AiC: Shepherds 3.0

11 September 2019

I would like to describe an idea that’s been kicking around in my head. I’m calling this idea “shepherds 3.0” – the 3.0 is to distinguish it from the other places we’ve used the term in the past. This proposal actually supplants both of the previous uses of the term, replacing them with what I believe to be a preferred alternative (more on that later).


This is an idea that has been kicking around in my head for a while. It is not a polished plan and certainly not an accepted one. I’ve not talked it over with the rest of the lang team, for example. However, I wanted to put it out there for discussion, and I do think we should be taking some step in this direction soon-ish.


What I’m proposing, at its heart, is very simple. I want to better document the “agenda” of the lang-team. Specifically, if we are going to be moving a feature forward1, then it should have a shepherd (or multiple) who is in charge of doing that.

In order to avoid unbounded queues, the number of things that any individual can shepherd should be limited. Ideally, each person should only shepherd one thing at a time, though I don’t think we need to make a firm rule about it.

Becoming a shepherd is a commitment on the part of the shepherd. The first part of the lang team meeting should be to review the items that are being actively shepherded and get any updates. If we haven’t seen any movement in a while, we should consider changing the shepherd, or officially acknowleding that something is stalled and removing the shepherd altogether.

Assigning a shepherd is a commitment on the part of the rest of the lang-team as well. Before assigning a shepherd, we should discuss if this agenda item is a priority. In particular, if someone is shepherding something, that means we all agree to help that item move towards some kind of completion. This means giving feedback, when feedback is requested. It means doing the work to resolve concerns and conflicts. And, sometimes, it will mean giving way. I’ll talk more about this in a bit.

What was shepherds 1.0 and how is this different?

The initial use of the term shepherd, as I remember it, was actually quite close to the way I am using it here. The idea was that we would assign RFCs to a shepherd that should either drive to be accepted or to be closed. This policy was, by and large, a failure – RFCs got assigned, but people didn’t put in the time. (To be clear, sometimes they did, and in those cases the system worked reasonably well.)

My proposal here differs in a few key respects that I hope will make it more successful:

  • We limit how many things you can shepherd at once.
  • Assigning a shepherd is also a commitment from the lang team as a whole to review progress, resolve conflicts, and devote some time to the issue.
  • We don’t try to shepherd everything – in contrast, shepherding marks the things we are moving forward.
  • The shepherd is not something specific to an RFC, it refers to all kinds of “larger decisions”. For example, stabilization would be a shepherd activity as well.

What was shepherds 2.0 and how is this different?

We’ve also used the term shepherd to refer to a role that is moving towards full lang team membership. That’s different from this proposal in that it is not tied to a specific topic area. But there is also some interaction – for example, it’s not clear that shepherds need to be active lang team members.

I think it’d be great to allow shepherds to be any person who is sufficiently committed to help see something through. The main requirement for a shepherd should be that they are able to give us regular updates on the progress. Ideally, this would be done by attending the lang team meeting. But that doesn’t work for everyone – whether it because of time zones, scheduling, or language barriers – and so I think that any form of regular, asynchronous report would work jsut fine.

I think I would prefer for this proposal – and this kind of “role-specific shepherding” – to entirely replace the “provisional member” role on the lang team. It seems strictly better to me. Among other things, it’s naturally time-limited. Once the work item completes, that gives us a chance to decide whether it makes sense for someone to become a full member of the lang team, or perhaps try shepherding another idea, or perhaps just part ways. I expect there are a lot of people who have interest in working through a specific feature but for whom there is little desire to be long-term members of the lang team.

How do I get a shepherd assigned to my work item?

Ultimately, I think this question is ill-posed: there is no way to “get” a shepherd assigned to your work. Having the expectation that a shepherd will be assigned runs smack into the problems of unbounded queues and was, I think, a crucial flaw in the Shepherds 1.0 system.

Basically, the way a shepherd gets assigned in this scheme is roughly the same as the way things “get done” today. You convince someone in the lang team that the item is a priority, and they become the shepherd. That convincing takes place through the existing channels: nominated issues, discord or zulip, etc. It’s not that I don’t think this is something else we should be thinking about, it’s just that it’s something of an orthogonal problem.

My model is that shepherds are how we quantify and manage the things we are doing. The question of “what happens to all the existing things” is more a question of how we select which things to do – and that’s ultimately a priority call.

OK, so, what happens to all the existing things?

That’s a very good question. And one I don’t intend to answer here, at least not in full. That said, I do think this is an important problem that we should think about. I would like to be exposing more “insight” into our overall priorities.

In my ideal world, we’d have a list of projects that we are not working on, grouped somewhat by how likely we are to work on them in the future. This might then indicate ideas that we do not want to pursue; ideas that we have mild interest in but which have a lot of unknowns. Ideas that we started working on but got blocked at some point (hopefully with a report of what’s blocking them). And so forth. But that’s all a topic for another post.

One other idea that I like is documenting on the website the “areas of interest” for each of the lang team members (and possibly other folks) who might be willing shepherds. This would help people figure out who to reach out to.

Isn’t there anything I can do to help move Topic X along?

This proposal does offer one additional option that hadn’t formally existing before. If you want to see something happen, you can offer to shepherd it yourself – or in conjunction with a member of the lang team. You could do this by pinging folks on discord, attending a lang team meeting, or nominating an issue to bring it to the lang team’s attention.

How many active shepherds can we have then?

It is important to emphasize that having a willing shepherd is not necessarily enough to unblock a project. This is because, as I noted above, assigning a shepherd is also a commitment on the part of the lang-team – a commitment to review progress, resolve conflicts, and keep up with things. That puts a kind of informal cap on how many active things can be occurring, even if there are shepherds to spare. This is particularly true for subtle things. This cap is probably somewhat fundamental – even increasing the size of the lang team wouldn’t necessarily change it that much.

I don’t know how many shepherds we should have at a time, I think we’ll have to work that out by experience, but I do think we should be starting small, with a handful of items at a time. I’d much rather we are consistently making progress on a few things than spreading ourselves too thin.

Expectations for a shepherd

I think the expectations for a shepherd are as follows.

First, they should prepare updates for the lang team meeting on a weekly basis (even if it’s “no update”). This doesn’t have to be a long detailed write-up – even a “no update” suffices.

Second, when a design concern or conflict arises, they should help to see it resolved. This means a few things. First and foremost, they have to work to understand and document the considerations at play, and be prepared to summarize those. (Note: they don’t necessarily have to do all this work themselves! I would like to see us making more use of collaborative summary documents, which allow us to share the work of documenting concerns.)

They should also work to help resolve the conflict, possibly by scheduling one-off meetings or through other means. I won’t go into too much detail here because I think looking into how best to resolve design conflicts is worthy of a separate post.

Finally, while this is not a firm expectation, it is expected that shepherds will become experts in their area, and would thus be able to give useful advice about similar topics in the future (even if they are not actively shepherding that area anymore).

Expectations from the lang team

I want to emphasize this part of things. I think the lang team suffers from the problem of doing too many things at once. Part of agreeing that someone should shepherd topic X, I think, is agreeing that we should be making progress on topic X.

This implies that the team agrees to follow along with the status updates and give moderate amounts of feedback when requested.

Of course, as the design progresses, it is natural that lang team members will have concerns about various aspects. Just as today, we operate on a consensus basis, so resolving those concerns is needed to make progress. When an item has an active shepherd, though, that means it is a priority, and this implies then that lang team members with blocking concerns should make time to work with the shepherd and get them resolved. (And, is always the case, this may mean accepting an outcome that you don’t personally agree with, if the rest of the team is leaning the other way.)


So, that’s it! In the end, the specifics of what I propose are the following:

  • We’ll post on the lang team repository the list of active shepherds and their assigned areas.
  • In order for a formal decision to be made (e.g., stabilization proposal accepted, RFC accepted, etc), a shepherd must be assigned.
    • This happens at the lang team meeting. We should prepare a list of factors to take into account when making this decision, but one of the key ones is whether we agree as a team that this is something that is high enough priority that we can devote the required energy to seeing it progress.
  • Shepherds will keep the lang-team updated on major developers and help to resolve conflicts that arise, with the cooperation of the lang-team, as described above.
    • If a shepherd seems inactive for a long time, we’ll discuss if that’s a problem.


  1. I could not find any single page that documents Rust’s feature process from beginning to end. Seems like something we should fix. But what I mean by moving a feature forward is basically things like “accepting an RFC” or “stabilzing an unstable feature” – basically the formal decisions governed by the lang team. ↩︎