This document describes the (auto) build setup for maemo-on-devuan. Jenkins is used, together with jenkins-debian-glue, and a set of custom scripts

This document hopefully helps those seeking to either help our maemo-on-devuan by adding more repositories and packages, or those who want to set up similar infrastructure.

All repositories currently exist on (git.devuan.org/maemo|github.com/fremantle-gtk2 - pick one). If the source for a package does not yet exist, please create such a repository on the right git host.

There are various different cases when one wants to add dependencies / import packages:

1. Source exists for the package, and it's already available and prepared (e.g. anything on github.com/fremantle-gtk2)
2. Source exists but it's not yet available/packaged in a sensible place for us
3. Only a .deb exists, but the package is source only
4. Only a .deb exists, but it contains binaries and no source code is available.

In cases 1-3, you should be able to import a package with relative ease. In the last case (4), we need to reverse engineer the package before we attempt to import it.

The repository name does not have to correspond to the exact package name, as the package names are described in the debian/ directory instead.

Steps:

1. Add maemo/RELEASE branch (where RELEASE is ascii, ascii-devel, beowulf) that contains the debian/ directory

2. Ensure debian/control follows recent practices (if not, you'll find out during build)

3. Add debian/gbp.conf

   [DEFAULT]
   # this is the format of the git tag where we take the actual software version
   upstream-tag=%(version)s

4. To build a package, please see Making a release of a package in the repository

Jenkins will always fetch the latest maemo/<release name> branch, read debian/gbp.conf and the debian/changelog file. It will pick the latest version from the debian/changelog, and determine what tag to load the code from, based on the upstream-tag provided above in debian/gbp.conf. The source code branch needs to be tagged with such tags.

1. Add a tag to the branch we want to take the source code from. The tag should match the version in the debian/changelog.
2. Add a new debian/changelog entry if needed. Make sure the version you're adding has a proper tag in the git log.
3. To actually get Jenkins to build the package, read Jenkins job management, and specifically Defining jobs.

Once we reach an agreement on the exact version strings, document that here.

All jenkins jobs descent (are copies of) the following three jobs:

1. jenkins-debian-glue-source
2. jenkins-debian-glue-binaries
3. jenkins-debian-glue-repos

The jenkins-debian-glue-source job will fetch any new sources and trigger a jenkins-debian-glue-binaries job. The jenkins-debian-glue-binaries job will build packages on the machines labeled with the respective architecture. Finally, jenkins-debian-glue-repos will copy all artifacts to the master node and update the repository with packages.

Section Defining jobs documents how to add a job to the currently existing set of jobs. This should be all that is required to start building a package, given that it's dependencies are present.

Currently the jobs are all described in a python config file.

The job name does not have to correspond to the exact package name, as the package names are described in the debian/ directory instead.

config.py

_jobs = {
    # job name is key, values can be:
    # 'repo-name': required if repo name is not the same as job name
    # 'host': required if host is not https://github.com/maemo-leste/%s
    # 'releases': {'leste': 'ascii', 'extras': 'beowulf'}
    'libcal': {'repo-name': 'libcal'}
}

https://jenkins-debian-glue.org/faq/#slave_nodes

For slave nodes, we install the system the same way we did with the master node. However, Jenkins does not have to be running on the slave node. Our master node will handle this when we introduce the node to it. The only thing that matters is having an accessible ssh port and ssh keys setup between the two jenkins accounts (master - slave).

We can add a slave node using the HTTP interface of the master node.

1. Manage Jenkins
2. Manage Nodes
3. New node

On the New node screen, we have to configure our slave node:

• Node name should be the architecture we are building for.
• Set Permanent Agent

Once this is set, proceed with configuring as follows:

• Name: armhf
• Description: armv7 machine
• # of executors: 2
• Remote root directory: /var/lib/jenkins (this is ~jenkins on master)
• Labels: armhf
• Usage: Only build jobs with label expressions matching this node
• Launch method: Launch slave agents via SSH
• Host: armhf.slave.maemo.org
• Credentials: setup proper credentials here
• Host key verification strategy: Known hosts file (this requires a manual login once beforehand)
• Port: 22 (or whatever is set up on the slave)

Click Save, and this should setup the node.

Since we have two nodes now, this means the master node needs to be correctly labeled as well. Simply go to the configuration menu in the UI and add a label to the master node matching the architecture that will be built using the master node. In our case this is amd64.

Make sure you add the new architecture to the reprepro configuration(s) where they are located on the master node. In our case this is /srv/repository/conf and /srv/repository/release/leste/conf.

The dch_hook.sh file is a shell script containing our custom versioning logic. It's defined in our sources.xml template, and executed at some point in the process of generate-git-snapshot. For it to work as we have intended, some changes are required in /etc/jenkins/debian_glue:

USE_ORIG_VERSION=true
SKIP_DCH=true
UNRELEASED_APPEND_COMMIT=false

The concept of this hook is that all our built packages will have a maemo version appended on the end of the filename and in the changelog. With this we avoid polluting the git repo, but it still allows us to see how many builds have been triggered for a specific package.

Let's take libcal as an example. In git, our changelog has the version of 0.3-2. When the source job is built, generate-git-snapshot combined with dch-hook.sh will append +0m7 to it for the initial build. This results in the actual package version being 0.3-2+0m7. If we rebuild this package without bumping the version in git, the following build will have +0m7.1 appended. Each subsequent build will increment the latter .1 by one, until the version in git is changed. It will then reset back to +0m7 and repeat the process.