[RFC 0029] Backports team

rfcs/0029-backports-team.md

@@ -0,0 +1,256 @@
+---
+feature: backports_team
+start-date: 2018-07-28
+author: samueldr
+co-authors: ...
+related-issues: ...
+---
+
+Summary
+=======
+[summary]: #summary
+
+This intends to formalize the backports process and formalize the formation of
+a team managing backports.
+
+
+Motivation
+==========
+[motivation]: #motivation
+
+The way it is currently handled makes it so too often fixes for lesser-used
+software are not backported to the *current-stable* revision of NixOS.
+Furthermore, even some more-often used software sometimes lag behind since the
+updates were simply not cherry-picked. While there are [general
+guidelines][grahamc_backport] written up, no canonical process is documented.
+
+Backports is not only a security feature, but also a user-experience feature.
+Upstream software is getting updated; it would be best if current-stable does
+not get the reputation of being stale.
+
+By formalizing the process, it is expected to create a good hygiene that will
+keep NixOS current-stable fresher, instead of presenting stale software to the
+end-users.
+
+Finally, by appointing a team, even if fluid, it is expected that the
+responsibilities given to the users will _nudge them_ toward doing the work
+required of maintaining the backports. The work of integrating backports should
+be the release manager's, as per [RFC 0015][rfc0015]. Through [dubious
+stats][stats_release], it seems that for the 18.03 release their work has
+been minimal, and instead filled-in by ad-hoc work by other members.
+
+
+Detailed design
+===============
+[design]: #detailed-design
+
+Backports process
+-----------------
+
+Expected to be backported to current-stable are the following:
+<sup>[1][grahamc_backport]</sup>
+
+ * Patch-level and minor-level updates.
+ * Security patches from downstream projects (other distributions) when
+   no patch-level or minor-level update is made available.
+ * Any update when the current stable version is completely broken.
+ * Extremely security-sensitive software, in particular web browsers,
+   mail user-agents and the kernel, can and should be updated to the version
+   expected by upstream to be secure.
+
+### Examples
+
+An example of broken software needing a major update to be backported is
+Spotify, which regularly breaks with old versions.
+
+Don't backport if the patch is just for Darwin, they use nixpkgs-unstable not a
+stable branch.
+
+> `FIXME` Is this true?? I see nixpkgs-18.03-darwin in the channels list.
+
+Backports team
+--------------
+
+The idea behind the team is more to rally people behind a common mindset and
+banner than forcibly organizing and assigning roles to people. There is also
+the advantage of creating an explicit communication channel reducing the risks
+of overlaps in the backport efforts. Since the members of the team are working
+towards the same goal, better coordination with tooling and documentation can
+hopefully happen.
+
+Being on the backports team is not expected to be too much of an investment,
+as long as there are multiple members on the team. Members will not be expected
+to do anything particular, only to sometimes help out in the duties. At first,
+no rotated duties are expected to be given.
+
+While reading the following sections, keep in mind that there is expectation
+that there will be tooling developer along the way to automate and expedite
+part of the process of backporting.
+
+### Duties
+
+The duties of the backport team members are as follow
+
+ * Identify pull requests and commits subject to backport.
+ * Prepare contained change sets with backports.
+
+Identifying pull requests subject to backport is expected to be done through
+figuring out, from the set of recently closed pull requests, which ones are
+to be backported. Nothing more, preparing the change sets is a separate task.
+
+Identifying the commits subject to backport is a similar task, but may not be
+done depending on the difficulty. It is expected that changes and fixes needing
+to be backported will be applied through PRs.
+
+Preparing the change sets will generally consist of taking the commit(s) and
+using `git cherry-pick -x` on the current-stable branch. Then, creating a PR
+with a generally standardized layout.
+
+As for "*who merges?*", this is something that may be handed by any NixOS
+member having the commit bit, in the backports team or not. It is not expected
+to be in the duties of the backports team until there is a critical mass of
+members, especially members having commit access. It is hoped that the
+standardized work from the backports team will make merging their pull requests
+an easy task.
+
+
+Backports tooling
+-----------------
+
+While having a team working together is great, collaborating still is an issue
+without both conventions and tooling. Initially the author's (@samueldr) duty,
+figuring out both a workflow and tools to help the team collaborate efficiently
+is required. 
+
+The processes and tooling is expected to:
+
+ * Distribute the effort as much as possible while reducing overlap.
+ * Minimize the size of individual efforts.
+ * Reduce the amount of busywork.
+ * Standardize to expedite shipping and reviewing backports.
+
+### Collaboration
+
+> (This is a high-level non-bikesheddable overview of the functionalities)
+
+Through a networked-service, with high probability of it being a web-based
+service, team members will be able to file merged PRs as either on-topic of
+off-topic for the backports team. The system is expected to fetch the newly
+merged PRs from github and provide them for the "figuring out" task previously
+described. The PRs would be presented from oldest to newest.
+
+Once filed, the collaboration system will allow members to assign themselves
+to do the backports. It is expected that team members will go from oldest to
+newest, but not enforced. Working on implementing a backport would first have
+the team member "take" the task, allowing parallel work to not conflict.
+
+A task will be worked on by making use of a command line tooling, allowing
+most tediousness to be automated away.
+
+### Making the backport
+
+As previously said, the team member will use a command line tooling.
+
+The team member will be presented with the commands in the interface, allowing
+less thoughts to be given to the task.
+
+When using the tooling, pretty much all tasks will be as automated as possible.
+
+ * Fetching
+ * (Checking out current-stable)
+ * Creating the branch
+ * Cherry-picking
+     * If it fails to auto-apply, this is where work starts; applying.
+     * After applying, resuming automation will be possible.
+ * Build of affected attributes (*may be limited somewhat*)
+
+One part that is **not** automated is testing. It is expected that the team
+member will at least try to launch affected software `result/bin/*`. Only a
+general glance would be required.
+
+For patches affecting libraries, affected attributes will be listed, allowing
+some general testing to be done.
+
+Once some general testing has been done and conclusive, the team member will
+use the tooling to automate pushing and creating the PR.
+
+ * Push to team member's repository.
+ * Create the PR targeting current-stable.
+     * The text will be auto-filled with data according to the build.
+     * Some machine-usable data may be provided for the tooling.
+ * Automatically mark the task as done and add with the newly opened PR number to it.
+
+The standardized PRs created by the tooling are expected to be easier to be
+approve by members of NixOS. The reduced number of manual steps is expected
+to make it easy for non-maintainers, even only enthusiasts to take part in
+the effort.
+
+
+Drawbacks
+=========
+[drawbacks]: #drawbacks
+
+Formalizing a process always reeks of red tape. This may turn off some
+volunteers away from doing backports.
+
+It may well happen that once the team formed, nothing changes and the backports
+are not maintained any more than they are.
+
+Writing and maintaining tooling may slow down or halt progress toward actually
+maintaining backports.
+
+
+Alternatives
+============
+[alternatives]: #alternatives
+
+Continuing as-it-is, with ad-hoc updates, sometimes missed. This has proved not
+to be the most successful way to manage backports.
+
+It could also be possible to only implement parts of the RFC. Either the team
+or the tooling. Both are of equal value and generally independent. They would,
+though, work best if working together.
+
+
+Unresolved questions
+====================
+[unresolved]: #unresolved-questions
+
+ * Actual team organization (if any).
+ * Specifying processes for all software updates and fixes.
+ * Keeping patch-releases up-to-date when a major is out at upstream.
+ * How to work with non-PR commits with fixes or updates needing backport.
+ * Conflict with non-team members' own backports.
+
+
+Future work
+===========
+[future]: #future-work
+
+ * Specifying collaboration with automated updates (@r-ryantm)
+
+
+Definitions
+===========
+[definitions]: #definitions
+
+### Backport
+
+Taking fixes and features from a newer version and applying them to an older
+maintained version to provide the equivalent fix or feature.<sup>
+[1][1]
+[2][2]
+[3][3]
+</sup>
+
+### Current-stable
+
+A version-agnostic way to specify the current stable branch of NixOS.
+
+
+[1]: https://en.wikipedia.org/wiki/Backporting
+[2]: https://en.wiktionary.org/wiki/backport
+[3]: https://access.redhat.com/security/updates/backporting
+[rfc0015]: https://github.com/NixOS/rfcs/blob/master/rfcs/0015-release-manager.md
+[stats_release]: https://gist.github.com/samueldr/7ec402f71d3bb2ac2e059f33d29d95bb
+[grahamc_backport]: https://gist.github.com/grahamc/c60578c6e6928043d29a427361634df6#what-to-backport