|Title:||autils project plan: An unified standalone repository for utility libraries|
|Author:||Beraldo Leal <email@example.com>, Cleber Rosa <firstname.lastname@example.org>|
|Discussions-To:||email@example.com or PR thread|
|Reviewers:||Plamen Dimitrov <firstname.lastname@example.org>, Xu Han <email@example.com>, Chunfu Wen <firstname.lastname@example.org>, Dan Zheng <email@example.com>|
Table of Contents
This blueprint describes the plan for migrating the utilities libraries present in Avocado and related projects to a standalone and central repository. Such a repository, called autils, is about having well tested utilities that will keep test writers efforts in the features to be tested, and away from debugging and troubleshooting those libraries. While this repository initially targets Avocado users writing tests, it should be generic enough for consumption by third-party developers outside of the Avocado community. It presents a glimpse of the infrastructure that will guarantee quality criteria and will ease the onboarding and maintenance of such libraries.
Besides the plan, with the migration steps for the current libraries, this blueprint also presents the tasks already known to be necessary in order to achieve this goal.
When reading here, in this document, “avocado.utils”, please take into consideration Avocado and Avocado-VT utility libraries.
More than just test utility libraries
Our current libraries are not only useful for writing tests, they are basically system level utilities that could help not only test writers but also day to day tasks. Splitting the repository can increase the adoption by other projects that will choose to use this standalone repo because of its inherent value.
Better reuse of libraries and code between Avocado, Avocado-VT and third party projects
Currently we are duplicating efforts by maintaining common libraries in Avocado and Avocado-VT. By concentrating those libraries in a standalone repository, we will make sure they will be well suited for different use cases. By having more users, the quality and collaboration of those libraries will be improved. Also, third-party software will be able to import those libraries without the need of having the entire Avocado framework as a dependency. This will reduce the learning curve of future potential users that don’t need the entire framework.
Making it clear to our users that Avocado framework is not Python-centric:
Currently, those utility libraries can only be used by Python tests and the Avocado framework was designed to support tests written in multiple languages. By having those libraries outside the core harness codebase we reinforce the language agnostic characteristic of Avocado.
Reduce traffic in the main framework repository and allow for a scalable community on its own standalone project
As evidenced by “avocado-misc-tests”, a more focused standalone project will attract a larger number of similar minded developers and users. By being a simpler standalone repository, with its own specific contributions guidelines, the entry barriers are also lowered.
To avoid the need to change our current imports, our first attempt was to keep the “avocado.utils” namespace regardless of the location of its codebase by using the Implicit Package Namespace (PEP-420) approach. Namespace packages are a mechanism for splitting a single Python package across multiple directories on disk. Unfortunately, to be able to convert the avocado.utils.* modules and packages into a namespace package, we would need to remove the __init__.py files from multiple places. That would break, among other things, the current Avocado egg based deployment model used on spawners such as podman.
Because of that, we decided to have a new namespace for the utility libraries. We ran a poll to choose the name and autils (Avocado Utils) was the winner. So our goal is to host common libraries under avocado-project/autils repository on GitHub and make them also available on PyPI under python-autils package name, since the autils name is unavailable.
The primary goal of autils is to be a central place for modules that ease the interaction with system level features and interfaces. As such, if a user needs to interact with such a system level feature or interface, he/she may propose a new module.
For instance, if a user finds himself/herself constantly interacting with disk partitions, he/she may want to propose a “partitions” module that provides basic APIs for that. In this case it makes sense to have it as part of this central repository.
However, if the utility intends to ease the interaction with an esoteric feature, system or device, this may not be a good place for it.
As a general rule, if a system level feature or interface is generically available to a multitude of users, such as the partition example above, it is a good candidate for this repository. Common sense is much appreciated here and we can always review the proposals during the PR submission or within a Github discussion thread (even before the proposal is ready for submission).
Besides that, the proposed utility must have at least one MAINTAINER that will be responsible for keeping track of Pull Requests, tests and quality requirements as described below:
It’s expected that both new modules and changes to existing modules will need to observe a full set of quality requirements and general standards that will either be described (documentation) and enforced as automated checks as much as possible.
Among the topics related to the quality requirements and standards, the contribution documentation and automated checks will cover, at least:
Logical, stylistic, analytical and formatter checks
- Docstring checks
- PEP-8, PEP-257, etc…
Unit and Functional Tests Coverage
Due to the current code quality and given this is a new repository we would like to reach at least 90% of coverage. This way, users writing tests relying on these libraries will have a guarantee of a good level of correctness and quality, redirecting the debugging and troubleshooting efforts away from the well tested libraries.
The new standalone repository will have predefined pipelines to run those checks. Newly added libraries will automatically be subject to those checks.
Every module will need to provide a description including:
The utility category (ie: network, cpu, security, etc..)
Maintainers information * Main contact (We need a contact point, a single maintainer or a group) * Maintainer Group members (this will be used to block merges)
Supported execution method (Local vs Remote) We intend to flag modules that have the capability of acting on remote systems. This will probably be a broad classification, just for informational purposes because unfortunately we have way too many ways of executing code remotely.
We don’t hope to address a complete standardization of the remote execution APIs just yet, but we want to at least bring some visibility to it.
For instance, the textual category and description will allow users to search for modules based on their interest. Also, our pipelines will parse that data field containing the list of supported platforms, and will trigger tests on those platforms. Users will then be able to see the results of those tests on each of the platforms defined, giving them quality assurance and coverage there.
The exact format of the metadata file is still to be defined, but a possibility is to define a JSON schema that will be provided on this new repository.
As described above, each library must follow the new standard and keep that same level for future versions. Because of that, new PRs must be approved by at least one MAINTAINER and pass the basic checks pipelines that will be provided as part of the autils repository.
Besides the automatic checks, each MAINTAINER is responsible for keeping the consistency of that library with the overall project.
As a new and standalone repository, there will not be any type of release version sync between Avocado and autils. Avocado may release, say, version 100.0, while autils won’t have a release. Because of that, autils will be versioned at a different pace than Avocado.
Every week an automatic job will be executed to determine if there is a potential need for a new release:
If the number of commits since last release has reached a threshold; AND
There is no open discussion with the tag “release-decision”: * Creates a new discussion (with the “release-decision” tag) to run a poll
between all the MAINTAINERS, so they will have the opportunity to thumbs up if there is a need for a new release;
- The link to the new discussion should be sent to all MAINTAINERS;
- If the decision is positive, a round-robin selected maintainer should be responsible for coordinating the release it self;
- And the discussion should be closed
If there is any existing open discussion, the bot job should comment on the same discussion with the updated list of commits and ping again the MAINTAINERS, give them the opportunity to thumbs up based on the new status;
Manual release process could also be started if a MAINTAINER think it is a good timing (i.e: Security bug or critical bug it was fixed).
The current libraries are not going to be removed right away. A transition period is expected for migrating each library into this new standard, and because of that libraries in transition should be available, receive bug fixes for at least one LTS cycle.
Each library in our current code base, should receive a status flag, informing users what their current status is, according on the following table:
The library is using the legacy mode, and it is not yet converted to the autils repository. This should be the status for all avocado.utils.* (including Avocado-VT utils) libraries upon the creation of the autils repository.
A maintainer has picked up this library to be ported to the autils repository. It’s not expected yet that a working version is available on autils, but someone is working on the migration. During this period, the avocado.utils (or Avocado-VT) library should continue to receive both bug fixes and improvements. It’s the maintainer’s responsibility to make sure that the work in progress version of the new autils library includes the same bug fixes and improvements merged during this phase.
This means the current avocado.utils library is deprecated in favor of the migrated version at the autils repository. During at least one Avocado LTS cycle period, the avocado.utils version of the libraries should:
- Include warnings in each deprecated method or function, pointing users to the equivalent method or function in the new autils library
- Keep receiving critical bug fixes backported from the new autils version
At this point, this library has an official maintainer who will be the main point of contact for.
- There are no plans to migrate the library to autils.
Once at least one LTS cycle has passed after the “migrated” status, the legacy library can be permanently removed from its repository of origin.
We could provide a few support material as reference:
- An “onboarding guide” on how to send a new library to this repository
- A template library directory structure that can be used as bootstrap for a
- new library
- Existing libraries in this new repository can also be used as examples