The AiiDAlab in general definition is not a single tool but an infrustructure contains multiple componets. It is run or deployed on an tengible machine. We don't have a version tag for AiiDAlab specificly which cause the problem that it is hard to measure the progress of development.
All base dependencies have their version tag and developed incrementally. Once there are breaking features or important fixes, we shouldn't wait and postpone the release of the new version.
Defined in https://github.com/aiidalab/aiidalab-docker-stack/blob/main/build.json, the full-stack which we used for all production deployment and as the default stack of aiidalab-launch, contain the base dependencies that when new version relesae, the aiidalab-docker-stack
required version bump with the corresponding package version bump. These dependencies are:
aiidalab-home
.aiidalab-home
is the app that not provide the app entry point in the index page but ensential for the tools show in the index page and also the app management widgets.aiida-core
, and use the second latest version defined in aiida-core
. For example, when 3.11
is supported by aiida-core
, we then start to support 3.10
for aiidalab ecosystem.
All version bump of the packages above will eventually lead to the version update of the aiidalab-docker-stack
. (TBD) the version of aiidalab and aiidalab-home is later introduced to the build.json
file. It means if as defined above, we keep the aiidalab-home in the list, we better to have a tag for it as well. This is a good practice in my opinion. Because the backend python version and aiida-core version are more reveal in background which means users have no intuitive feeling how the AiiDAlab is changing. But the aiidalab-home have potential to bring new layout, new tools to the index page which will be highlight for the updates.
The new release of the aiidalab-docker-stack will create new docker stacks to be used in deployments. We now have following core maintained deployments that required to having either the stable deployment for users running production and robust calculation or the edge deployment where we want to have all the edge features. The edge deployment is very important for us to show the progress quickly to the project reviews. It also enssential for developers to have an integrated pack of packges to run and test the edge features.
aiidalab-launch
, need to have latest image supported and backward for running the old images (Now, we support to run aiidalab/aiidalab-docker-stack
images.).Based on time, just like ubuntu version 23-04, 24-10 for next two releases.
The AiiDAlab team can keep its own pace on making the progress and anouncement. However, we have encountered the issue that to support the AiiDA 2.x we delay a lot for having the release annouce since plenty of unexpected issues appread and achitecture of docker stacks cahnges.
Now we introduce the concept of core dependencies, which is designed for apps that if the core dependencies are not meet, the app manager will prevent it from installing. This relieve the issuet that the core developers need to coordinates with app developers to wait the new app support. Meanwhile the empty list of the app is a very good reminder for app developers to priorities there work on support the new AiiDAlab.
At the moment, we only have aiida-core
in the core dependencies. Which means that once the app support the aiida-core version of new AiiDAlab, the app will have the newly supported version show in the app manager.
This can always go in slow but stady pace and make the anouncement in half year base…
The problem is the breaking changes are hard to measure the working load since sometime the changes tangle in between varias core packages.
Things need to be done:
During a lunchtime discussion with Xing, we agreed that we should strategically release QeApp. I acknowledged that I have been responsible for releasing many versions that were not robust, and we concluded that rather than releasing QeApp on a monthly basis, we should aim to release two promising versions every year.
To implement this plan, we will use the calendar versioning semantic that we have used for QeApp thus far. Specifically, we will release versions in April and October of each year, with the version numbers following the format [YEAR].04.xx and [YEAR].10.xx.
To ensure that the upcoming releases are well-planned, we will document new big features on a wiki page, and all bug fixes will go to the patched version of the latest published release. The principle behind this approach is that new features will only go into upcoming releases, while bug fixes will focus on the published release.
This approach offers several advantages. First, we can continue working on new features without breaking the in-use release. Second, it will help trim down the list shown in the app manager so that users can easily see what they have installed (though this will require additional work on the app register filter to allow for the display of the latest patch of releases). Third, the new features will be foreseeable and tracked. Finally, the current release will become more robust and will not break any workflows, allowing reviewers (from MARVEL or Nicola M.) to use it at any time without issue.
However, it is not entirely clear how we will manage the branches to support this approach. We will need to maintain two branches - one for published release support and one for upcoming releases. Although it is normal to find and fix trivial bugs during new feature implementation, separating out such bugs to the released supported branch will increase our workload by requiring another pull request. It will also make it harder to quickly complete new features. Furthermore, when it comes time to merge the branches for the new release, it will be quite challenging.
We do not want to treat QeApp as a large project, as it is ultimately an AiiDAlab app consisting only of GUI wrappers for some workflows. This is why we have opted for calendar versioning to encourage frequent development iterations. If we need to focus on cherry-picking commits and supporting two branches, it could potentially slow down our development pace.