For the design-driven contribution process it is required to specify features upfront in a design doc.
Design Doc Structure
A design doc should discuss the following aspects:
Use-Cases: The interactions between a user and a system to attain particular goals.
Acceptance Criteria: Conditions that must be satisfied to consider the feature as done.
Background: Stuff one needs to know to understand the use-cases (e.g. motivating examples, previous versions and problems, links to related changes/design docs, etc.)
Possible Solutions: Possible solutions with the pros and cons, and explanation of implementation details.
Conclusion: Which decision was made and what were the reasons for it.
As community we want to collaborate on design docs as much as possible and write them together, in an iterative manner. To make this work well design docs are split into multiple files that can be written and refined by several persons in parallel:
index.md: Entry file that links to the files below (also see 'dev-design-doc-index-template.md').
use-cases.md: Describes the use-cases, acceptance criteria and background (also see 'dev-design-doc-use-cases-template.md').
solution-<n>.md: Each possible solution (with the pros and cons, and implementation details) is described in a separate file (also see 'dev-design-doc-solution-template.md').
conclusion.md: Describes the conclusion of the design discussion (also see 'dev-design-doc-conclusion-template.md').
It is expected that:
An agreement on the use-cases is achieved before solutions are being discussed in detail.
Anyone who has ideas for an alternative solution uploads a change with a
solution-<n>.mdthat describes their solution. In case of doubt whether an idea is a refinement of an existing solution or an alternative solution, it’s up to the owner of the discussed solution to decide if the solution should be updated, or if the proposer should start a new alternative solution.
All possible solutions are fairly discussed with their pros and cons, and treated equally until a conclusion is made.
Unrelated issues (judged by the design doc owner) that are identified during discussions may be extracted into new design docs (initially consisting only of an
use-cases.mdfile). Doing so is optional yet can be done by either the design owner or reviewers.
Changes making iterative improvements can be submitted frequently (e.g. additional uses-cases can be added later, solutions can be submitted without describing implementation details, etc.).
After a conclusion has been approved contributors are expected to keep the design doc updated and fill in gaps while they go forward with the implementation.
How to propose a new design?
To propose a new design, upload a change to the
homepage repository that adds a new folder under
which contains at least an
index.md and a
uses-cases.md file (see
design doc structure above).
Pushing a design doc for review requires to be a contributor.
When contributing design docs, contributors should make clear whether they are committed to do the implementation. It is possible to contribute designs without having resources to do the implementation, but in this case the implementation is only done if someone volunteers to do it (which is not guaranteed to happen).
Only very few maintainers actively watch out for uploaded design docs. To raise awareness you may want to send a notification to the repo-discuss mailing list about your uploaded design doc. But the discussion should not take place on the mailing list, comments should be made by reviewing the change in Gerrit.
Design doc review
Code-Review votes on changes that add/modify design docs are
sticky. This means any
Code-Review+2 vote is
preserved when a new patch set is uploaded. If a new patch set makes
significant changes, the uploader of the new patch set must start a new
review round by removing all positive
Code-Review votes from the
Ideas for alternative solutions should be uploaded as a change that describes the solution (see above). This should be done as early as possible during the review process, so that related comment threads stop there and do not clutter the current review. It is up to the alternative reviews to then host their related comments.
Verification should be based on the generated
jekyll site using the
docker, rather than via the rendering in
Changes which make a conclusion on a design (changes that add/change
conclusion.md file, see Design Doc Structure)
should stay open for a minimum of 10 calendar days so that everyone has
a fair chance to see them. It is important that concerns regarding a
feature are raised during this time frame since once a conclusion is
approved and submitted the implementation may start immediately.
Other design doc changes can and should be submitted quickly so that collaboration and iterative refinements work smoothly (see above).
For proposed features the contributor should hear back from the engineering steering committee within 30 calendar days whether the proposed feature is in scope of the project and if it can be accepted.
If the Gerrit review doesn’t start efficiently enough, stalls, gets off-track too much or becomes overly complex, one can use a meeting to refocus it. From that review thread, the organizer can volunteer oneself, or be proposed (even requested) by a reviewer. Community managers may help facilitate that if ultimately necessary.