The Snapcraft Summit Montréal took place between June 11th and June 13th, 2019 and people from different open-source projects gathered to iterate and work in a flawless manner, that would have taken far more time and resources if it wasn’t for the summit. The shared goal was to improve the presence of each individual project in the snapcraft.io store. The plus side of the story was being able to meet the people behind different projects and share past experiences and current ideas.
Day by day, there were steady updates for the projects. Some were not new to the platform and sought improvements and problem solving. Others learnt about the confinement capabilities, the different pieces that make the store come to life, and take the first steps towards integrating them. At the end of each day a massive status check of almost forty parties was held, and a summary was dropped in forum.snapcraft.io.
Before jumping into the technical changes for Crystal itself, I would like to mention the people with whom I shared some chats, ideas and work that definitely can impact in future proposals. These people are Daniel Silverstone from rustup, Ana Rosas & María de Antón from TravisCI, Alex Arslan & Jeff Bezanson from Julia, Graham Christensen from NixOS, and Sergio Schvezov from Snapcraft. Meeting them helped me get to know better some aspects of the projects, challenges, and solutions that are not in clear sight. I can’t stress enough how much I enjoyed meeting them all.
- Crystal has a new official distribution method in snapcraft.io/crystal that applies to 10 Linux distros (and counting…)
- Crystal applications can be packaged and distributed as snaps.
- Nightly native packages have now a straightforward way to be installed.
- TravisCI is using the above method when
crystal: nightlyis specified.
You can jump and find the snap installation instructions in the new shinny snapcraft.io/crystal page.
The store has the notion of channels, which helps support the release process with edge, beta, candidate, and stable versions.
The CI script has been updated to publish the content of master to the edge channel every night. So now nightlies are available there, as well as in
crystal-lang/crystal:nightly docker image, and in the artifacts of the build itself.
Upon tagging and releasing a new version, the package will be available also in the edge channel and will be moved to the stable channel manually. For now, beta and candidate channel are not expected to be used, but the ability to do so in the future is great.
Since the Crystal snap runs in the classic confinement level (more about this later), the installation from the terminal in Ubuntu package is:
Travis needs to deal with a lot to support the different languages and configurations. Having a more unified way to install them, considering also the new upcoming releases, would definitely simplify their human and computer workload.
Travis was eager to start giving support to languages through snaps, and we were eager to reestablish the availability of nightlies.
From now on, when using
crystal: nightly in the
travis.yml file, the snap in the edge channel will be used. For a couple of months the
crystal: latest will still use the current
.deb packages that are hosted in our package repository. Eventually, the snap in the stable channel will be used.
A bonus point is that our package repository will not get traffic from CI builds.
Packaging your Crystal app as a snap
is merged and released (Edit: it was been released in Snapcraft 3.7!), building a snap that will package the application will be easy. If you are unfamiliar with snapcraft don’t miss its introduction.
Assuming there is a
shard.yml that declares targets, dependencies, etc.
snapcraft.yaml file to declare all the parts will look as follows:
After installing the generated snap,
$ crystal-hello will invoke
./bin/hello. Of course you can tweak the names and avoid the
When building the
.snap the following things will happen:
- Installing dependencies via
$ shards install --production.
- Building all targets via
$ shards build --production.
- Packaging all executables in
./binin the final
- Packaging all linked libraries required to run those binaries (detected via
In case the app requires some C libraries that are not available by default, you will need to list them as build-packages.
If you can’t wait, you can grab the PR code and use it as a local plugin.
Some technical details of the crystal snap
While starting to develop the Crystal snap we try to make it work in the strict confinement. The challenge to make it work as strict is how to get libraries and toolchain of host and snap to coexist.
When installing a snap there is no package dependencies that can be declared (like in the
.deb packages). So it would require to have all the libraries included by default in order to have a smooth fresh experience.
As soon as the program to compile becomes more advanced, or a shard library links against a C library, the user will need to deal with some abstraction leaks in packaging related to how those libraries are to be found.
That is the main reason why we stick to the classic confinement model. The caveat is how to let the user know that some native package of the Linux distribution needs to be available as a post-installation step.
The current solution is that the crystal command installed with the snap is actually a wrapper that checks the compiler can be used successfully for simple programs. A message regarding required packages is shown if something goes wrong with that check.
The store works with a concept called channel. For each channel a single version is available. A channel full name is
<track>/<risk>/<branch> where the default track is called latest.
As stated before, the most recent release will be available in the
latest/stable channel, and nightlies in
latest/edge. We have already set up the CI to be able to deliver some
latest/edge/<feature-branch> on demand.
While crystal does not reach
1.0, this schema is enough. You are able to access the latest stable release, and the future version.
There were some discussions regarding how to handle explicit version availability in the store for projects following semver. The general consensus is to create a track for each
Major.minor release, leaving only the latest
Major.minor.patch available in the store. In this schema some projects might leave the
latest track empty to force the user to make an explicit decision. What schema we will follow is still open.
To Canonical, Travis CI, and all the people involved, thanks for an awesome week!