On Jun 4, 2007, at 6:58 PM, Patrick Linskey wrote:
To date, OpenJPA has done minimal repackaging (currently only the EDU classes). I'm wary of changing this policy, especially considering how few classes we repackage. In fact, I think that not repackaging the backport classes is a good thing, as it lets people easily plug in the faster Java 5 version without having to then re-repackage those classes and recompile them. Any other opinions?
I'm also allergic to repackaging (re-namespacing). I agree that it's inconvenient if the dependency changes its interfaces but that's usually in a major release.
[If interfaces change in an incompatible way on most minor releases, as in your ehcache example, I'd think twice about using the package at all. Not brilliant change management IMHO.]
Craig
-Patrick On 6/4/07, Brian McCallister <[EMAIL PROTECTED]> wrote:On Jun 4, 2007, at 6:39 PM, Patrick Linskey wrote: > I'm allergic to re-namespacing... why do you think that we should > do so? Avoiding collisions. The majority case is that people don't care about the extra 327k but they care a lot about not hitting conflicts with libraries. Dug Lea's libraries are not a great example of this, but Hibernate *is* a good example -- it relies on EHCache by default, going from 3.0 to 3.1 to 3.2 is a major non-backwards compatible change, and you cannot use EHCache 1.2 with 3.0, so you are trapped unable to upgrade dependencies. Weblogic and ANTLR (a couple versions back) is another great example. Basically, if you are a library (which OpenJPA is) you want to minimize the degree to which you place constraints on the runtime environment of your users. I can easily imagine someone using a home- rolled build of the concurrent backport which was subtly incompatible. Yes, your user could renamespace then, but it is better if they *never have the issue* -Brian > > -Patrick > > On 6/4/07, Brian McCallister <[EMAIL PROTECTED]> wrote: >> I would suggest using backports and repackaging -- though I have>> trouble imaging the interfaces on backports changing. I, personally, >> am of the opinion that if at all possible, small dependencies should>> be re-namespaced and bundled. >> >> -Brian >> >> On Jun 4, 2007, at 4:22 PM, Patrick Linskey wrote: >> >> > Hi, >> > >> > In the process of doing some concurrency-related work on >> OpenJPA, I've >> > run across the need for a ReentrantReadWriteLock, akin to what >> is in >> > Java 5's java.util.concurrent package, Emory University's>> > edu.emory.mathcs.backport package, and Doug Lea's EDU.oswego.cs.dl>> > package. >> >>> > Currently, OpenJPA has repackaged copies of some of the code from>> > EDU.oswego.cs.dl, but not everything. I'd like to get rid of the >> > repackaged copies, and move to the versions in >> > edu.emory.mathcs.backport. According to Doug Lea's website, the>> > backport classes are preferable to the EDU.oswego.cs.dl classes at>> > this point. >> > >> > This change is independent of future changes to allow for >> pluggability >> > of the concurrent implementation, and only impacts those classes >> that >> > we are already directly repackaging. >> > >> > Thoughts? >> > >> > -Patrick >> > >> > -- >> > Patrick Linskey >> > 202 669 5907 >> >> > > > -- > Patrick Linskey > 202 669 5907-- Patrick Linskey 202 669 5907
Craig Russell Architect, Sun Java Enterprise System http://java.sun.com/products/jdo 408 276-5638 mailto:[EMAIL PROTECTED] P.S. A good JDO? O, Gasp!
smime.p7s
Description: S/MIME cryptographic signature
