Hello GraphAr community, I am writing about the Java side of GraphAr. As far as I can tell there is no usable, up-to-date Java library right now: the existing Java code is the FastFFI binding, which has not kept up and is not something a JVM project can depend on today. I would like to help fix that, and I am ready to put serious and sustained work into it.
I have a working prototype of a pure-Java implementation in a fork. It covers immutable metadata, a dependency-light storage API with local and S3 adapters, a physical IO contract, a Parquet projection and row-range backend, an ordered-by-source reader, and a writer with a validator. It is tested and measured against the C++ implementation and the canonical fixtures, which I treat as the format oracle. Following the contribution guide, I opened issues first and then narrow PRs: #943, #944, #947, #953, #955 for the Java work #946 as the first small implementation PR Where things stand: SemyonSinchenko reviewed #946 and I addressed the comments, but the PR has not moved since. The larger Java issues (#947, #944, #953, #955) have had no replies at all. PR #952 has had its CI workflows in "action_required" since 11 August, so it cannot even get a build. I am not asking anyone to merge things quickly. I understand that a project has a process and that a contribution of this size needs agreement on direction before code. So my questions are: 1. Is a pure-Java implementation something the project wants? If yes, where should it live, and should it replace the FastFFI binding or sit next to it? 2. What do you need from me to make this reviewable: a design document, a thread on this list, a smaller first slice, something else? 3. Is there someone willing to be the committer who looks at the Java PRs? Without that, the queue will keep growing and I would rather not add to it blindly. Happy to start from scratch on the process side if that is what it takes. I just want the Java library to exist. Thanks, Alex (github: keksmd) --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
