Phil Steitz a écrit :
> Henri Yandell wrote:
>> On Sat, Mar 27, 2010 at 4:30 PM, sebb <seb...@gmail.com> wrote:
>>> On 27/03/2010, Gary Gregory <ggreg...@seagullsoftware.com> wrote:
>>>> Hi Hen, well done to get the ball rolling. More below.
>>>>
>>>>
>>>> > -----Original Message-----
>>>> > From: Henri Yandell [mailto:flame...@gmail.com]
>>>> > Sent: Saturday, March 27, 2010 14:08
>>>> > To: Commons Developers List
>>>> > Subject: [LANG][COLLECTIONS] Beta releases
>>>> >
>>>> > Possibly a query for IO too if it's 2.0 has large changes.
>>>> >
>>>> > Given the large API changes in Lang 3.0 and Collections 4.0, a beta
>>>> > release seems like a very useful thing (kudos to pbenedict for
>>>> > convincing of me that months ago on IM :) ).
>>>> >
>>>> > I'm interested in what advice and thoughts people might have on the
>>>> > subject. Areas I can think of are:
>>>> >
>>>> > 1) versioning, does JIRA identify the version as 3.0-beta1; or just
>>>> > have a 3.0 and treat the beta as an invisible release? I'm preferring
>>>> > the latter.
>>>>
>>>>
>>>> I think there is also "nightly-build" available. If bugs are logged
>>>> against "3.0" and fixed in "3.0", then the understanding is that the bug
>>>> was fixed in the alpha/beta/RC cycle. It seems fine if a little mysterious
>>>> though. +0.
>>>>
>>> Surely you can just add whatever versions you like to the JIRA
>>> configuration?
>> I don't want someone to come along later and have to figure out what
>> 3.0 was from the release notes from N 3.0 alpha, beta, gamma releases.
>> And I don't want to have to modify lots of issues later to roll into a
>> 3.0.
>>
> +1 - best to leave JIRA 3.0
I don't understand the problem.
>
>>>> > 2) Maven - does the beta go to the main Maven repo, or just tell
>>>> > people to pull from snapshot (and make sure there are current
>>>> > snapshots in the snapshot repo)? I'm thinking the latter.
>>>>
> +1 - snapshot
+0
Despite these versions are expected to be less stable and don't want
people to use them for production, they will be officially released, so
they are worth being preserved.
>>>> +1
>
>>> Agreed, should go to snapshot repo. If there are subsequent API breaks
>>> then having the beta in the main repo is likely to cause grief to
>>> others and to us at some point.
>>>
>>>> > 3) Announcements - blogging, announce@ type announcements presumably.
>>>>
>>>>
>>>> +1. Same as for a release I would think since we are talking about
>>>> important changes and asking for feedback. A broad audience is required.
>>>>
>>> +1
>>>
> +1 - with the right disclaimers in the announcements / messages
+1
>
> We would need to follow the normal release process, including PMC
> vote to release. We should also talk to the infra@ ppl to make sure
> they are OK with the snapshot repo (non-mirrored) hosting. I don't
> think this should be an issue because we are not likely talking
> about a large volume of downloads, but we should at least ask about it.
>
>>>> > 4) Length of time spent in beta. I think we should define this up
>>>> > front.
>>>>
>>>>
>>>> +1. At least 1 month? I would also like to see at least 1 week pre-beta
>>>> warning to allow interested committers to put this on their radar and
>>>> contribute before the beta goes out.
>> Fair enough. And I'm thinking 3->6 months. Possibly 3 months with a
>> 2nd beta then for 3 more months.
This seems very long. A long-lived version may be considered stable
enough to users who may think it is a final version despite its name.
Alpha and beta should not live more than a few weeks. One month may be a
good compromise.
>
> Sounds about right.
>>>> > The intent would be to get early adopters using and finding bugs, but
>>>> > more importantly drive conversation around the API changes and suggest
>>>> > new ones. I want us to be able to change an API without having to say
>>>> > "Yeah, that was dumb - sadly we have to wait 'til 5.0".
>>>>
>>>>
>>>> That sounds like a good intention, IMO this means at least 2 betas, 1) ask
>>>> for feedback, 2) provide new alpha/beta with feedback changes.
>>>>
>>>> Tangent: If we are talking about changing APIs, shouldn't these really be
>>>> called Alphas and leave a Beta out for a stable API and bug finding only?
>>>> The drawback is that it might harder to get interest from a wide audience
>>>> on more than one pre-release version.
>>>>
>>> +1 Alpha to be used for fluid APIs.
>>> -1 API changes allowed in Betas.
>> Betas sound pretty pointless then in this case :) You'd be mad to sign
>> up for no API changes in beta for a release with major API changes
>> (unless you follow beta-1 with alpha-4).
>>
> I think we need to be clear on this from the beginning - the API
> itself is being {alpha, beta}-tested.
>
>>> So long as the rationale for the naming is well explained - i.e. that
>>> Alpha is only used because the API might change, not because of the
>>> intrinsic quality of the implementation - I'm hopeful Alpha/Beta won't
>>> put people off.
>> *shrug* :)
>>
>> I don't tend to see public alpha releases anymore, "beta" seems to be
>> the phrase for 'this is out, but we're making no promises' even if it
>> doesn't make logical sense for there never to be visible alpha
>> releases. Call it early-access, or developer-preview :)
>>
> That could actually be a good reason to call them alphas - sticks
> out as something different from beta. We are asking users not just
> to test the functionality, but actually more importantly to review
> and test the API.
-0 to alpha. This would scare people and we would end up having done a
release for commons developers only, when we want to get some feedback
from outside. Alpha are not released, they correspond to the current
state in the subversion repository, and beta are versions that can go to
the adventurous public.
>
> I think this is a good idea that will help other components once we
> find a way to do it. We probably should have done something similar
> for [math] 2.0 and I can imagine the same kind of situation in the
> future for both [pool] and [dbcp]. Getting major version release
> APIs correct requires somehow cajoling more than just the usual
> suspects into playing with the new APIs and pointing out what is,
> well, "suboptimal" about them.
I agree. We have done some mistakes with [math] 2.0 that could have been
avoided with such a process.
Luc
>
> Thanks, Hen!
>
> Phil
>
>
>> Hen
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
>> For additional commands, e-mail: dev-h...@commons.apache.org
>>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
> For additional commands, e-mail: dev-h...@commons.apache.org
>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org
