|Title:||Blueprint specification and requirements|
|Author:||Beraldo Leal <email@example.com>|
|Reviewers:||Cleber Rosa <firstname.lastname@example.org>, Jan Richter <email@example.com>, Plamen Dimitrov <firstname.lastname@example.org>, Willian Rampazzo <email@example.com>|
Table of Contents
Having better records of architectural decisions in our repository is a good way to socialize our technical decisions, improve the code review process, avoid unnecessary code, and balance the workload among other developers. We are already using the “blueprint” documents for a while and still learning how to write them. Although we have a basic idea of what is a blueprint, some may not understand the concept because we are missing one meta-document that describes the blueprint’s basic notions. This document is a blueprint to specify how we should write blueprints from now on.
Depending on the project size, having very well defined and structured documents about the architecture decisions seems like an overkill, but could help projects of any size, including Avocado, save time, make better decisions and improve the way we socialize those decisions.
Today in the Avocado project we have the good practice to submit an RFC to our mailing list or to use/open a GitHub issue when we have a new idea. RFCs are a widespread way to disclose the architecture decisions, but they are just one part of a longer process. During the RFC phase, we argue in favor of a proposal, and we are mostly concerned about collecting feedback. After this phase, we could go one step forward and consolidate the discussion in Blueprints (sometimes called ADRs - Architecture Decision Records). This could be the next step so we could better socialize our decisions for future readers. A very well defined and structured document has some advantages over an RFC, but it is not intended to replace it, just be a later stage to follow from it.
With blueprints, we could not only, but mainly:
- Create better documents for future members and future reference, when we are trying to answer the following questions:
- why are you doing this? (the “problem” or the “motivation”);
- what are you proposing to solve the problem? (your “solution”)?
- And how are you going to implement the proposed solution?
Depending on the type of your blueprint, the answer for the last question (c) could be written in pseudocode, general text or might even not be necessary (although desired) — more details on the section named Specification.
When using RFCs as email threads, there are no sections or headers, each contributor will try to send the RFC without following formal sections and headers. RFCs, as we use them today are just thread discussions and are not focused on future review/reading.
- Make sure that another peer will be able to implement a feature/requirement.
Blueprints are not for you; they are for the community. If you know that a problem exists and know how to fix it, the most natural course of action would be to start coding this fix and submitting a Pull Request. While this is still valid for most of the cases, some important architectural changes must be discussed first to explain the “why”, “what” and “how” to keep everyone on the same page, avoid unnecessary coding, and most importantly: allow others to implement it in case you are not available.
- Improve the code review quality and time.
Having a better understanding of the problem and the big picture is better for code review. It is harder to capture that from the Pull Request (PR) description and PR changes. Developers that are aware of the problem tend to review your changes with the problem in mind and hence more quickly.
- Reduce onboarding time for new members.
Having a history of how we made an architectural decision and why we are implementing it this way will give new members reading material to understand our project, avoiding unnecessary discussions and meetings to explain something.
- Create a common standard that will make it easier for readers.
With an open RFC, authors tend to organize the ideas in different ways with different sections. Having a very well structured document with common sections will make it easier for readers to understand the problem and the solution.
- Track the progress of a significant implementation.
We could use the blueprints header “status” line to track the progress of some features. We could even have a page parsing and listing all the blueprints with the title, author, status, and target milestone for that feature.
- Find the middle ground between “overthinking” and “auto-pilot.”
Last but not least: we are not trying to overthink here and/or slow down our development processes. The idea is to have a lightweight document, very objective that will save us time in the medium and long run. We don’t have to overthink by trying to handle any possible scenario outside of ones we actually have a use case for. But we should also avoid the “auto-pilot” mode in our contributions where we are fixing issues as quick as possible without thinking about the big picture, it is not healthy for the project.
One blueprint per topic¶
Try to follow the minimalist approach and be concise with content relevant to one particular topic. If you have a more general topic to discuss, you should set the type as “Epic Blueprint” (more below) but still try to be concise and focused on the subject.
File format and location¶
Our current documentation already uses ReStructuredText (.rst) format, so we will adopt .rst format here too. All blueprints will be located inside docs/source/blueprints with the filename BPXXX.rst, where XXX is the number of the blueprint. Just pick the next number available for your blueprint.
It’s recommended that you use
as a starting point.
Write for your audience¶
As mentioned before, your blueprint will be read by your peers, future members, and future yourself. Keep in mind that your audience is developers with a minimal understanding of the Avocado internals and be kind providing any necessary context to understand the problem.
Currently, we have the following blueprint types:
- Architectural Blueprint: Any blueprint changing or introducing a new core feature or architectural change to Avocado.
- Process Blueprint: Any blueprint that is not implementing a new core feature, but changing how the project works. This could be, for instance, related to the repositories or processes.
- Meta Blueprint: A blueprint about blueprints. Like this one and any future blueprint that changes our blueprint’s styles and methods.
- Epic Blueprint: A blueprint that is touching on multiple areas and is too big to have all the documentation in one single blueprint. We could split epic blueprints into smaller blueprints or issues (if they are small and easy to understand). Epic Blueprints are not a merge of all sub-blueprints. Like an epic issue, epic blueprints don’t need to detail “how” (or provide details) that the sub-blueprints could have.
- Component Blueprint: A blueprint with the intent to describe a new utility module or a new plugin.
Python PEPs (Python Enhancement Proposals) uses RFC822 for describing the headers. This could be useful here too, especially when parsing those headers to display our list of blueprints with the current status.
The current list of items of our blueprint headers is below:
- Number: Usually, the blueprint number in the format BPXXX
- Title: A short descriptive title, limited to 80 characters
- Author: The author or authors of blueprint. Following the format: [FIRST NAME] [LAST NAME] - <email@domain>
- Reviewers: All reviewers that approved and helped during the review process
- Created: Date string when the blueprint first draft was submitted. Please use the following format: DD-MMM-YYYY.
- Type: One of the types described during the previous section
- Status: One of the types described during the next section
Here is an example of a header:
:Number: BP001 :Title: Configuration by convention :Author: Beraldo Leal <firstname.lastname@example.org> :Reviewers: Cleber Rosa, Lukáš Doktor and Plamen Dimitrov :Created: 06-Dec-2019 :Type: Epic Blueprint :Status: WIP
- Draft: All blueprints should be created in this state. This means the blueprint is accepting comments, and probably there is a discussion happening. Blueprints in draft mode can be part of our repository.
- Approved: Blueprint was approved after discussions, and all suggestions are already incorporated on the document. Nobody has started working on this yet.
- Assigned: This status is not about the blueprint itself, but about the proposal that is the subject of the BP. This means that the blueprint was approved, and someone is already working on implementing it. A BP status can change from Draft to Assigned if the work has started already.
- WIP: Blueprint was approved and someone is working on it. Work in Progress.
- Implemented: This means the BP is already implemented and delivered to the Avocado’s master branch.
- Rejected: Rejected status means the idea was not implemented because it wasn’t approved by everyone or has some technical limitations.
- Deprecated: Deprecated means it was approved, implemented, and at some point, makes no more sense to have it. For example, anything related to the legacy runner. Usually, Deprecated means that it was replaced by something else.
As you can see, there is no status to accommodate any future change in a blueprint. Blueprints should not be “voided.” Any improvement on an old blueprint should be presented as a new blueprint, changing the status of the original to “deprecated”.
In order to facilitate the reading and understanding of the problem, all blueprints must have the following sections:
- Backwards Compatibility
- Security Implications
- How to Teach This
- Related Issues
Below you can find a brief description of what you should write in each section:
TL;DR: Should be a short description of your blueprint. Like an abstract. We recommend writing this at the end of your first draft. This will give you a better overview of it.
Motivation: This should be the motivation of your proposed solution, not the motivation of the blueprint itself. It describes the problem. Here, you should answer “why” your solution is needed.
Specification: In this section, you should describe how you are going to solve the problem. You can create subsections here to organize your ideas better. Please keep in mind that it is useful to mention the details, with code snippets, examples, and/or references. This will save you time, making sure that everyone is in agreement with the proposed solution.
Backwards Compatibility: How is your proposal going to affect older versions of Avocado? Should we deprecate some modules, classes, or methods? Are we going to keep backwards compatibility or not?
Security Implications: Do you have any concerns about security with your proposed solution and what are they? If there’s functionality that is insecure but highly convenient, consider how to make it “opt-in”, disabled by default.
How to Teach This: What is the best way to inform our devs and users about your new feature/solution? Consider both “how-to” and reference style documentation, and if appropriate, examples (under
examples/) using the feature.
Related Issues: Here, you should mention Github links for both: a) current open issues that are blocking while waiting for your BP and b) all open issues that will render this BP as “implemented” when closed.
Issues to address this BP
Would be nice, if possible, to open issues on GH that covers all aspects of your Blueprint.
Issues this BP will solve
What are the issues already existent on Avocado project that your proposal will solve?
References: Any external reference for helping understand the problem and your solution.
So far, we are on our 3rth blueprint (BP003 was the last one). This BP000 should have been released before those blueprints. So probably those three blueprints are not 100% compliaant with this meta blueprint, and that is fine. We were learning on the fly. We don’t need to change any of those blueprints after BP000 gets approved.
No security implications found so far.
How to Teach This¶
Getting used to writing blueprints is not an easy task. And probably we are going to find unplanned issues with this process on the way. The general rule of thumb is to use common sense. To make this more public, we could consider the following:
- If approved, BP000 should be on top of our blueprints lists for reference.
- We could also have a template inside the blueprints directory to help people when submitting their own blueprints.
- Also, we could include pointers and instructions in our development guide for this BP.
- Another good practice would be to make comments in Avocado’s source code with some pointers to specific blueprints.