This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Revert Abortion joke removal.
On 05/08/2018 06:38 AM, Alexandre Oliva wrote:
I guess we've somehow miscommunicated, because I'm actually pointing out
a weakness in the current rules, that was proponents of the removal of
the snippet under dispute were taking advantage of.
The argument brought forth by some of them was precisely that, because
Richard did not speak up quickly enough, he would then have to overcome
objections, because he'd have been turned into a proponent of a change
to revert what had been sneaked past him, whereas had he spoken up very
very quickly, those having to overcome objections would be those seeking
to make the change.
At the same time, it was argued that the rules were made so that those
proposing a change were the ones who had to overcome objections.
See, the *current* rules, per their argument, only serve their stated
goals for those who reply very very quickly. That's a bug.
Not really, because commits are not necessarily final. They can be
reverted (as you have demonstrated), just that there needs to be a
proper discussion if the opposition comes up after the commit has
already happened. It seems that you see the commit as being some sort
of strategic advantage, which it isn't.
It becomes an irreversible issue if the commit in question has an ABI
change and we are near a release (in which case our 1 month freeze
protects us) but in those cases we as a community stick to the more
conservative stand of backing out the change.
The patch in question is clearly not an ABI change.
And yet, I do not take advantage of the hole to denounce it. I waited
more than twice as long as the sneaked-in patch for objections.
The opposition was apparent in every email everyone sent, so it is
clearly not a consensus decision. You specifically mentioned RMS' super
powers in the project as one of the driving reasons for the change, at
which point there is no real discussion or dissent, just annoyance. If
you do the commit as you just did, you do it on the basis of those super
powers, not on the basis of consensus.
The current rules adopted by the community, however, speak of consensus
in terms that can be exploited by sufficiently stubborn people (say,
like DJ, RMS and myself :-) as DJ has just demonstrated by his blanket
objection to any of my future proposals.
I read DJ's comment as being a reaction to your comment about assumed
consensus. In the normal case (i.e. the last 7-8 years!) we have
successfully put trust in each other to act in good faith.
and don't encourage people to quickly pile on to every patch they
dislike for fear that consensus will be misjudged.
*nod*
I will take this opportunity to suggest two improvements for the
consensus rules:
- establish reasonable time limits for objections to be raised before
one can conclude that consensus was reached, or, for expediency,
explicitly allow for (not too) late objections to be raised, in such a
way that the burden of overcoming objections is not reversed just
because there *appeared* to be consensus
This assumes commits to be some sort of strategic advantage, it is not.
We are not a hostile community, as we demonstrated over the past years.
The only reason this discussion is turning hostile is because (1)
commits are being seen as some sort of strategic advantage and (2) there
are repeated calls to authority in the absence of actual merit in the
opposition.
- relax the consensus requirements such that even a sustained objection
by a (stubborn) concerned interest won't block changes indefinitely,
with fallback decision-making procedures such as collective decisions
by maintainers, escalating to GNU or somesuch
BUT
documenting that, being part of GNU, the first and only operating
system developed for ethical reasons, GNU libc may be asked to install
or to back out changes that might not seem relevant under the
technical light of the developer community, but that are deemed
important by the GNU project.
That's not the consensus part, that's the 'dear leader does not like it'
part.
Siddhesh