Re: [Building Sakai] 2.6.x pom <version> after 2.6.0 release

Beyond the issue of patch pain and the ease with which one should be able to do an svn update on a branch without some crazy incrementing versioning pattern for my poms getting in the way (which, I am not against), I think it is important to bear in mind that it is a risky assumption (and will remain a risky assumption whatever versioning guidelines we adopt) to assume that one can add (or should be able to add) contrib projects to a Sakai build without first ensuring that the base poms of each contrib project points to the correct Sakai parent pom.

If today I take the head of 2.5.x branch (version M2) and then drop in two well-managed contrib projects,  Mneme 1.2m2 and QNA 1.0 (or QNA 2.5.x) without first confirming the parent pom to which these contrib project expects to bind, will I achieve a successful Maven build?  Nope.

So the idea that deployers can avoid or ignore some sort of manual or programmatic intervention when combining core and contrib projects into a new build because we choose a non-incrementing versioning scheme, is I think unrealistic.

Cheers,

Anthony

On Mar 27, 2009, at 5:08 AM, Stephen Swinsburg wrote:

A minor comment on the branch version being out of date, bear in mind the branch DOES stay the same, theoretically identical. No API changes, UI changes or tool behaviours should change, unless they are bugfixes. So the constant version of the branch is still valid as it really is only one version, just with bugfixes.

That being said, if the version in contrib projects can be set to an official release with the appropriate maven repo definition setup, like you said David, then this whole problem could go away. This is where we need some "Guidelines for Contrib projects" perhaps.

regards,

Steve

On 27/03/2009, at 8:24 AM, David Horwitz wrote:

Hi Guys,

A couple of thoughts - I'm generally against the idea of the branch version staying the same for the lifetime of the branch - it leads to the version becoming devalued and introduces increasing uncertainty about what version of a dependency your project may actually be using.  We need to remember that by the standards of open source project our release cycles are long (years), and that we're using the same version number to describe a wide range of code of varying maturity and stability.

On the issue that Seth mentioned about maintaining contrib projects - there is no reason for 99.9% of contrib projects to bind their versions to a non-release Sakai version. If you set your Sakai version to a release (e.g. 2.5.3) and add the definition of the Sakai maven repo to your projects base pom - it will build and run for any 2.5.* version (and probably all 2.6 versions too)

David

Stephen Swinsburg wrote:

I really do feel that the maintenance releases should have a stable <version> number associated with them, which does not change over time as tags are released. So like 2.6-SNAPSHOT or just 2.6.x. But not 2.6.1-SNAPSHOT as that would later change to 2.6.2-SNAPSHOT, then to 2.6.3-SNAPSHOT and so on.

Other projects (eg Apache Wicket) use a branch <version> similar to this. Tagged releases are like 1.3.3, 1.3.4, 1.3.5, just like us, and there is one 1.3.x branch with it's version at 1.3-SNAPSHOT. This branch version is stable and as fixes go into the branch, a new version is tagged, 1.3.6, but the branch remains at 1.3-SNAPSHOT as it's still the same singular 1.3 branch. Trunk is the only moving version which would be at 1.4-SNAPSHOT in this example.

If we have a changing branch <version>, it's going to mean a lot more manual intervention in removing deployed artifacts from the previous 'branch' (ie as it changes from 2.6.1-SNAPSHOT to 2.6.2-SNAPSHOT). So you couldn't just do an svn update in a branch, build and be on your way as the version might have changed. One of the main requirements behind the current maintenance branches is that they remain very stable.

There is currently no undeploy goal in our build process like there was in 2.4.x which would clean up an old version. Perhaps we need to look at this again (http://bugs.sakaiproject.org/jira/browse/SAK-13280).

Also, when did we shift to suggesting people use point releases rather than the maintenance branch in production?

regards,

Steve

On Thu, Mar 26, 2009 at 6:04 PM, Anthony Whyte <arwhyte@...> wrote:

Opening this conversation up to the dev list for further comments:

Steve--well, in a world were we could use the Maven release plugin

with the whole of Sakai (which does not exist at present; although I

think we can sort out the problem with some project/pom naming

realignments) we could perform releases from the 2.6.x branch as we

do now from K1.  In such a case the release plugin would generate a

2.6.0 tag and then the plugin would increment the pom <version>

number of the 2.6.x branch to 2.6.1-SNAPSHOT and commit the changes

automatically.  Then 2.6.0 artifacts are created and placed in the

repo.  This is how K1 <versioning> works and I expect Ian intends for

K2 to work the same way.   All of this you know so I apologize here

for stating the obvious.

The point I am trying to make above is that the maintenance branch

should be viewed as a SNAPSHOT set of code that by definition is

rather more fluid in nature than a point release (using M2 as a fixed

version number as you recommend obscures this).  Indeed, it is no

longer the case that we (the Foundation) actively advise people to

run their production instances off a maintenance branch.  Our goal

has been to undercut the old adage that friends don't let friends run

Sakai point releases in production by producing reliable maintenance

releases that are produced regularly to a well understood timeline

(the latter still a goal).  We have had a modicum of success here

with the 2.5 maintenance series as I see now that a fair number of

schools are running 2.5.2, 2.5.3 and 2.5.4 in production.  I

recognize that more experienced production houses tend to run off the

maintenance branch but over time I expect this to become the

exception rather than the rule given the number of smaller

institutions that run (and will run) point releases of Sakai.

From my perspective, I think consistency in our versioning practices

is important and I believe the "Maven" practice first adopted by Ian

works well.

trunk:  [major.minor]-SNAPSHOT

release tag: [major.minor.revision]

1.0.x branch [major.minor.revision]-SNAPSHOT

This is an area were I believe it would be good to settle on a

general practice since there may be advantages to the community of

having a few other core projects adopt their own release cycles

independent of a general Sakai release.  Our practices are a bit

inconsistent at present as a few examples will demonstrate:

Examples:

Sakai (after 2.6.0 release)

trunk:  [major.minor.revision]-SNAPSHOT  (e.g., currently 2.7.0-

SNAPSHOT, IMHO should simply be 2.7-SNAPSHOT)

tag: [major.minor.revision] (e.g. 2.6.0)

2.6.x branch [major.minor.revision]-SNAPSHOT (e.g., 2.6.1-SNAPSHOT)

K1 (after 1.0.4 release)

trunk:  [major.minor]-SNAPSHOT  (e.g., 1.1-SNAPSHOT)

release tag: [major.minor.revision] (e.g. 1.0.4)

1.0.x branch [major.minor.revision]-SNAPSHOT (e.g., 1.0.5-SNAPSHOT)

K2 (current)

trunk:  [major.minor]-SNAPSHOT  (e.g., 0.1-SNAPSHOT)

release tag: [major.minor.revision] (no tag yet)

branch [major.minor.revision]-SNAPSHOT (no branch yet)

SiteStates (current)

trunk:  [major.minor]-SNAPSHOT  (e.g., 2.0-SNAPSHOT)

release tag: [major.minor.revision] (e.g., 1.2.1)

branch (no 2.6 branch yet; I assume this would be 1.2.2-SNAPSHOT)

EntityBroker (current)

trunk:  [major.minor.revision]-SNAPSHOT  (e.g., 1.3.7-SNAPSHOT, IHMO

should simply be 1.3-SNAPSHOT)

release tag: [major.minor.revision] (e.g., 1.3.6)

2.6.x branch [major.minor.revision]-SNAPSHOT (currently, 1.3.6-

SNAPSHOT, IHMO should be 1.3.7-SNAPSHOT)

Cheers,

Anth

On Mar 26, 2009, at 12:20 PM, Steve Swinsburg wrote:

My only worry with this is is that the number will change, rather

than be stable like the 2.5.x series of M2. So then someone doing a

simple SVN update of just one module perhaps will get an updated

POM which doesn't match the rest of their dependencies.

My feeling is that the branch version number should be more stable

since we advise people to run it in production?

Hmm,

Steve

On 26 Mar 2009, at 16:13, Anthony Whyte wrote:

Currently, 2.6.x poms have a version of 2.6.0RC1-SNAPSHOT (it

really should have just been 2.6.0-SNAPSHOT).  Steve has enquired

what the <version> for the *x branch will be after the release of

2.6.0 (the release to occur from a 2.6.0 branch that I will create

when we do the first RC tag).

My recommendation is:  2.6.1-SNAPSHOT, the revision number to be

incremented by +1 whenever we do a maintenance release (e.g. 2.6.2-

SNAPSHOT, etc.).

Any objections?

Cheers,

Anthony

_______________________________________________

sakai-dev mailing list

sakai-dev@...

http://collab.sakaiproject.org/mailman/listinfo/sakai-dev

TO UNSUBSCRIBE: send email to sakai-dev-unsubscribe@... with a subject of "unsubscribe"

--

Aaron Zeckoski (aaronz@...)

Senior Research Engineer - CARET - Cambridge University

[http://bugs.sakaiproject.org/confluence/display/~aaronz/]

Sakai Fellow - [http://aaronz-sakai.blogspot.com/]

_______________________________________________ sakai-dev mailing list sakai-dev@... http://collab.sakaiproject.org/mailman/listinfo/sakai-dev TO UNSUBSCRIBE: send email to sakai-dev-unsubscribe@... with a subject of "unsubscribe"