Seeds

The contents of the Ubuntu distribution, and others, are managed by means of seeds. At their simplest, these are lists of packages which are considered important to have in the main component of the distribution, without explicitly listing all their dependencies and build-dependencies.

Seed lists are typically divided up by category: a base or minimal seed might list the core set of packages required to make the system run at all, while a desktop seed might list the set of packages installed as part of a default desktop installation. germinate takes these seeds, adds their dependency trees, and produces an output for each seed which contains a dependency-expanded list of package names. These outputs may be handed on to archive maintenance or CD-building tools.

Some seeds may inherit from other seeds: they rely on those seeds to be installed. For example, a desktop seed will typically inherit from a minimal seed. germinate understands these inheritance relationships. If a package in the desktop seed depends on ‘foo’, but ‘foo’ is already part of the minimal seed or dependency list, then ‘foo’ will not be added to the desktop output.

Seeds are stored in text files downloaded from a given URL. Lines not beginning with ‘ * ’ (wiki-style list markup) are ignored.

Seed entries may simply consist of a package name, or may include any of the following special syntax:

%

Seed entries beginning with ‘%’ expand to all binaries from the given source package.

...|...

Seed entries may list alternatives to the seeded package to allow metapackages to list them, and users to use those alternatives without having to remove the metapackages.

[...]

Seed entries may be followed with ‘ [arch1 arch2 ...]’ to indicate that they should only be used on the given architectures, or with ‘ [!arch1 !arch2 ...]’ to indicate that they should not be used on the given architectures.

(...)

Seed entries in parentheses indicate that the seed should be treated as a recommendation of metapackages generated from this seed, rather than as a dependency.

!

Seed entries beginning with ‘!’ cause the given package to be blacklisted from the given seed and any seeds from which it inherits; this may be followed by ‘%’ as above to blacklist all binaries from the given source package. Note that this may result in uninstallable packages whose dependencies have been blacklisted, so use this feature sparingly. The purpose of a blacklist is to make it obvious when a package that is not supposed to be installed ends up in germinate's output, so that package relationships can be fixed to stop that happening. It is not intended for the purpose of working around buggy package relationships, and attempts to do so will not work because apt has no way to know about blacklist entries in seeds.

snap:name

Seed entries beginning with ‘snap:’ are snap packages. These are different from deb packages in that they do not have (build-)dependencies, cannot be recommended, and do not end up in any resulting metapackages. (If you try to recommend a snap package, it will be ignored completely.) Snaps specified in seeds will be output in a .snaps file named after the corresponding seed, as software processing the output of germinate will typically need to treat snaps differently from debs. germinate will not check remotely to see if a given snap is available, therefore seeds are expected to explicitly list all architectures a snap is to be seeded on. ‘snap:’ entries can also be suffixed with "/classic" to indicate that the snaps need to be installed with classic confinement on end-user systems.

key: value

Some seeds also contain headers at the top of the file, in “key: value” format. For the most part, these are not parsed by germinate itself. The Ubuntu tasksel package uses keys beginning with ‘Task-’ to define fields of similar names in its .desc files. germinate-update-metapackage(1) uses some of these headers to reduce the need for fragile configuration; see its documentation for further details.

A STRUCTURE file alongside the seeds lists their inheritance relationships. It may also include lines beginning with ‘include’, causing other collections of seeds to be included as if they were part of the collection currently being germinated, or lines beginning with ‘feature’, which set flags for the processing of seeds. Features may also be set on a per-seed basis using lines beginning with ‘ * Feature:’ in the seed file.

The following flags are currently defined:

follow-build-depends

Follow Build-Depends fields. This flag is only recognised in individual seed files, not in STRUCTURE.

follow-build-depends-all

Follow Build-Depends fields for Architecture: all packages. This has no effect if no-follow-build-depends is set.

follow-recommends

Treat Recommends fields as if they were Depends.

no-follow-build-depends

Do not follow Build-Depends fields.

no-follow-build-depends-all

Do not follow Build-Depends fields for Architecture: all packages, even though Build-Depends are followed for other packages.

no-follow-recommends

Do not treat Recommends fields as if they were Depends. This flag is only recognised in individual seed files, not in STRUCTURE.

Build-dependencies and ‘supported’

There is typically no need for a default desktop installation to contain all the compilers and development libraries needed to build itself from source; if nothing else, it would consume much more space. Nevertheless, it is normally a requirement for the maintainers of a distribution to support all the packages necessary to build that distribution.

germinate therefore does not add all the packages that result from following build-dependencies of seed packages and of their dependencies (the “build-dependency tree”) to every output, unless they are also in the seed or in the dependency list. Instead, it adds them to the output for the last seed in the STRUCTURE file, conventionally called supported.

Like any other seed, the supported seed may contain its own list of packages. It is common to provide support for many software packages which are not in the default installation, such as debugging libraries, optimised kernels, alternative language support, and the like.

Outputs

The output files are named after the seed to which they correspond. An additional output file is needed for supported, namely ‘supported+build-depends’, which contains the supported list and the build-depends lists of the other seeds all joined together. An ‘all’ output is produced to represent the entire archive.

Some other files are produced for occasional use by experts. See the README file for full details on these.