I will check you trick ad soon as possible. I am sorry, this is a very busy week for me and do not have enough cycles. But I think that we should address this problem in order to ease the adoption of the new code and APIs.
Did you evaluate to eventually rollback the breaking changes? Another alternative, if we want to let users use both the old and the new APIs is to simply rename all of the packages and start a brand new system. This approach was done in Apache Commons and IIRC it will be done with Netty5. We also did it with the new Apache Bookkeeper API. Pros: No need to preserve compatibility, we are free to clean up all of the tech debt. The switch to Curator 5 will be explicit opted in Cons: Cherry picks won't be straightforward. Enrico Il Ven 22 Mag 2020, 23:40 Jordan Zimmerman <jor...@jordanzimmerman.com> ha scritto: > Hi Everyone, > > I've coded a possible solution in the test project. See here: > > https://github.com/Randgalt/curator_5_0_test/blob/master/combo/pom.xml#L49 > > It uses the Maven dependency plugin to create a small compatibility JAR > that contains the Curator 4.3.0 versions of the classes that have changed > in 5.0.0 (i.e. the ones that no longer return ListenerContainer). If this > JAR is included in a CLASSPATH before Curator 5.0.0's JARs, these old > classes will take precedence and thus old binaries will continue to work. > The curator_5_0_test shows this. run.sh is the previous way with the > error. run-compatibility.sh is with the compatibility JAR. > > Thoughts? Notable, this doesn't change the master code of Curator at all. > We could add it to the 5.0 release. I don't think there's an issue with > this "hack". Can anyone think of one? I'd really appreciate people testing > with it. Try a build with just Curator 5.0 and then install and include > this curator-5_0-test:combo:1.0-SNAPSHOT early in the CLASSPATH - it should > work. > > -Jordan > > On May 21, 2020, at 10:43 AM, Jordan Zimmerman <jor...@jordanzimmerman.com> > wrote: > > Hello All, > > Sorry for the cross-posting but this is important enough to justify it. > > Apache Curator is in the process of releasing version 5.0. We've taken the > opportunity to address some long standing tech debt but this causes > breaking changes. We've detailed the breaks here: > http://curator.apache.org/staging/breaking-changes.html. The Clirr report > shows the exact API changes: > http://curator.apache.org/staging/curator-recipes/clirr-report.html. The > first two of these are the most worrisome. NodeCache's and > PathChildrenCache's getListenable() methods now have a different return > type. This has far reaching implications. If a Curator user were to drop in > Curator 5.0 without any code changes they will get runtime exceptions when > these methods are called. > > I've written a test that shows the problem: > > git clone https://github.com/Randgalt/curator_5_0_test.git > cd curator_5_0_test > ./run.sh > > You will see: > > java.lang.NoSuchMethodError: > org.apache.curator.framework.recipes.cache.PathChildrenCache.getListenable()Lorg/apache/curator/framework/listen/ListenerContainer; > at binary.Curator50Test.run(Curator50Test.java:26) > at test.Test.main(Test.java:9) > at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method) > at > sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62) > at > sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43) > at java.lang.reflect.Method.invoke(Method.java:498) > at org.codehaus.mojo.exec.ExecJavaMojo$1.run(ExecJavaMojo.java:297) > at java.lang.Thread.run(Thread.java:748) > > Enrico Olivelli brought this to our attention. Curator 5.0 is a major > version bump so breaking changes are implied. But, maybe this is blocker? > What do people think? If this is a serious enough concern we can come up > with a workaround. > > Please discuss and let's hold off completing the current release until > this has been fully discussed. > > -Jordan > > >
