Re: pkgsrc-2013Q4 freeze started

[Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost>]

wiz@ wrote:

> I trust most NetBSD developers to be able to decide for themselves if
> an update is likely to break stuff and if it is important to go in
> before the branch. Not every package update is really that important
> that it can't wait until the next branch.

I'd just avoid vague complaints against leaf package update.

> > > let's try fixing stuff instead, we can update all packages again in ~2
> > > weeks.
> > 
> > Please consider stable branch users and binary packages users.
> > 
> > If no updates are committed during these two weeks, they won't be
> > able to get new versions for three months, while small breakages
> > in leaf packages can be easily fixed by pullups in any time.
> 
> True, but pullups are also work, for releng. Better to not break stuff :)

In my opinion, package systems should rather consider about users
and market, not internal developers..

> > I wonder if we can have simple but defined rules for leaf packags like:
> > - no updates are allowed in the last two (or three) days of freeze
> 
> That sounds good.
> 
> > - "minor" updates is allowed
> > - "major" updates require prior approval (by MAINTAINER or PMC)
> 
> But then you have the same issue as above: you have to trust the
> developer to decide what is a minor update and what is a major one.

We can define major/minor if neccesary per version numbers, changes logs.
But I'm afraid people won't read complex rules.

> How is that different from 'ok, but not recommended.'?

It's more specific than "not recommended" I believe.

> >  - all updates must have "complete changes list" in the commit log
> 
> That's not special for the freeze, that should always happen.

Yes, it should, but not so much people do so unfortunately.

> > - any update that breaks build will be reverted silently
> 
> Well, I'd even give people a day to fix stuff.
>
> > etc.
> 
> So you agree that no list will ever clarify all cases :)

I really hate such NetBSD's "best or nothing" strategy.

> qemu is not a leaf package,

What's the actual definition of "leaf pacage"?  How can I check it?

---
Izumi Tsutsui