Platform Team/Package Management System

From Sugar Labs
Jump to navigation Jump to search
Pencil.png NOTICE:  This page is a draft in active flux...
Please contribute to these contents and discuss issues on the discussion page.


Zero Sugar Distribution

Zero Sugar provides a method that is transparent for users and convenient for doers to deploy software across varying platforms as well as across Sugar releases. Zero Sugar is mainly targeted to support a direct, doer-to-user interaction model, i.e., shortcut the chain of transfer of doer-to-distributor-to-user (still, Zero Sugar could be beneficial for distributors, since it unifies the deployment workflow).

Benefits

How Zero Sugar might be useful depending upon your point of view:

See also other use cases.

Users

Users can run Sugar effortlessly in multiple environments (machine architectures, GNU/Linux distributions, etc.). And since Sugar also stimulates users to be doers, there can be tons of handmade software in the field (mostly duplicates and reinvented wheels, but this is the nature of learning with Sugar—one tries some software, realizes some missing or additional feature, modifies or mimics some code, learns something new, and eventually produces a new creation). It is becoming critical to provide an easy and robust launching scheme for all these pieces of code, without writing long HOWTOs somewhere on a wiki, like "Please, if you are on Ubuntu, install these packages before starting my activity...".

Relying only on a particular distributor, i.e., using activities only from one distributor's channel, makes Sugar less useful or universal. Zero Sugar provides the most convenient and direct deployment method possible between the author of a piece of code and its users.

Doers

Authors of code are always interested in supporting as large an audience of users as possible. And since people can use Sugar in different environments, it is critical to provide an instrument that permits an activity to run on all Sugar platforms. In the case of Python-based activities, the existing practice of just zipping the activity code into .xo files works pretty well. But for arbitrary scripting languages, such as Ruby, that are not part of the Sugar Platform, for binary based activities and for activities that have dependencies that are not included in the Sugar Platform, there is the problem of providing the user with the necessary code.

Zero Sugar tries to solve all the possible issues that prevent activity launching in an heterogeneous Sugar environment.

Also, if you have some development idea in mind, you can start pushing it through the Sucrose release queue to get it into an upcoming (though maybe not the nearest) release. However, since the release process is binary—either your feature is included or not—Zero Sugar offers more flexibility for everyone. For example:

  • some features are of arguable general benefit; Zero Sugar would allow interested end users to efficiently test (and even deploy) such features;
  • some feature are only stable in limited environments; Zero Sugar would allow deployment in limited environments without the risk of destabilizing the rest of Sugar.

Distributors

The Sugar Platform cannot grow endlessly, e.g., installing Sugar from packages should not fetch half of all available packages including several UI toolkits, several programming languages, etc. So, components like Qt or Ruby are (at least at present) not in the Sugar Platform list. But with Zero Sugar, activities could still use such dependencies by installing them from native packaging systems on demand, and without bundling their binaries.

How it works at a glance

Instruments and services that are used within Zero Sugar:

  • 0sugar, the main tool, everything happens via the 0sugar command.
  • 0install decentralized deployment infrastructure.
  • OBS, openSUSE Build Service, build farms and repository of native packages for the GNU/Linux distributions and architectures that OBS supports.
  • PackageKit to install software from native packaging systems.
  • Sugar Labs's resources to host various files.

Identification

Every Zero Sugar package is identified by a Web url like GCompris. The file, url points to, is regular 0install feed (in other words, 0install package) that contains metadata about package itself and all its implementations.

Implementations are Web links to tarballs. Packages are split into implementations to:

  • let user launch not only last version e.g. package can contain implementations for several branches or/and stabilities,
  • split binary implementations per OS and platform,
  • split implementations per language.

Zero Sugar natively supports forks (better name?) that are similar to regular forks in FOSS but could be just results of doers' experiments i.e. without intension to push changes to an upstream. Forks are regular Zero packages and identified by unique Web url but linked to an upstream package (by mentioning upstream url in feed-for metadata field). So, having an upstream activity http://go.sugarlabs.org/Record, http://A.doer.org/My_Record from doer A and http://B.doer.org/My_Record from doer B, user C will have http://go.sugarlabs.org/Record implementations from three sources.

Spec file

The process departing point is a spec file which is used within Zero Sugar to accomplish two major tasks:

  • how to prepare the code to launch,
  • how to share the code.

The Zero Sugar spec file is an all-sufficient entity. With only the spec file and tools like 0sugar and 0distro, it is possible to prepare the code necessary to launch in various environments like the major rpm/deb-based GNU/Linux distributions or just launch locally.

Prepare the code

The preparation step can be trivial, unless the code requires a building stage. Building might occur:

  • on the developer's workstation, to deploy to environments similar to the developer's,
  • on OBS, to build for the GNU/Linux distributions that OBS supports, or
  • building might happen on the user's side, if other methods don't work.

Share the code

Sharing step might be:

  • local, if code needs to be run only in the doer's environment.
    Zero Sugar spec file will be handled as a regular activity.info file.
  • peer-to-peer, direct sharing between doer and users.
    • Code can be launched if on-line users in the Neighborhood View have a copy of shared code, i.e., the doer only needs to be on-line to let other people launch his code.
    • Just by transfering a bundle with the code, e.g., using .xo bundles.
  • client-server, doer needs to upload code to the server, and users will download it.
    The particular method might be different:
    • via the 0install infrastructure,
    • via OBS repositories with native packages,
    • by uploading bundles to servers like ASLO.
  • distributor, most likely similar to client-server, but different from the doer's point of view, since only the distributor is responsible for a particular distribution method.

Regular workflow

The regular workflow within Zero Sugar, in the case of coding a Python-based activity, will look like the following:

  • Create activity spec file.
  • Code the activity.
  • Try current code in Sugar just by selecting an icon in the activities list.
  • When a milestone is achieved, call:
    • 0sugar dist to create sources tarball
    • 0sugar commit to let Sugar know that the activity can be shared in peer-to-peer mode between on-line users
  • If doer wants to support server-client sharing model for a broad audience of users:
    • request an account on sunjammer.sugarlabs.org (or use other servers to rsync files to),
    • call, 0sugar push to publish 0install feeds and files
  • If doer wants to support OBS based sugar distributions or users that prefer activities from native packages:
    • create a project on OBS,
    • call 0sugar push <obs-project>
  • options for distributors:
    • If it is an OBS-based distribution, there is no need for any packaging-related work at all, just link/branch the activity to your distribution project.
    • Reuse the activity spec file to build native packages on non-OBS build farms. Most likely, the resulting packages will not conform to all the requirements for inclusion in an official repository, but this feature could still be useful when strong packaging rules are not required, e.g., in various Sugar/edu derivates like Trisquel-edu, SoaS or USR.
    • Using the sources tarball generated by the 0sugar dist command, and information about dependencies from the spec file, create a regular package for any particular GNU/Linux distribution.

Documentation

Start your exploration of Zero Sugar beginning with these resources:

  • How Zero Sugar looks from the GNU/Linux distribution packaging point of view
  • How Zero Sugar looks from the 0install point of view
  • How Zero Sugar looks from the Sugar Activity developer point of view

See also the Zero Sugar example use cases .