- Feature Name: north_star
- Start Date: 2016-08-07
- RFC PR: #1728
- Rust Issue: N/A
A refinement of the Rust planning and reporting process, to establish a shared vision of the project among contributors, to make clear the roadmap toward that vision, and to celebrate our achievements.
Rust's roadmap will be established in year-long cycles, where we identify up front - together, as a project - the most critical problems facing the language and its ecosystem, along with the story we want to be able to tell the world about Rust. Work toward solving those problems, our short-term goals, will be decided by the individual teams, as they see fit, and regularly re-triaged. For the purposes of reporting the project roadmap, goals will be assigned to release cycle milestones.
At the end of the year we will deliver a public facing retrospective, describing the goals we achieved and how to use the new features in detail. It will celebrate the year's progress toward our goals, as well as the achievements of the wider community. It will evaluate our performance and anticipate its impact on the coming year.
The primary outcome for these changes to the process are that we will have a consistent way to:
- Decide our project-wide goals through consensus.
- Advertise our goals as a published roadmap.
- Celebrate our achievements with an informative publicity-bomb.
Rust is a massive project and ecosystem, developed by a massive team of mostly-independent contributors. What we've achieved together already is mind-blowing: we've created a uniquely powerful platform that solves problems that the computing world had nearly given up on, and jumpstarted a new era in systems programming. Now that Rust is out in the world, proving itself to be a stable foundation for building the next generation of computing systems, the possibilities open to us are nearly endless.
And that's a big problem.
In the run-up to the release of Rust 1.0 we had a clear, singular goal: get Rust done and deliver it to the world. We established the discrete steps necessary to get there, and although it was a tense period where the entire future of the project was on the line, we were united in a single mission. As The Rust Project Developers we were pumped up, and our user base - along with the wider programming world - were excited to see what we would deliver.
But 1.0 is a unique event, and since then our efforts have become more diffuse even as the scope of our ambitions widen. This shift is inevitable: our success post-1.0 depends on making improvements in increasingly broad and complex ways. The downside, of course, is that a less singular focus can make it much harder to rally our efforts, to communicate a clear story - and ultimately, to ship.
Since 1.0, we've attempted to lay out some major goals, both through the internals forum and the blog. We've done pretty well in actually achieving these goals, and in some cases - particularly MIR - the community has really come together to produce amazing, focused results. But in general, there are several problems with the status quo:
We have not systematically tracked or communicated our progression through the completion of these goals, making it difficult for even the most immersed community members to know where things stand, and making it difficult for anyone to know how or where to get involved. A symptom is that questions like "When is MIR landing?" or "What are the blockers for
?stabilizing" become extremely frequently-asked. We should provide an at-a-glance view what Rust's current strategic priorities are and how they are progressing.
We are overwhelmed by an avalanche of promising ideas, with major RFCs demanding attention (and languishing in the queue for months) while subteams focus on their strategic goals. This state of affairs produces needless friction and loss of momentum. We should agree on and disseminate our priorities, so we can all be pulling in roughly the same direction.
We do not have any single point of release, like 1.0, that gathers together a large body of community work into a single, polished product. Instead, we have a rapid release process, which results in a remarkably stable and reliable product but can paradoxically reduce pressure to ship new features in a timely fashion. We should find a balance, retaining rapid release but establishing some focal point around which to rally the community, polish a product, and establish a clear public narrative.
All told, there's a lot of room to do better in establishing, communicating, and driving the vision for Rust.
This RFC proposes changes to the way The Rust Project plans its work, communicates and monitors its progress, directs contributors to focus on the strategic priorities of the project, and finally, delivers the results of its effort to the world.
The changes proposed here are intended to work with the particular strengths of our project - community development, collaboration, distributed teams, loose management structure, constant change and uncertainty. It should introduce minimal additional burden on Rust team members, who are already heavily overtasked. The proposal does not attempt to solve all problems of project management in Rust, nor to fit the Rust process into any particular project management structure. Let's make a few incremental improvements that will have the greatest impact, and that we can accomplish without disruptive changes to the way we work today.
Rust's roadmap will be established in year-long cycles, where we identify up front the most critical problems facing the project, formulated as problem statements. Work toward solving those problems, goals, will be planned as part of the release cycles by individual teams. For the purposes of reporting the project roadmap, goals will be assigned to release cycle milestones, which represent the primary work performed each release cycle. Along the way, teams will be expected to maintain tracking issues that communicate progress toward the project's goals.
At the end of the year we will deliver a public facing retrospective, which is intended as a 'rallying point'. Its primary purposes are to create anticipation of a major event in the Rust world, to motivate (rally) contributors behind the goals we've established to get there, and generate a big PR-bomb where we can brag to the world about what we've done. It can be thought of as a 'state of the union'. This is where we tell Rust's story, describe the new best practices enabled by the new features we've delivered, celebrate those contributors who helped achieve our goals, honestly evaluate our performance, and look forward to the year to come.
Key terminology used in this RFC:
problem statement - A description of a major issue facing Rust, possibly spanning multiple teams and disciplines. We decide these together, every year, so that everybody understands the direction the project is taking. These are used as the broad basis for decision making throughout the year, and are captured in the yearly "north star RFC", and tagged
R-problem-statementon the issue tracker.
goal - These are set by individual teams quarterly, in service of solving the problems identified by the project. They have estimated deadlines, and those that result in stable features have estimated release numbers. Goals may be subdivided into further discrete tasks on the issue tracker. They are tagged
retrospective - At the end of the year we deliver a retrospective report. It presents the result of work toward each of our goals in a way that serves to reinforce the year's narrative. These are written for public consumption, showing off new features, surfacing interesting technical details, and celebrating those who contribute to achieving the project's goals and resolving it's problems.
release cycle milestone - All goals have estimates for completion, placed on milestones that correspond to the 6 week release cycle. These milestones are timed to corrspond to a release cycle, but don't represent a specific release. That is, work toward the current nightly, the current beta, or even that doesn't directly impact a specific release, all goes into the release cycle milestone corresponding to the time period in which the work is completed.
The full planning cycle spans one year. At the beginning of the cycle we identify areas of Rust that need the most improvement, and at the end of the cycle is a 'rallying point' where we deliver to the world the results of our efforts. We choose year-long cycles because a year is enough time to accomplish relatively large goals; and because having the rallying point occur at the same time every year makes it easy to know when to anticipate big news from the project. Being calendar-based avoids the temptation to slip or produce feature-based releases, instead providing a fixed point of accountability for shipping.
This planning effort is problem-oriented. Focusing on "how" may seem like an obvious thing to do, but in practice it's very easy to become enamored of particular technical ideas and lose sight of the larger context. By codifying a top-level focus on motivation, we ensure we are focusing on the right problems and keeping an open mind on how to solve them. Consensus on the problem space then frames the debate on solutions, helping to avoid surprises and hurt feelings, and establishing a strong causal record for explaining decisions in the future.
At the beginning of the cycle we spend no more than one month deciding on a small set of problem statements for the project, for the year. The number needs to be small enough to present to the community managably, while also sufficiently motivating the primary work of all the teams for the year. 8-10 is a reasonable guideline. This planning takes place via the RFC process and is open to the entire community. The result of the process is the yearly 'north star RFC'.
The problem statements established here determine the strategic direction of the project. They identify critical areas where the project is lacking and represent a public commitment to fixing them. They should be informed in part by inputs like the survey and production user outreach, as well as an open discussion process. And while the end-product is problem-focused, the discussion is likely to touch on possible solutions as well. We shouldn't blindly commit to solving a problem without some sense for the plausibility of a solution in terms of both design and resources.
Problem statements consist of a single sentence summarizing the problem, and one or more paragraphs describing it (and its importance!) in detail. Examples of good problem statements might be:
- The Rust compiler is too slow for a tight edit-compile-test cycle
- Rust lacks world-class IDE support
- The Rust story for asynchronous I/O is very primitive
- Rust compiler errors are difficult to understand
- Rust plugins have no clear path to stabilization
- Rust doesn't integrate well with garbage collectors
- Rust's trait system doesn't fully support zero-cost abstractions
- The Rust community is insufficiently diverse
- Rust needs more training materials
- Rust's CI infrastructure is unstable
- It's too hard to obtain Rust for the platforms people want to target
During the actual process each of these would be accompanied by a paragraph or more of justification.
We strictly limit the planning phase to one month in order to keep the discussion focused and to avoid unrestrained bikeshedding. The activities specified here are not the focus of the project and we need to get through them efficiently and get on with the actual work.
The core team is responsible for initiating the process, either on the internals forum or directly on the RFC repository, and the core team is responsible for merging the final RFC, thus it will be their responsibility to ensure that the discussion drives to a reasonable conclusion in time for the deadline.
Once the year's problem statements are decided, a metabug is created for each on
the rust-lang/rust issue tracker and tagged
R-problem-statement. In the OP of
each metabug the teams are responsible for maintaining a list of their goals,
linking to tracking issues.
Like other RFCs, the north star RFC is not immutable, and if new motivations arise during the year, it may be amended, even to the extent of adding additional problem statements; though it is not appropriate for the project to continually rehash the RFC.
During the regular 6-week release cycles is where the solutions take shape and are carried out. Each cycle teams are expected to set concrete goals that work toward solving the project's stated problems; and to review and revise their previous goals. The exact forum and mechanism for doing this evaluation and goal-setting is left to the individual teams, and to future experimentation, but the end result is that each release cycle each team will document their goals and progress in a standard format.
A goal describes a task that contributes to solving the year's problems. It may or may not involve a concrete deliverable, and it may be in turn subdivided into further goals. Not all the work items done by teams in a quarter should be considered a goal. Goals only need to be granular enough to demonstrate consistent progress toward solving the project's problems. Work that contributes toward quarterly goals should still be tracked as sub-tasks of those goals, but only needs to be filed on the issue tracker and not reported directly as goals on the roadmap.
For each goal the teams will create an issue on the issue tracker tagged with
R-goal. Each goal must be described in a single sentence summary with an
end-result or deliverable that is as crisply stated as possible. Goals with
sub-goals and sub-tasks must list them in the OP in a standard format.
During each cycle all
R-unstable issues assigned to each team
must be triaged and updated for the following information:
- The set of sub-goals and sub-tasks and their status
- The release cycle milestone
Goals that will be likely completed in this cycle or the next should be assigned to the appropriate milestone. Some goals may be expected to be completed in the distant future, and these do not need to be assigned a milestone.
The release cycle milestone corresponds to a six week period of time and contains the work done during that time. It does not correspend to a specific release, nor do the goals assigned to it need to result in a stable feature landing in any specific release.
Release cycle milestones serve multiple purposes, not just tracking of the goals
defined in this RFC:
R-goal tracking, tracking of stabilization of
R-RFC-approved features, tracking of critical bug fixes.
Though the release cycle milestones are time-oriented and are not strictly tied
to a single upcoming release, from the set of assigned
R-unstable issues one
can derive the new features landing in upcoming releases.
During the last week of every release cycle each team will write a brief report summarizing their goal progress for the cycle. Some project member will compile all the team reports and post them to internals.rust-lang.org. In addition to providing visibility into progress, these will be sources to draw from for the subsequent release announcements.
The retrospective is an opportunity to showcase the best of Rust and its community to the world.
It is a report covering all the Rust activity of the past year. It is written for a broad audience: contributors, users and non-users alike. It reviews each of the problems we tackled this year and the goals we achieved toward solving them, and it highlights important work in the broader community and ecosystem. For both these things the retrospective provides technical detail, as though it were primary documentation; this is where we show our best side to the world. It explains new features in depth, with clear prose and plentiful examples, and it connects them all thematically, as a demonstration of how to write cutting-edge Rust code.
While we are always lavish with our praise of contributors, the retrospective is the best opportunity to celebrate specific individuals and their contributions toward the strategic interests of the project, as defined way back at the beginning of the year.
Finally, the retrospective is an opportunity to evaluate our performance. Did we make progress toward solving the problems we set out to solve? Did we outright solve any of them? Where did we fail to meet our goals and how might we do better next year?
Since the retrospective must be a high-quality document, and cover a lot of material, it is expected to require significant planning, editing and revision. The details of how this will work are to be determined.
As a result of this process the Rust roadmap for the year is encoded in three main ways, that evolve over the year:
- The north-star RFC, which contains the problem statements collected in one place
- The R-problem-statement issues, which contain the individual problem statements, each linking to supporting goals
- The R-goal issues, which contain a hierarchy of work items, tagged with metadata indicating their statuses.
Alone, these provide the raw data for a roadmap. A user could run a
GitHub query for all
R-problem-statement issues, and by digging through them
get a reasonably accurate picture of the roadmap.
However, for the process to be a success, we need to present the roadmap in a way that is prominent, succinct, and layered with progressive detail. There is a lot of opportunity for design here; an early prototype of one possible view is available here.
Again, the details are to be determined.
The timing of the events specified by this RFC is precisely specified in order to set clear expectations and accountability, and to avoid process slippage. The activities specified here are not the focus of the project and we need to get through them efficiently and get on with the actual work.
The north star RFC development happens during the month of September, starting September 1 and ending by October 1. This means that an RFC must be ready for FCP by the last week of September. We choose September for two reasons: it is the final month of a calendar quarter, allowing the beginning of the years work to commence at the beginning of calendar Q4; we choose Q4 because it is the traditional conference season and allows us opportunities to talk publicly about both our previous years progress as well as next years ambitions. By contrast, starting with Q1 of the calendar year is problematic due to the holiday season.
Following from the September planning month, the quarterly planning cycles take place for exactly one week at the beginning of the calendar quarter; likewise, the planning for each subsequent quarter at the beginning of the calendar quarter; and the development of the yearly retrospective approximately for the month of August.
The survey and other forms of outreach and data gathering should be timed to fit well into the overall calendar.
- [Refining RFCs part 1: Roadmap] (https://internals.rust-lang.org/t/refining-rfcs-part-1-roadmap/3656), the internals.rust-lang.org thread that spawned this RFC.
- [Post-1.0 priorities thread on internals.rust-lang.org] (https://internals.rust-lang.org/t/priorities-after-1-0/1901).
- [Post-1.0 blog post on project direction] (https://blog.rust-lang.org/2015/08/14/Next-year.html).
- [Blog post on MIR] (https://blog.rust-lang.org/2016/04/19/MIR.html), a large success in strategic community collaboration.
- ["Stability without stagnation"] (http://blog.rust-lang.org/2014/10/30/Stability.html), outlining Rust's philosophy on rapid iteration while maintaining strong stability guarantees.
- [The 2016 state of Rust survey] (https://blog.rust-lang.org/2016/06/30/State-of-Rust-Survey-2016.html), which indicates promising directions for future work.
- [Production user outreach thread on internals.rust-lang.org] (https://internals.rust-lang.org/t/production-user-research-summary/2530), another strong indicator of Rust's needs.
- [rust-z] (https://brson.github.io/rust-z), a prototype tool to organize the roadmap.
The yearly north star RFC could be an unpleasant bikeshed, because it simultaneously raises the stakes of discussion while moving away from concrete proposals. That said, the problem orientation should help facilitate discussion, and in any case it's vital to be explicit about our values and prioritization.
While part of the aim of this proposal is to increase the effectiveness of our team, it also imposes some amount of additional work on everyone. Hopefully the benefits will outweigh the costs.
The end-of-year retrospective will require significant effort. It's not clear who will be motivated to do it, and at the level of quality it demands. This is the piece of the proposal that will probably need the most follow-up work.
Instead of imposing further process structure on teams we might attempt to derive a roadmap solely from the data they are currently producing.
To serve the purposes of a 'rallying point', a high-profile deliverable, we might release a software product instead of the retrospective. A larger-scope product than the existing rustc+cargo pair could accomplish this, i.e. The Rust Platform idea.
Another rallying point could be a long-term support release.
Are 1 year cycles long enough?
Are 1 year cycles too long? What happens if important problems come up mid-cycle?
Does the yearly report serve the purpose of building anticipation, motivation, and creating a compelling PR-bomb?
Is a consistent time-frame for the big cycle really the right thing? One of the problems we have right now is that our release cycles are so predictable they are almost boring. It could be more exciting to not know exactly when the cycle is going to end, to experience the tension of struggling to cross the finish line.
How can we account for work that is not part of the planning process described here?
How do we address problems that are outside the scope of the standard library and compiler itself? (See The Rust Platform for an alternative aimed at this goal.)
How do we motivate the improvement of rust-lang crates and other libraries? Are they part of the planning process? The retrospective?
'Problem statement' is not inspiring terminology. We don't want to our roadmap to be front-loaded with 'problems'. Likewise, 'goal' and 'retrospective' could be more colorful.
Can we call the yearly RFC the 'north star RFC'? Too many concepts?
What about tracking work that is not part of R-problem-statement and R-goal? I originally wanted to track all features in a roadmap, but this does not account for anything that has not been explicitly identified as supporting the roadmap. As formulated this proposal does not provide an easy way to find the status of arbitrary features in the RFC pipeline.
How do we present the roadmap? Communicating what the project is working on and toward is one of the primary goals of this RFC and the solution it proposes is minimal - read the R-problem-statement issues.