Agreed. This is a change that’s fine to include when editing related (and new) code, but doesn’t come close to warranting a wide scale change. On 9 May 2025, at 18:32, Blake Eggleston <bl...@ultrablake.com> wrote:
This seems like a cool feature that will be useful in future development work, but not something we should be proactively refactoring the project to make use of.
On Fri, May 9, 2025, at 10:18 AM, Vivekanand Koya wrote: I would say that https://openjdk.org/jeps/394 (instanceOf) aims to
provide safer and less poisoning in the code by default. Instead of
having a production server halt/impaired due to a RuntimeException, instead it is verified at compile time. If a new language feature makes code more robust and addresses a hazardous, historical design choice, I believe it's time has arrived. Curious to see what everyone thinks.
Thanks, Vivekanand K.
I would like to refactor the codebase (Trunk 5+) to eliminate unsafe explicit casting with instanceOf.
We have a rich history of broad sweeping refactors dying on the rocks of the community's aversion to instability and risk w/out a concrete outcome we're trying to achieve. :)
All of which is to say: do we have examples of instanceOf casting blowing things up for users that would warrant going through the codebase to tidy this up? Between src/java and test/unit and test/distributed we have around 2,000 occurrences of this pattern.
On Fri, May 9, 2025, at 10:14 AM, Vivekanand Koya wrote: Sounds great. I would like to refactor the codebase (Trunk 5+) to eliminate unsafe explicit casting with instanceOf.
Thanks, Vivekanand
Yep, that approach seems more than sufficient to me. No need for lots of ceremony, but good to keep everyone in the decision loop.
I think it doesn’t cost us much to briefly discuss new language features before using them.
I had that thought as well but on balance my intuition was there were enough new features that the volume of discussion to do that would be a poor cost/benefit compared to the "lazy consensus, revert" approach.
So if I actually do the work required to have an opinion ;):
JDK21: JDK17: JDK16: JDK15: JDK14: JDK11:
Assuming we just lazily evaluate and deal with new features as people actually care about them and seeing them add value, a simple "[DISCUSS] I'm thinking about using new language feature X; any objection?" lazy consensus that we then dumped onto a wiki article / code style page as "stuff we're good to use" would probably be fine?
On Fri, May 9, 2025, at 7:58 AM, Benedict wrote:
I think it doesn’t cost us much to briefly discuss new language features before using them. Lambdas, Streams and var all have problems - and even with the guidance we publish some are still misused.
The flow scoping improvement to instanceof seems obviously good though. For new feature work on trunk, targeting the highest supported language level featureset (jdk17 right now, jdk21 within the next couple of weeks) makes sense to me. For bugfixing, targeting the oldest supported GA branch and the highest language level that works there would allow maximum flexibility with minimal re-implementation.
If anyone has any misgivings with certain features (i.e. the debate around usage of "var") they can bring it up on the dev ML and we can adjust, but otherwise I'd prefer to see us have more modern evolving options on how contributors engage rather than less.
On Fri, May 9, 2025, at 1:56 AM, Vivekanand Koya wrote: Hello,
I want to understand the community's thoughts on using newer features (post JDK11) in upcoming releases in Cassandra. An example is flow scoping instead of explicitly casting types with instanceOf: https://openjdk.org/jeps/395. I want your thoughts on JDK requirements for the main Cassandra repository, Accord, and Sidecar.
Much appreciated. Thanks, Vivekanand K.
|
