Wild and crazy times for the development tree

[Jeremy Katz]

On Tue, 2006-03-21 at 09:14 +0100, David Nielsen wrote:
> To sum up the point, we'd have 6 months of development time followed by
> 3 months of API stable bugfixing and updating (e.g. like what was done
> with GNOME 2.14 in FC5) then we release that as the technology preview
> Fedora, it would be considered stable. Then we follow that up with a 4
> month API stable polish cycle.

And then we have a release with basically extremely out of date software
for the next 9 months.  This ends up being pretty difficult to maintain
over the longer term.

> Pros and cons of such an out of balance approach to releasing are many
> but when I thought it up I had just had a lenghty conversation with one
> of the FC developers who was telling me that with the 9 month cycle he
> was fighting fatigue and burnout. It struck me as a user that the 9
> month cycle was completely awesome but for developers it might be hard -
> thus the need for faster paced releases and periods of light pressure.

Yep.  For all of the pressure that the 6 month cycle entails (and it
does), it definitely seemed less so than 9.  Mostly because the end is
more clearly in sight the entire time :)

> We can all agree that major surgery like replacing the Init system,
> reworking the installer, switching to modular X, switching GCC versions,
> etc. requires more time than a 6 month cycle would allow for in terms of
> testing (implementation I'm sure could be done in the 6 month cycle but
> would it be well tested?). So we will need more cycles of this length,
> depending on the amount of surgery that is planned in the near future. 

Like I've said, I don't think that a nine month cycle actually does
anything to significantly help here.  The problem is that then, if I
have something else pop up, it's easier to have it interrupt working on
whatever big new thing.  And it still ends up getting done at the last
minute.  I'm convinced this is a fundamental law of software
development :-)

Jeremy