GLEP 39: Gentoo metastructure

Author Grant Goodyear <[email protected]>, Ciaran McCreesh <[email protected]>
Type Informational
Status Final
Version 3
Created 2005-09-01
Last modified 2023-05-08
Posting history 2005-09-01, 2006-02-09, 2007-10-12, 2008-01-19, 2022-11-25, 2023-04-10, 2023-04-16
Replaces 4
GLEP source glep-0039.rst

Status

Implemented. The metastructure proposal was accepted by a vote of all Gentoo developers on 2005-06-14 [1]. GLEP amended on 2006-02-09 to add the final bullet point to list B in Specification.

Updated by an all-developers vote on 2023-05-08:

  • Replace leaving council members by next in line [2].
  • Updating this document requires an all-developers vote [3].
  • Council members must be developers [4].
  • An inquorate council meeting cannot take any substantive action.
  • Drop hard requirement of yearly project lead elections.

Abstract

GLEP 4 is replaced with a new "metastructure" that retains established projects (and makes new projects easier to create), but adds a new "Gentoo Council" to handle global (cross-project) issues.

Motivation

The Fosdem and subsequent reform proposals shepherded by Koon are thorough, extremely detailed, and somewhat complicated. They have a lot of good ideas. For many who have been with Gentoo a long time, though, there's just something about them that they don't really like. More than a few Gentoo devs are almost entirely uninterested in metastructure as long as it doesn't get in their way, and because the current proposals impose at least some order on our unruly devs these proposals are guaranteed to "get in the way" to some degree. For example, a frequent comment that has been heard is that many Gentoo devs don't know who his/her manager (or project lead) is, which is a clear indication that our current system is broken. The existing proposals solve the problem by requiring that each dev belong to a project. Perhaps the part that is broken, though, is the belief that every dev should have a manager. The history of Gentoo is such that traditionally big advances have often been implemented by a single or a small number of dedicated devs (thus our long-standing tradition that devs have access to the entire tree), and surely we do not want to make things harder (or less fun) for such people. So here's a minimal proposal for those who remembers the "good ol' days" and thinks things aren't really so different now.

Synopsis of the current system

  • There are 13-15 top-level projects (TLPs). Top-level projects are comprised of sub-projects, and the goal was that every Gentoo project would be a sub-project of one of the TLPs. Supposedly each dev therefore belongs to one or more TLPs.
  • Each TLP has at least a "strategic" manager, and potentially also an "operational" manager. Only the strategic managers vote on global Gentoo issues.
  • The managers of each TLP were appointed by drobbins, the other TLP managers, or elected by their project members. These managers have no set term.
  • Within each TLP the managers are responsible for making decisions about the project, defining clear goals, roadmaps, and timelines for the project, and solving problems that arise within the TLP (see GLEP 4 for the specific list).
  • The strategic TLP managers are also responsible for deciding issues that affect Gentoo across project lines. The primary mechanism for handling global-scope issues is the managers' meetings.
  • Disciplinary action taken against erring devs is handled by the "devrel" TLP, unless the dev is a strategic TLP manager. In that case disciplinary action must be enacted by the other strategic TLP managers.

Problems with the existing system

  1. The assumption that TLPs are complete is either incorrect (there still is no "server" TLP) or just plain weird (but the lack of a server TLP is technically okay because all devs who don't have an obvious TLP belong to the "base" TLP by default).
  2. There is nothing at all to ensure that project leads actually do represent the devs they supposedly lead or satisfy their responsibilities. Indeed, should a TLP manager go AWOL it is not at all obvious how the situation should be resolved.
  3. Nothing is being decided at global scope right now. Some TLP strategic managers rarely attend the managers' meetings, and the managers as a whole certainly are not providing any sort of global vision for Gentoo right now.
  4. Even if the strategic TLP managers were making global decisions for Gentoo, the TLP structure is such that almost all devs fall under only one or two TLPs. Thus voting on global issues is hardly proportional, and thus many devs feel disenfranchised.
  5. Regardless of whether or not it is justified, devrel is loathed by many in its enforcement role.

Additional problems identified by the current metastructure reform proposals

  1. The current system has no mechanism for identifying either projects or devs that have gone inactive.
  2. Bugs that cut across projects often remain unresolved.
  3. GLEPs often linger in an undetermined state.

Specification

  1. A project is a group of developers working towards a goal (or a set of goals).
    • A project exists if it has a maintained Wiki project page as described below. ("Maintained" means that the information on the page is factually correct and not out-of-date.) If the Wiki page isn't maintained, it is presumed dead.
    • It should have at least one lead, and the leads are selected by the members of the project. This selection should occur at least once every 12 months, and may occur at any time. Any member can demand a lead election if the last election was more than 12 months ago.
    • It may have zero or more sub-projects. Sub-projects are just projects that provide some additional structure, and their Wiki pages are defined as sub-projects of the parent project.
    • Not everything (or everyone) needs a project.
    • Projects need not be long-term.
    • Projects may well conflict with other projects. That's okay.
    • Any dev may create a new project just by creating a new project page on the wiki.gentoo.org (see [5]) and sending a Request For Comments (RFC) e-mail to gentoo-dev. Note that this GLEP does not provide for a way for the community at large to block a new project, even if the comments are wholly negative.
  2. Global issues will be decided by an elected Gentoo council.
    • There will be a set number of council members. (For the first election that number was set to 7 by acclamation.)
    • Council members will be chosen by a general election of all devs once per year.
    • Council members (and their proxies) must be Gentoo developers.
    • The council must hold an open meeting at least once per month.
    • Council decisions are by majority vote of those who show up (or their proxies).
    • If a council member (or their appointed proxy) fails to show up for two consecutive meetings, they are marked as a slacker.
    • If a council member who has been marked a slacker misses any further meeting (or their appointed proxy doesn't show up), they lose their position.
    • Whenever a member of the council loses their position (the reason is irrelevant; e.g. they resign or they are booted for slacking), then the next person in line from the previous council election is offered the position. If they accept and the current council unanimously accepts the new person, they get the position. Otherwise, it is offered to the next person in line, and so forth. If the council does not accept that person, then a new election is held to choose a new member. The new member gets a 'reduced' term so that the yearly elections still elect a full group.
    • Council members who have previously been booted for excessive slacking may stand for future elections, including the election for their replacement. They should, however, justify their slackerness, and should expect to have it pointed out if they don't do so themselves.
    • The 'slacker' marker is reset when a member is elected.
    • If any meeting has less than 50% attendance by council members, a new election for all places must be held within a month. The 'one year' is then reset from that point. No substantive action can be taken in any such meeting.
    • Disciplinary actions may be appealed to the council.
    • A proxy must not be an existing council member, and any single person may not be a proxy for more than one council member at any given meeting.

Rationale

So, does this proposal solve any of the previously-mentioned problems?

  1. There is no longer any requirement that the project structure be complete. Some devs work on very specific parts of the tree, while some work on practically everything; neither should be shoehorned into an ad-hoc project structure. Moreover, it should be easy to create new projects where needed (and remove them when they are not), which this proposal should enable.
  2. By having the members choose their project leads periodically, the project leads are necessarily at least somewhat responsible (and hopefully responsive) to the project members. This proposal has removed the list of responsibilities that project leads were supposed to satisfy, since hardly anybody has ever looked at the original list since it was written. Instead the practical responsibility of a lead is "whatever the members require", and if that isn't satisfied, the members can get a new lead (if they can find somebody to take the job!).
  3. If the council does a lousy job handling global issues (or has no global vision), vote out the bums.
  4. Since everybody gets to vote for the council members, at least in principle the council members represent all developers, not just a particular subset.
  5. An appeal process should make disciplinary enforcement both less capricious and more palatable.
  6. This proposal doesn't help find inactive devs or projects. It really should not be that much of a problem. We already have a script for identifying devs who haven't made a CVS commit within a certain period of time. As for moribund projects, if the project page isn't maintained, it's dead, and we should remove it. That, too, could be automated. A much bigger problem is understaffed herds, but more organization is not necessarily a solution.
  7. The metabug project is a great idea. Let's do that! Adding a useful project shouldn't require "metastructure reform", although with the current system it does. With this proposal it wouldn't.
  8. This proposal has nothing to say about GLEPs.

Updates to this document

Any major updates to this document (that is, those that change its content rather than just fixing typos or adding small clarifications) require a vote of all developers. Eligible voters are all developers at the time when the proposed update is published. The vote passes if both of the following conditions are fulfilled:

  • The ratio of positive to negative votes is at least two to one, and
  • the number of positive votes is no less than one quarter of the number of eligible voters.