You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Karl Fogel <kf...@red-bean.com> on 2008/02/28 21:32:52 UTC
1.6 and getting back to our release policy.
Once upon a time, we had a release policy [1]. It said that releases
would be roughly time-based, that each release would include at least
one new feature or "significant differentiator", but that we wouldn't
say specifically which feature it would be. Whatever's ready when the
date arrived could be the release definer. It wouldn't matter so much
*what* the feature was, as long as there was one. After all, what
doesn't get shipped in this release would just get shipped later.
Unfortunately, we ignored this policy for 1.5.
Instead, we decided that Merge Tracking would define the release.
Then we delayed the release as long as necessary to get basic Merge
Tracking into it, even though other features were ready earlier
(WebDAV proxy, sparse directories, etc), or could have been ready if
we hadn't felt such mad pressure to finish Merge Tracking because we
decided that it -- and nothing else -- would be the release definer.
Can we please, please do things the other way for 1.6?
Let's say "1.6 will come out approximately six months after 1.5, and
whatever is ready will go into it." If improved tree-conflicts are
ready, great. If reflective merges are ready, great. If something
else is ready, great. But -- and here's the important thing -- if
only *one* thing is ready, the release should *not* wait for any other
thing. At six month windows, the price of delaying a not-quite-ready
feature is not so high, and when it does come out it will be mature.
As it was, we paid both coming and going for our indiscipline: because
we refused to release 1.5 for so long, we finally had to compromise on
some parts of Merge Tracking in order just to ship.
I know these are not new thoughts; Justin recently posted saying the
same thing. I just thought I'd point out that we wrote up this policy
years ago, and pretty much all thought it was a good idea, and then we
ignored it anyway. We can stop ignoring it any time we want :-).
-Karl
[1] http://subversion.tigris.org/roadmap.html#release-planning
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: 1.6 and getting back to our release policy.
Posted by Erik Huelsmann <eh...@gmail.com>.
> As it was, we paid both coming and going for our indiscipline: because
> we refused to release 1.5 for so long, we finally had to compromise on
> some parts of Merge Tracking in order just to ship.
>
> I know these are not new thoughts; Justin recently posted saying the
> same thing. I just thought I'd point out that we wrote up this policy
> years ago, and pretty much all thought it was a good idea, and then we
> ignored it anyway. We can stop ignoring it any time we want :-).
Thanks Karl for putting it forward as eloquent as you always do. I can
only say "hear, hear!".
> -Karl
>
> [1] http://subversion.tigris.org/roadmap.html#release-planning
Bye,
Erik.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: 1.6 and getting back to our release policy.
Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
On Thu, Feb 28, 2008 at 1:32 PM, Karl Fogel <kf...@red-bean.com> wrote:
> Once upon a time, we had a release policy [1]. It said that releases
> would be roughly time-based, that each release would include at least
> one new feature or "significant differentiator", but that we wouldn't
> say specifically which feature it would be. Whatever's ready when the
> date arrived could be the release definer. It wouldn't matter so much
> *what* the feature was, as long as there was one. After all, what
> doesn't get shipped in this release would just get shipped later.
>
> Unfortunately, we ignored this policy for 1.5.
>
> Instead, we decided that Merge Tracking would define the release.
> Then we delayed the release as long as necessary to get basic Merge
> Tracking into it, even though other features were ready earlier
> (WebDAV proxy, sparse directories, etc), or could have been ready if
> we hadn't felt such mad pressure to finish Merge Tracking because we
> decided that it -- and nothing else -- would be the release definer.
>
> Can we please, please do things the other way for 1.6?
>
> Let's say "1.6 will come out approximately six months after 1.5, and
> whatever is ready will go into it." If improved tree-conflicts are
> ready, great. If reflective merges are ready, great. If something
> else is ready, great. But -- and here's the important thing -- if
> only *one* thing is ready, the release should *not* wait for any other
> thing. At six month windows, the price of delaying a not-quite-ready
> feature is not so high, and when it does come out it will be mature.
>
> As it was, we paid both coming and going for our indiscipline: because
> we refused to release 1.5 for so long, we finally had to compromise on
> some parts of Merge Tracking in order just to ship.
>
> I know these are not new thoughts; Justin recently posted saying the
> same thing. I just thought I'd point out that we wrote up this policy
> years ago, and pretty much all thought it was a good idea, and then we
> ignored it anyway. We can stop ignoring it any time we want :-).
+1 to all of the above.
We have no rule that says we have to issue perfect releases as long as
we have a committment to do better next time. =) -- justin
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: 1.6 and getting back to our release policy.
Posted by Karl Fogel <kf...@red-bean.com>.
Talden <ta...@gmail.com> writes:
> I'd prefer to say "intend releases to span no more than 6 months"
> allowing for those occasions when some really significant piece
> gets finished that might be worth releasing without waiting for
> the next 6 month release.
Thank you, that's what I should have said. 6 months is an upper
bound, not a lower bound.
> I'd also like to see more attention to releasing the minimal
> useful incarnation of a feature sooner rather than attempt to
> cover every little variation - of course enough diligence must be
> paid to design to ensure that the approach taken does not
> unnecessariliy introduce incompatibilities for the completion of
> the feature. IMHO the suggestions the complete inability to get
> development traction on the 'obliterate' functionality is a
> poster-child for this situation.
+1 in general (although I think what's going on with 'obliterate' in
particular is a little more complicated than that -- we'd all be happy
to release even an incomplete implementation of it, if only we had at
least a basic design).
-Karl
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: 1.6 and getting back to our release policy.
Posted by Talden <ta...@gmail.com>.
> Can we please, please do things the other way for 1.6?
>
> Let's say "1.6 will come out approximately six months after 1.5, and
> whatever is ready will go into it." If improved tree-conflicts are
> ready, great. If reflective merges are ready, great. If something
> else is ready, great. But -- and here's the important thing -- if
> only *one* thing is ready, the release should *not* wait for any other
> thing. At six month windows, the price of delaying a not-quite-ready
> feature is not so high, and when it does come out it will be mature.
I'd prefer to say "intend releases to span no more than 6 months"
allowing for those occasions when some really significant piece
gets finished that might be worth releasing without waiting for
the next 6 month release.
As you've suggested, a release definer is the set of features
that are ready to go.
I'd also like to see more attention to releasing the minimal
useful incarnation of a feature sooner rather than attempt to
cover every little variation - of course enough diligence must be
paid to design to ensure that the approach taken does not
unnecessariliy introduce incompatibilities for the completion of
the feature. IMHO the suggestions the complete inability to get
development traction on the 'obliterate' functionality is a
poster-child for this situation.
--
Talden
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org