I think it would make life more difficult managing so many versions
and the compatibility between those version, however I think this
would be beneifical for updating the server at a more granulized level.
Another approach is to break out the server into components (multiple
modules making up a component identified by a groupId) and each group
Id could have its own version.
On Jun 15, 2006, at 12:34 AM, Matt Hogstrom wrote:
Not sure if this is already captured.
What do folks think about leaving the modules as independent pieces
with their own version numbers and the geronimo_version is just the
aggregate release to users? I expect this would make out life more
difficult but I haven't found the single version number to rule all
modules all that easy either.
Also, it would be nice that if a module hadn't changed then it
stays static and is a good indicator of where the activity is.
Thoughts?
Hiram Chirino wrote:
On 6/11/06, Alan D. Cabrera <[EMAIL PROTECTED]> wrote:
X.Y.z: patch release - bug fixes
X.y: minor release - bug fixes and compatible enhancements
x: major release - major enhancements, and incompatible
changes
I am very much against placing anything but patches in the .z
releases.
Let me explain why. When we make a minor release we basically
branch
the X.y release for the purposes of releasing patches. Any
changes to
this branch, e.g. patches, must also be immediately applied to the
trunk. If we make any enhancements to this branch, we must also
immediately apply the same enhancement to that branch. This adds
significant more risk that bug patching but more importantly when we
fall into the trap of putting minor enhancements into a patch
release,
we remove the most serious impetus to getting the minor release
done and
out the door.
+1. This allows us to time box the bug fix releases. If we can get
into the groove of doing regular x.y.z releases (at like 1 a month
intervals), then I think that also reduces the pressure on needing to
make the x.y releases perfect. I think we sometimes delay our x.y
releases because we are aiming for perfection.
The only problem with the above is that it does not solve the problem
of being able to time box the x.y release. The since dev branch of
the x.y release could have multiple new features at different levels
of completion it's hard to stabilize at any given time. Do you guys
consider this a problem?
I like Dain's suggestion of splitting up the modules. In theory in
progress work being done separately versioned project should not hold
up the time boxed release of a Geronimo x.y. Geronimo would just
release with the previous stable version. In practice, even for
independently versioned projects like ActiveMQ, Geronimo will hold up
it's releases to get new releases from ActiveMQ. This is bad if you
want to time box a release.
Another thought that might help Geronimo be able to stay on a time
box
release cycle is making more use of 'development' branches. We could
encourage develops to work on new features in development branches
that get merged in once the feature is fully working. The down side
to this is that it may not be obvious to other developers what
work is
going on where.
Or perhaps we need to do a a combination of independent versioned
modules where most of the work happens, and then having small
development branches of the main Geronimo module that holds the
integration code that enables the new features. So then then
development branches are used to do integration testing with in
progress features and they are merged in to trunk once the feature is
done and all integration testing is completed.
-sachin
