Github user parthchandra commented on the issue: https://github.com/apache/drill/pull/1144 To be honest, I don't know. There are too many paths available to write to direct memory, some which were developed over time and so not all vector code may be using them. Ideally, every bit of code should go thru DrillBuf which wraps a UDLE (UnsafeDirectLittleEndian). UDLE has one level of bounds check (which I feel is necessary) and uses PlatformDependent directly. Vector set safe methods provides vector level bounds checking and the vector set methods bypass bounds checking. ResultSetLoader methods provide guarantees that may make the set safe methods redundant. Our goal should be to not depend on the set safe methods except for debugging. What exactly are you thinking of when proposing using PlatformDependent directly.
---