If we have a dependency on a library that’s not in Maven Central, then using River in a Maven-based project (for example, the River Examples project) will effectively be impossible (it can be done but is a royal nuisance for downstream users).
As such, I’d vote against any release that has a dependency on a library that’s not in Maven Central. If Peter’s not able to put custard-apple into Maven Central, then I think we have no choice but to accept it as contributed code and roll it into River. A third option would be to accept it as a contributed code, but treat it as a separate deliverable and publish it on its own into Maven Central. The only caveat is that we would have to rename the packages to ‘org.apache.river….’, since we can only publish under the ‘org.apache’ and ‘net.jini’ group ids. Peter (or anyone here) could maintain it under River’s aegis, and other users could still get it out of Maven Central. Cheers, Greg > On Aug 12, 2015, at 3:24 PM, Patricia Shanahan <p...@acm.org> wrote: > > Thanks, Peter. > > It appears to me that we have three options for dealing with "custard apple": > > 1. Do not include it in the release, but include a download link in the > installation instructions. > > Pro: easy for us. Con: more work for users. > > 2. Include a selected source subset that River currently needs. > > Pro: minimize source code volume. Con: more complicated maintenance - if > Peter makes a change, we would need to check whether it involves the subset. > > 3. Include the whole library source. > > Pro: simpler maintenance - if Peter makes a change, we just copy in the > change. Con: Distributing source code that is not used by River. > > Any other options we should consider? Other pros or cons I have not thought > of? Opinions? > > On 8/12/2015 5:34 AM, Peter wrote: >> It's AL2 licensed, the source is in a jar file in dep-libs, consider it >> already contributed. >> >> The library contains more code than you need, as it covers every type of >> Java Collections interface, up to Java 7 (it will be updated at some >> point to support Java 8 & 9 collections interfaces also). >> >> The code itself is scalable, it has a single JVM thread that performs >> garbage collection of references, api calling threads don't perform this >> task. The references themselves are often created and destroyed without >> being shared among other threads, a large number never leave CPU cache >> and safe publication is used to avoid synchronization. Overhead is low >> and its public API is compact. >> >> For the most part use of soft references is advised against, however >> there is a case for a cache that degrades gracefully under load in the >> form of a Map where keys are timed and values softly reachable (or vice >> versa), to avoid extreme cases where a JVM is running out of memory, the >> jvm gc can clear entries that aren't strongly reachable (even though >> timed keys haven't expired) at its descretion potentially avoiding or at >> least delaying an OOME. Timed references will prune infrequently used >> cache values, even though they're still strongly reachable. >> >> Unlike other cache libraries, custard apple only implements reference >> capabilities, allowing the user to wrap this over their preferred >> collection, such as ConcurrentHashMap, ConcurrentSkipListMap, >> NonBlockingHashMap etc, it reduces code duplication and allows the user >> to benefit from new performance improvements in popular collections. >> >> Regards, >> >> Peter. >> >> On 12/08/2015 8:57 PM, Patricia Shanahan wrote: >>> Does it have a license that lets us do that? >>> >>> (If you are the writer, and copy it in yourself, it would be covered >>> by your ICLA, just like any other code you contribute to Apache.) >>> >>> On 8/12/2015 12:00 AM, Peter wrote: >>> ... >>>> I'm a little busy right now to consider moving custard apple. If you >>>> wan't you can always copy only the code in use into >>>> org.apache.river.impl >>> ... >>> >>
