) ». Il s'agit en fait généralement d'une notation commode pour désigner la page de manuel intitulée ar dans la section 1.

Il peut aussi arriver que cette notation soit utilisée pour lever des ambiguïtés, par exemple pour différencier la commande printf que l'on pourra désigner par printf(1) et la fonction printf du langage C, que l'on pourra désigner par printf(3).

Le Chapitre 7, Résolution de problèmes et sources d'informations revient plus longuement sur les pages de manuel (voir Section 7.1.1, « Les pages de manuel »).

Examinons le contenu d'un fichier .deb :

$ ar t dpkg_1.21.22_amd64.deb
debian-binary
control.tar.gz
data.tar.xz
$ ar x dpkg_1.21.22_amd64.deb
$ ls
control.tar.gz  data.tar.xz  debian-binary  dpkg_1.21.22_amd64.deb
$ tar tJf data.tar.xz | head -n 16
./
./etc/
./etc/alternatives/
./etc/alternatives/README
./etc/cron.daily/
./etc/cron.daily/dpkg
./etc/dpkg/
./etc/dpkg/dpkg.cfg
./etc/dpkg/dpkg.cfg.d/
./etc/logrotate.d/
./etc/logrotate.d/alternatives
./etc/logrotate.d/dpkg
./lib/
./lib/systemd/
./lib/systemd/system/
./lib/systemd/system/dpkg-db-backup.service
$ tar tJf control.tar.xz
./
./conffiles
./control
./md5sums
./postinst
./postrm
./prerm
$ cat debian-binary
2.0

Comme vous le voyez, l'archive ar d'un paquet Debian est constituée de trois fichiers:

debian-binary: This is a text file which simply indicates the version of the .deb file package format version. In Debian Bookworm it is still version 2.0.
control.tar.xz: This archive file contains all of the available meta-information, like the name and version of the package as well as some scripts to run before, during or after (un-)installation of it. Some of the meta-information allows package management tools to determine if it is possible to install or uninstall it, for example according to the list of packages already on the machine, and if files shipped have been modified locally.
data.tar.xz, data.tar.bz2, data.tar.gz: This archive contains all of the files to be extracted from the package; this is where the executable files, libraries, documentation, etc., are all stored. Packages may use different compression formats, in which case the file will be named differently for xz, bzip2 or gzip.