zhztheplayer commented on PR #13248: URL: https://github.com/apache/arrow/pull/13248#issuecomment-1211731895
> I think what needs to happen is that C Data Interface allocations need to get associated with an actual allocator, even if the allocator isn't actually doing the allocation. This is how we used to handle native-allocated blocks in Java dataset codes before migrating to ABI: `NativeUnderlyingMemory`: https://github.com/apache/arrow/commit/d25660ed5f57b167878c96cafc23361421a73ca8#diff-177060b1e9fd73540eacd8d55a32a81702ddbc39b6f1f406702d345620fbd1ceR25 Although the codes binding it to allocator was more like a workaround: https://github.com/apache/arrow/commit/d25660ed5f57b167878c96cafc23361421a73ca8#diff-6f45bb0f8f4da15fddfe7027b1983fe16dda4332b1cba9a9f1a6f28a0fa66a26R101-R104 I am not sure if introducing a new API `BufferAllocator#allocate(MemoryChunk chunk)` would be more reasonable here. The API is designed and implemented in our own forked Arrow repo: https://github.com/oap-project/arrow/blob/c4bd7cc1a45991aedf8e4ddd5533d798361d24c7/java/memory/memory-core/src/main/java/org/apache/arrow/memory/BufferAllocator.java#L55-L62 `MemoryChunk` interface is simplified from `AllocationManager`: https://github.com/oap-project/arrow/blob/c4bd7cc1a45991aedf8e4ddd5533d798361d24c7/java/memory/memory-core/src/main/java/org/apache/arrow/memory/MemoryChunk.java#L20-L42 -- 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]
