RFC: Blackout period for controversial design decisions

MLIR
1

Folks - since the LLVM dev summit, we have been inundated with what I’ll call a lot of very “high calorie” discussions in MLIR, both on PRs and RFCs. Due to my somewhat unique position in the community, I get direct feedback from a lot of folks on burnout (in addition to feeling it myself). We’re at epic levels right now.

It’s important to remember that this is a volunteer project with most people trying to do the right thing in addition to fulfilling the needs of their day job. With the project norms that exist right now, we have no way to pause or sequence design discussions, and they are becoming a very large drain on the mental health of a lot of people, since if you don’t show up and formulate the right argument at any given time, something might get done that will disadvantage you.

I propose that we have a moratorium on controversial decisions or weighty design changes in MLIR for the remainder of the year. Many people are going on holiday and are stressed about also needing to stay engaged in these discussions so as to not have mis-aligned outcomes.

Let’s just remove that. No one’s cheese gets moved until we reconvene in the new year. Garden, clean things up, prototype, dream… but let’s give folks a break on needing to be on call to defend positions.

(hopefully this does not trigger controversy as I have no real standing to make such a project wide plea, but I can see the stress and have to speak up)

15 Likes
2

I support this. Normally we don’t need this kind of request, but there’s just too much going on in MLIR at the moment. Last thing we want is for controversial decisions to be selected by “the ones still in the room”. Similarly, if consensus is formed on the already open threads, let’s wait until the new year comes and the well rested folks come back and have a gander.

What constitutes a controversial decision is fuzzy, but getting it half-right is already better than completely wrong. Hopefully early Jan we’ll have a good proposal for the MLIR governance process and we’ll not reach this scale of burnout again.

1 Like
3

+1 to Christmas truce on RFCs, as long as we don’t all come back to dozens of postponed RFCs arriving at the same time.

1 Like
4

One of the things that governance of the areas needs to solve is the ability to throttle and prioritize design discussions. These discussions are not all created equal and represent real work by the contributors. When there are too many of certain types or a false sense of urgency, stress and burnout go up.

Hopefully some more structure gives the groups a better ability to manage their agendas and avoid that.