Hi all,
Let me start by apologizing for taking this long to send this email. The
attentive reader will have noticed my name in Russ' original draft as
one of the people who reviewed it. When Russ sent his initial proposal,
I started drafting a large response that I lost due to a silly mistake
on my end. Life then intervened, and I didn't have time to follow up
until now.
That said,
I think introducing a fixed time limit on a GR is a bad idea, for
various reasons.
First of all, no matter how careful we choose a time limit based on
historical precedence, I think it is naive to assume we will be able to
come up with a time limit that will always work for all future GR
proposals. Some issues are contentious, and they take time to work
through them. In my reading, the longest we have ever taken to decide on
a vote was for GR 2006-003, where the initial proposal was sent in June
but the eventual vote only opened in September[1].
An argument that has been brought forward to avoid that problem is that
it is theoretically possible to discuss matters on this list before
starting the formal procedure. While that is true, I would like to point
out that the whole reason for the introduction of this time limit is so
that people can't game the system by using procedural measures to delay
the vote, possibly indefinitely. If we don't want to allow people to
delay the vote, we should *also* not allow people to game the system by
forcing a vote prematurely, through proposing a formal GR when someone
else offers incomplete ideas that they would have preferred to see
discussed first. Again, I would point to GR 2006-003 where something
along those lines happened[1].
I fear that in order to avoid that pitfall, people may wish to discuss
things in private amongst themselves rather than posting something to
the -vote list when they want to start looking at a problem, which will
give them an unfair time advantage.
Additionally, it is not always possible to foresee all of the complexity
in a problem space; we may be quite a bit into the formal discussion of
the ballot when we decide that there are some significant issues that
require exploring and which would benefit from more time. If everyone
involved agrees that this is a good thing, then I think we should allow
for that possibility; the proposed procedure does not do so, and I fear
that this may result in rushed proposals that are suboptimal and do not
resolve the issue at hand.
An argument that has been brought forward to remedy this is that it is
theoretically possible to advance a vote for the default option in such
a case. While this is true, that is problematic: first of all, it will
delay the resolution of the situation by a significantly larger amount
of time (you will need to go through a complete vote only to have to do
the whole process over again). I think it is relevant in this context
that we only managed to do this one time in the history of the project,
in a vote where I can't help but feel like the proponents of the vote
tried to game the system (2006-005[2]).
We might have been able to use this for 2004-004, but alas.
Additionally, I believe that proposing we vote the default option more
often is antithesis to what we *should* be doing. I think a GR vote
should generally be the final answer to an issue we are dealing with,
and that in order to do that, we should ensure that the ballot is
complete, with all relevant options represented. I don't think we get
that if we introduce a rigid timeline that cannot be diverted from in
exceptional situations.
I hear and agree with the argument against such a procedure; having a
way to delay the vote which everyone can trigger opens the system up to
abuse, which could allow the vote to be delayed indefinitely if
formulated badly. I believe the answer to that is not to remove the
option to delay the vote entirely, but to restrict the conditions under
which such a delay can be invoked; only provide it to the DPL, or
provide only a limited number of delays, or allow a majority of people
who proposed options that are already on the ballot to object, or
something along those lines. The goal should be to end up with a
procedure where *can* extend the discussion period if discussions are
actually still happening and we believe it is valid, without allowing
people who want to avoid any vote from happening entirely to delay
things indefinitely.
Additionally, this proposal does not remedy what I think is another
issue we have with our procedure, which I have been wanting to resolve
one way or the other for quite a while but have no idea how to do so;
the "I want to create a ballot with all possible options" antipattern.
We had a few cases of this over the years (at least two that I can
remember), usually by well-intentioned people who want to get things to
a vote quickly, but I honestly believe it creates more problems than it
attempts to solve.
Writing an option that does not represent your own personal opinion is
extremely difficult at best, and is probably not even possible at all.
This means you'll get proposals from people who actually hold an opinion
very close to what you wrote down that you'll then need to evaluate to
decide whether this improves the ballot option, or whether it changes
the option into something different entirely and thus should be a
separate ballot option instead. To deal with such a proposal from the
point of view of someone who feels one of the options is
almost-but-not-quite something you can vote for can be very frustrating,
as I experienced first-hand in
https://lists.debian.org/debian-vote/2019/11/msg00032.html .
I also believe that a ballot with options that were written by people
who do not support that option will usually result in a cluttered
ballot, with various options that are almost but not quite the same
thing, and options that are irrelevant noise and which will never win. I
think this behavior should be discouraged if not outright forbidden
(although, again, I'm not sure how to forbid them), and would note that
adding a strict time limit seems more likely to create private
discussions (as I've explained above) and therefore to me seems more
likely to result in this antipattern.
I'm not submitting a formal draft to go against yours at this point
(although I do have the beginnings of one), because I would like to see
whether I'm alone in this opinion or not, where only in the latter case
it would make sense to continue down this path.
Thanks for your thoughts,
[1] In the case of 2006-003, I started a discussion on -vote in order to
start the debate before formally proposing a GR, intending to
explore the problem before starting to build the ballot. The first
follow-up to that mail however was a formal GR proposal by Manoj,
which then started the procedure. It was not contentious vote
though, and I think technically the vote may have expired at least
once, although I'm not 100% sure about that -- it involved a few
amendments resetting the timer and the DPL extending the minimum
discussion period after one of them, so the details are a bit
muddy.
[2] In 2006-005, Denis Barbier proposed a vote to recall the project
leader. The DPL then delegated the authority to vary the discussion
and voting periods to him and Loïc Minier. Denis chose to not accept
any amendments and reduced the discussion time to the minimum, and
called for a vote while, in my opinion, the discussion was still in
full force.
--
w@uter.{be,co.za}
wouter@{grep.be,fosdem.org,debian.org}
Attachment:
signature.asc
Description: PGP signature