On May 18, 2009, at 6:10 PM, Stephen Colebourne wrote:
As a minimum, major changes require a package name change, and as
has been pointed out a maven id change.
So:
(1) minor incompatible changes (remove long deprecated methods or
fix details at the edges). These are compatible for 99% of users and
can be a major version (or perhaps minor) with no package or maven
name change.
(2) major incompatible changes (moving classes between packages,
changing how the code is used, anything that causes the user to need
to reread docs, inability to use new version as a dependency for
something compiled using old version). These can be thought of as a
major version change with associated package name change and maven
name change.
(3) major reboot (same problem space, but lots of rewrite), I'd
still strongly prefer to see these in the sandbox (eg cli2).
Socially, they are often a different set of committers with a subtly
different set of design ideas (all good things). Although the
sandbox approach is preferable, my minimum requirement however is a
major version change with package name change and maven name change.
Hopefully, each set of changes can be classified into one of those
three groups. However, IMO [collections] and [cli] are both cases
that would have been better off taking the sandbox route.
Personally I'm surprised that the sandbox route isn't more popular
as it allows proper innovation with no limits (ie. no one like me
looking over the shoulder...). Maybe, its a fear of difficulty in
getting it promoted. Maybe its a desire to retain the brand name of
the original. Maybe a separate thread should identify the fears.
What is going on with commons configuration is a great example.
experimental-configuration2 won't be completely API compatible but
will be somewhat. It will require and take advantage of Java 5
features. It isn't clear yet whether this will be a 2) or a 3)
(probably somewhere in between). What is clear is that any time a
patch is made to trunk the same patch is also applied, when possible,
to the experimental branch. As unit tests are added to trunk they are
also added to the experimental branch if they can be. If this work was
going on in sandbox it would, IMO, be far more likely that the sandbox
would get out of synch with the "production" version. And why bother
working in sandbox when we can just collaborate on a branch?
Ralph
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org
