Distributions [LWN.net]

News and Editorials

Gentoo Package Management with Portage

June 9, 2004

This article was contributed by Ladislav Bodnar

Before Gentoo Linux emerged (pun intended) as a powerful force on the Linux distribution scene, software installation on most distributions was a subject of much controversy and frequent criticism. The RPM package manager was often at the receiving end of the blame due to its complex dependency structures; typing "rpm dependency hell" into Google's search engine returns over 700 entries. Debian's implementation of its own package management was often seen as technically superior to those found in RPM-based distributions, but many potential users were put off by Debian's other complexities and a lack of graphical installation and configuration utilities. And while Slackware's own package management does not suffer from dependency issues (simply because it doesn't care about them), many will argue that its pkgtool and other package utilities are rather too basic to qualify as a package management tool in the true sense of the word.

Enter the world of Gentoo Linux. With one short command, a user can install any application, without ever having to worry about libraries and other dependencies. The command will download the source code of the requested package, it will also download the source codes of all the dependent packages (if any), then it will configure, compile and install the package(s) without any further intervention on the user's part. Because the vast majority of packages are compiled directly from source code (with some exceptions, such as OpenOffice, Java, Opera, etc.), the user has complete control over package versions, optimizations, even optional dependencies, in some cases. In fact, the only drawbacks when compared to binary package management tools are longer installation times and fairly stringent hardware requirements - the pleasure of using Gentoo Linux rises exponentially with the power of the CPU at hand.

Portage. The main force behind Gentoo's convenient package management is Portage. Written in Python, Portage not only provides the all-important "emerge" command, it also groups all packages, or ebuilds in Gentoo speak, into a logical tree-like structure in /usr/portage/. Ebuilds can be unmasked, masked or hard masked, with unmasked ebuilds considered stable and well-tested, while masked ebuilds are stable, but not yet tested extensively. Hard masked ebuilds are usually reserved for alpha or beta packages. Only unmasked packages are installed by default, but this can be overridden either in /etc/make.conf or on the command line like this:

# ACCEPT_KEYWORDS="~x86" emerge packagename

The list of current masked packages is stored in /usr/portage/profiles/package.mask. To see the available versions of a package and their masked status in a nicely colored output, we can use the etcat command (part of the gentoolkit package):

# etcat -v mozilla

USE-flags. The USE-flags is a very clever concept, serving as a central place to configure some of the most basic aspects of the operating system and its behavior. As an example, if you prefer to have support for ALSA for all the relevant applications, you have to specify it as a USE-flag. Once done, these applications will be compiled with ALSA support included. If you don't specify it centrally in /etc/make.conf, you would have to compile the applications with the following command:

# USE="alsa" emerge xmms

The above example is still useful in case a user wants to override the global setting. There are over 200 USE-flags listed in /usr/portage/profiles/use.desc; they provide many "uses" that range from Java and SELinux support to specifying the default Japanese input server or enabling potentially offensive items in packages. The USE-flags also specify one of the CPU architectures - the currently available choices include x86, ppc, ppc64, sparc, alpha, mips, hppa, arm and amd64.

CFLAGS. Much has been said about CFLAGS on various forums, especially by those who never read the GCC manual. Probably the most important decision here is the use of march or mcpu when specifying the type of processor. The former will result in faster binaries at the expense of compatibility, while the latter will be produce somewhat slower binaries, but will cover a wider range of processors. As an example, specifying -mcpu=i586 will result in binaries optimized for i586, but they will also execute on i386; however specifying -march=i586 means that the binaries will only execute on i586, but not on i386 CPUs. Another common CFLAG is -O, which specifies a level of optimization ranging from -Os (optimized for small binaries) to -O1, -O2 or -O3 (optimized for speed). GCC provides over a hundred different CFLAGS.

Examples. Finally, let's have a look at some practical examples to illustrate the range of available options in Portage. The handful of the commands listed below are just a tip of the iceberg.

# emerge-webrsync

This is the same as emerge sync (it synchronizes the local portage tree with the latest available portage tree on official repositories), but useful for those users who are behind a firewall blocking the standard rsync port 873

# nice -n19 emerge -u world

This is the same as emerge -u world (it updates all installed packages to their latest available versions), but with a low priority so that the compilation process has a limited effect on the user's ability, in terms of processor resources, to perform other tasks.

# emerge -pv packagename

Lists packages that are dependent, but not yet installed, on packagename.

# emerge /mnt/cdrom/packagename*tbz2

Installs a binary package packagename from the mounted CD-ROM.

# emerge info

Lists all the Portage-specific variables, such as USE, CFLAGS, COMPILER, GENTOO_MIRRORS, etc.

If you are used to the rpm command, you can install Gentoo's epm to query packages. It takes many of the same arguments as rpm itself:

# epm -q packagename

Returns the installed version of packagename

# epm -qa

Returns all installed packages

# epm -qi packagename

Returns information about packagename

# epm -Va

Checks the integrity of all installed packages and reports conflicts

# epm -e packagename

Uninstalls packagename, same as emerge unmerge packagename

Learning about Portage is a journey in itself, a road paved with many pleasant surprises behind every turn. There is no doubt that Portage is one of the most powerful package management utilities available today.

Comments (28 posted)

Distribution News

Gentoo Linux

The Gentoo Weekly Newsletter for the week of May 31, 2004 is out. This issue covers the release of Wasabi, and several other topics.

Here's the announcement for the release of Wasabi 0.1, a custom log monitoring tool for Gentoo systems.

Distributions

News and Editorials

Distribution News

New Distributions

Minor distribution updates

Distribution reviews