I think the overwhelming need is that more than the author
understands and agrees with the main thrust of a major change. I
think that the apply-and-test RTC requirement we've been struggling
under has severely inhibited thorough review, but that's just my
experience.
I'm in favor of (3) for trunk and (2) for bug-fix branches and also
in favor of reviewing this change in a month or two to see if we are
actually reviewing changes sufficiently and in a sufficiently timely
manner.
thanks
david jencks
On Sep 7, 2006, at 12:59 PM, Kevan Miller wrote:
Thanks Matt and Jason.
I'll update the proposals to remove the PMC requirement from
Relaxed RTC.
I'm also going to update to make clear that the focus of this vote
is trunk development. Initially, it would apply to branches, also.
If we feel that branches need special controls, we should handle by
refining our general process. I don't wan't to get hung up fixing
*everything*...
Any other comments?
More inline...
On Sep 6, 2006, at 7:27 PM, Jason Dillon wrote:
Thanks Kevan for the summary, I think this really helps.
More comments below inline...
On Sep 6, 2006, at 1:34 PM, Kevan Miller wrote:
1. Relaxed RTC
...
* 3 +1 votes from committers on the project (1 of these
committers needs to be a member of the PMC) with no outstanding
-1 votes.
I don't think that we need a requirement of a PMC vote here. The
PMC can provide enough oversight w/o a required vote. IMO, a vote
is a vote is a vote... I don't put any weight over a committers
vote to a PMC members vote. I value them both equally. Forcing a
PMC vote here is artificial and promotes separatism rather than
unity.
Totally agree. I almost made that change, but was consciously
trying *not* to edit the proposals as I pulled them out of the
various discussion threads.
2. RTC with Lazy Consensus
Geronimo follows a Review-Then-Commit model with Lazy consensus.
Patches for new function are provided by developers for review
and comment by their peers. Feedback is conducted through JIRA
comments. The goal of this interaction is to solicit suggestions
from the community and incorporate their feedback as appropriate.
A patch is accepted if:
* 3 +1 votes from committers on the project with no outstanding
-1 votes and no significant, ongoing discussion
* 72 hours pass with no outstanding -1 votes and no significant,
ongoing discussion. A 24 hour warning should be sent to the dev
list.
For the most part I like this model... but only for non-trivial
changes, see below.
It's not bad and I think we can operate quite reasonably using
this. However, I don't think it's necessary. I'd prefer to see the
benefits of this approach handled by more up-front communication
using CTR. By the time something is committed, it's not much of a
surprise...
3. CTR with documentation guidelines
Geronimo follows a Commit-Then-Review model. There should be an
emphasis of community communication. Community-based policing and
persuasion will be used to remedy any problem areas. Guidelines
are not strict dogma -- common sense should prevail. Community
communication is the key, not a process. General guidelines are:
* Non-trivial changes (and certainly potentially controversial
changes) should be announced on the dev list. This announcement
should be well in advance of the change being committed. The
community should be given the opportunity to understand and
discuss the proposal.
* Concurrent with the commit of a significant change, the
committer should document the change on the dev list. You should
describe what you are doing, describe why you are doing it, and
provide an overview of how you implemented it.
This feels a whole lot like common-sense for how to participate on
an open-source project. In most cases, I think this is also the
best model to run under... though I do believe that RTC has some
merit as well.
If it was up to me (which it isn't, but here is my opinion
anyways), I would use a hybrid model, which would default to CTR
(with emphasis on common sense and communication) and for non-
trivial or potentially controversial changes follow the RTC with
Lazy Consensus as described in #2 (with the addition of inclusion
of development branches or patches, depending on the complexity).
I actually think that this is common-sense too.
IMO... this is the best of both worlds, without being too
overbearing on policy and process, leveraging the trust of the
developers and fostering our community with sufficient oversight
and freedom to allow real progress to be achieved.
I think RTC w/ lazy consensus is always available under a CTR
model. However, I'd prefer it be self-imposed rather than expected.
I wouldn't be upset if someone didn't follow RTC for a significant
change. I would be upset if they failed to adequately communicate
and discuss with the community...
* * *
But Jason... how do I know what is non-trivial or controversial?
Jason, Perhaps it's time for a little holiday... ;-)
--kevan
