Share null mask with constant null arg vector #17234
Conversation
| @@ -43,6 +43,9 @@ static void ConstantOrNullFunction(DataChunk &args, ExpressionState &state, Vect | |||
| if (ConstantVector::IsNull(args.data[idx])) { | |||
| // input is constant null, return constant null | |||
| result.Reference(info.value); | |||
| auto &result_mask = ConstantVector::Validity(result); | |||
There was a problem hiding this comment.
Can we remove the reference altogether, and just set the result to a constant NULL? (i.e. result.SetVectorType(VectorType::CONSTANT_VECTOR);)
There was a problem hiding this comment.
Thanks. I've updated the code.
There was a problem hiding this comment.
The Initialize should not also not be necessary anymore now, i.e. we just need the two lines (SetVectorType and ConstantVector::SetNull).
There was a problem hiding this comment.
The constant_or_null function may accept three or more inputs (e.g., constant_or_null(true, c1, c2)).
In cases where: c1 is a flat vector, and c2 is a constant vector :
we first access c1 and share its mask with the result through result_mask.Combine(input_mask).
Later, when we access c2 and find it's a constant null. If we simply call SetNull at this stage, it would corrupt c1's mask (since we're still sharing c1's mask state from the earlier Combine operation).
So we should call Initialize to switch the mask reference to c2's mask (the constant null vector) before calling SetNull.And c2 will not be affected because it's already null.
There was a problem hiding this comment.
I see - makes sense. Thanks!
|
Thanks! |
Share null mask with constant null arg vector (duckdb/duckdb#17234)
Share null mask with constant null arg vector (duckdb/duckdb#17234)
Share null mask with constant null arg vector (duckdb/duckdb#17234)
Share null mask with constant null arg vector (duckdb/duckdb#17234)
Fixes #17217
When the constant_or_null function encounters a constant NULL input, it will set the null mask of result vector for row 0 to invalid.
If the null mask is shared with other function arg vector (due to a previous Combine() call), modifying the null mask for row 0 would incorrectly affect the arg vector. This leads to wrong query result.
Share the null mask with the constant NULL arg vector should fix the issue.