Abhishek-Mohan commented on issue #4419: URL: https://github.com/apache/datafusion-comet/issues/4419#issuecomment-6359786570
Thanks for writing this up. I've taken this and plan to land it as a style guide page (#7151), a terminology pass over the remaining pages (#7152), and the #4420 skill (#7153). While checking the proposal against the current code I found a few places where I'd like to adjust the vocabulary before it becomes a rule, and I'd welcome your view: 1. **Hybrid** already appears in `expressions.md` (from `GenerateDocs`) with a different meaning: an expression that has both a Rust path and a codegen-dispatch path. I'd keep that meaning and not reuse the word for JVM kernels calling Rust. 2. **Expressions are not plan nodes.** Codegen-dispatched expressions and UDFs run inside `CometProject` and similar operators. I'd keep the three node kinds from #4650 and describe the expression kinds (Native, Codegen dispatch, Hybrid) separately, instead of using a four-category node table. 3. **Vectorized:** Spark's vectorized reader means batch-at-a-time decoding rather than SIMD. I'd allow the word in proper names and avoid using it as a synonym for columnar. 4. **Arrow-native:** I'd use it only for claims about data format. The codegen kernels and JVM shuffle still do per-row work. 5. **Contributor docs:** I'd keep "native side", "native plan" and "native code" for the `native/` crates and the JNI boundary. 6. **Operator renames:** these change plan output and golden files, so I'd suggest splitting them into a separate issue. #7152 would then say "Part of #4419", and this issue would close when the docs work is done. AI assistance: OpenCode helped check the proposal against the code. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
