(
lhs: &ArrayData,
rhs: &ArrayData,
lhs_start: usize,
rhs_start: usize,
len: usize,
)
| 44 | } |
| 45 | |
| 46 | pub(super) fn list_equal<T: ArrowNativeType + Integer>( |
| 47 | lhs: &ArrayData, |
| 48 | rhs: &ArrayData, |
| 49 | lhs_start: usize, |
| 50 | rhs_start: usize, |
| 51 | len: usize, |
| 52 | ) -> bool { |
| 53 | let lhs_offsets = lhs.buffer::<T>(0); |
| 54 | let rhs_offsets = rhs.buffer::<T>(0); |
| 55 | |
| 56 | // There is an edge-case where a n-length list that has 0 children, results in panics. |
| 57 | // For example; an array with offsets [0, 0, 0, 0, 0] has 4 slots, but will have |
| 58 | // no valid children. |
| 59 | // Under logical equality, the child null bitmap will be an empty buffer, as there are |
| 60 | // no child values. This causes panics when trying to count set bits. |
| 61 | // |
| 62 | // We caught this by chance from an accidental test-case, but due to the nature of this |
| 63 | // crash only occurring on list equality checks, we are adding a check here, instead of |
| 64 | // on the buffer/bitmap utilities, as a length check would incur a penalty for almost all |
| 65 | // other use-cases. |
| 66 | // |
| 67 | // The solution is to check the number of child values from offsets, and return `true` if |
| 68 | // they = 0. Empty arrays are equal, so this is correct. |
| 69 | // |
| 70 | // It's unlikely that one would create a n-length list array with no values, where n > 0, |
| 71 | // however, one is more likely to slice into a list array and get a region that has 0 |
| 72 | // child values. |
| 73 | // The test that triggered this behaviour had [4, 4] as a slice of 1 value slot. |
| 74 | // For the edge case that zero length list arrays are always equal. |
| 75 | if len == 0 { |
| 76 | return true; |
| 77 | } |
| 78 | |
| 79 | let lhs_child_length = lhs_offsets[lhs_start + len].to_usize().unwrap() |
| 80 | - lhs_offsets[lhs_start].to_usize().unwrap(); |
| 81 | |
| 82 | let rhs_child_length = rhs_offsets[rhs_start + len].to_usize().unwrap() |
| 83 | - rhs_offsets[rhs_start].to_usize().unwrap(); |
| 84 | |
| 85 | if lhs_child_length == 0 && lhs_child_length == rhs_child_length { |
| 86 | return true; |
| 87 | } |
| 88 | |
| 89 | let lhs_values = &lhs.child_data()[0]; |
| 90 | let rhs_values = &rhs.child_data()[0]; |
| 91 | |
| 92 | let lhs_null_count = count_nulls(lhs.nulls(), lhs_start, len); |
| 93 | let rhs_null_count = count_nulls(rhs.nulls(), rhs_start, len); |
| 94 | |
| 95 | if lhs_null_count != rhs_null_count { |
| 96 | return false; |
| 97 | } |
| 98 | |
| 99 | if lhs_null_count == 0 && rhs_null_count == 0 { |
| 100 | lhs_child_length == rhs_child_length |
| 101 | && lengths_equal( |
| 102 | &lhs_offsets[lhs_start..lhs_start + len], |
| 103 | &rhs_offsets[rhs_start..rhs_start + len], |
nothing calls this directly
no test coverage detected