Re: Stable release state

[David Kastrup]

Werner LEMBERG <address@hidden> writes:

>>> Declare a feature freeze for a few weeks, with only trivial
>>> additions and fixes for regressions and documentation issues.
>> 
>> The time for a formal decision was too late, the more active
>> developers informally agreed to try and nobody bothered actually
>> doing so.
>
> Hmm.  We discussed that, indeed, but noone stepped forward and
> *declared* it.

There is nobody with that kind of authority.

> I would like to make you our benevolent dictator who can decide
> this...

If you take a close look, the "benevolent dictator" model of the Linux
kernel is actually non-existent.  It's just that people like to call
Linus' personal but published repository canonical.  But it's not really
privileged in comparison with other people's personal public
repositories.

His "dictatorship" does not give him any special powers or requires any
enforcement.  So what are his infamous tirades about that make headlines
about twice a year?  They are not about people bypassing his power (how
could they?), but rather about breakdowns of his network of technical
trust: he himself applied some patch in the expectation that whoever
passed it to him would have reasonable expectations for it to meet the
required quality criteria.

We don't have something like a network of skilled active subsystem
maintainers.  In consequence, gatekeeping the central repository would
be a lot of effort.

We _did_ apply the "benevolent dictator" model to the stable branch in
the past.  Where is the difference as opposed to Linux kernel
development?  Basically testing coverage.  Developers are interested in
working with Linus' version of Linux.  At the minimum, it is what
patches tend to be based on.  In addition, any patch of suitable
interest is sure to get some people pounding on it even before it gets
into Linus' view.  Few if any people bother checking out
release/stable/*, in contrast, and our reviewing tools and processes
don't bother with it, either.

Also we have far fewer people interested in whole "add-on" patch series,
and our review tools make dealing with them awkward as well, leading to
a high amount of resistance against "can you put this in a branch for
now" requests.  The skyline stuff was kept in a branch for quite a
while, though, and was tested from there by several people.

So in a nutshell, it's easy to declare me the benevolent dictator of the
stable branch, but it has rather limited effects.  In contrast,
declaring me dictator over LilyPond master would, due to our
master-centric development tools, put more resources exclusively under
my control (meaning that you can't work well with them except under my
supervision) than useful for continued parallel/independent development.

So I don't really see a policy choice that can be implemented without
some major drawbacks, due to limitations of our processes and tools.

Basically, I think we are out of pretty solutions, and I am wary of the
repercussions of some uglier solutions.  That's frustrating, and
frustration does not really increase the prettiness of what I am able to
come up with.

-- 
David Kastrup