0.15 release date changes

[John Arbash Meinel]

Erik Bågfors wrote:
...

>> Probably in the future it make sense to use 2-3 months between releases.
>>
> 
> Release early, release often. I really believe that while bzr is being
> developed this much, it's good to release often.  In each release
> there has been performance enhancements, as well as other
> enhancements.  Since lots of people compare bzr to hg in performance,
> it's really good to get updates out as fast as possible to show that
> things are happening in this regards.
> 
> I also think that having a "this is the next thing we do"-list for
> people to look at would be good.
> 
> /Erik - still alive...

There actually is a bit more overhead to a release than most people
realize. Especially since we want to report what the performance
improvements are, etc. So we need to spend a few days quantifying it,
and building up the nice performance graphs, etc.

If you look, you'll see that I didn't create Performance/ pages since
0.12. Part of that is because Wouter has been doing the releases. But
the other part is because it takes me a couple days to produce (between
setting up the tests, running them overnight to ensure consistency, and
then analyzing the results and turning them into a nice looking summary.)

There are other mental overheads of having a freeze week, etc.

It ends up that 4-weeks is probably a little too fast. We've also come
close to getting real regressions because of the last-week's "let's get
this merged before we freeze".

So I might actually recommend extending the freeze cycle to be 2-weeks.
1 with frozen features, and 1 with only regression fixes (I guess that
was supposed to be the rc week, but I'm not sure if that is really
working as expected).

I also think it might be good to switch to a 6-week release cycle.

John
=:->