Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix Null Mask Handling in ArrayData, UnionArray, and MapArray #1589

Merged
merged 5 commits into from
Apr 28, 2022
Merged

Fix Null Mask Handling in ArrayData, UnionArray, and MapArray #1589

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
ecf753b
Fix ListArray and StructArray equality (#626)
tustvold Apr 19, 2022
95b2ed3
Simplify null masking in equality comparisons
tustvold Apr 20, 2022
feda7b6
Miscellaneous fixes
tustvold Apr 20, 2022
4aff5fe
Fix structure null equality
tustvold Apr 23, 2022
f67b650
Review feedback
tustvold Apr 23, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
72 changes: 35 additions & 37 deletions arrow/src/array/array_union.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,7 +61,6 @@ use std::any::Any;
/// type_id_buffer,
/// Some(value_offsets_buffer),
/// children,
/// None,
/// ).unwrap();
///
/// let value = array.value(0).as_any().downcast_ref::<Int32Array>().unwrap().value(0);
Expand Down Expand Up @@ -94,7 +93,6 @@ use std::any::Any;
/// type_id_buffer,
/// None,
/// children,
/// None,
/// ).unwrap();
///
/// let value = array.value(0).as_any().downcast_ref::<Int32Array>().unwrap().value(0);
Expand Down Expand Up @@ -140,7 +138,6 @@ impl UnionArray {
type_ids: Buffer,
value_offsets: Option<Buffer>,
child_arrays: Vec<(Field, ArrayRef)>,
bitmap_data: Option<Buffer>,
) -> Self {
let (field_types, field_values): (Vec<_>, Vec<_>) =
child_arrays.into_iter().unzip();
Expand All @@ -152,13 +149,11 @@ impl UnionArray {
UnionMode::Sparse
};

let mut builder = ArrayData::builder(DataType::Union(field_types, mode))
let builder = ArrayData::builder(DataType::Union(field_types, mode))
.add_buffer(type_ids)
.child_data(field_values.into_iter().map(|a| a.data().clone()).collect())
.len(len);
if let Some(bitmap) = bitmap_data {
builder = builder.null_bit_buffer(bitmap)
}

let data = match value_offsets {
Some(b) => builder.add_buffer(b).build_unchecked(),
None => builder.build_unchecked(),
Expand All @@ -171,7 +166,6 @@ impl UnionArray {
type_ids: Buffer,
value_offsets: Option<Buffer>,
child_arrays: Vec<(Field, ArrayRef)>,
bitmap: Option<Buffer>,
) -> Result<Self> {
if let Some(b) = &value_offsets {
if ((type_ids.len()) * 4) != b.len() {
Expand Down Expand Up @@ -216,7 +210,7 @@ impl UnionArray {
// Unsafe Justification: arguments were validated above (and
// re-revalidated as part of data().validate() below)
let new_self =
unsafe { Self::new_unchecked(type_ids, value_offsets, child_arrays, bitmap) };
unsafe { Self::new_unchecked(type_ids, value_offsets, child_arrays) };
new_self.data().validate()?;

Ok(new_self)
Expand Down Expand Up @@ -512,7 +506,7 @@ mod tests {
builder.append::<Int32Type>("a", 1).unwrap();
builder.append::<Int64Type>("c", 3).unwrap();
builder.append::<Int32Type>("a", 10).unwrap();
builder.append_null().unwrap();
builder.append_null::<Int32Type>("a").unwrap();
builder.append::<Int32Type>("a", 6).unwrap();
let union = builder.build().unwrap();

Expand All @@ -522,29 +516,29 @@ mod tests {
match i {
0 => {
let slot = slot.as_any().downcast_ref::<Int32Array>().unwrap();
assert!(!union.is_null(i));
assert!(!slot.is_null(0));
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think it is good to check null in the current slot, but does union.is_null(index) not work anymore? Can we check both?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The nullability is only a property of the slot, and not the Union itself, at least that is my interpretation of the specification

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

https://arrow.apache.org/docs/format/Columnar.html#union-layout

Says

Unlike other data types, unions do not have their own validity bitmap. Instead, the nullness of each slot is determined exclusively by the child arrays which are composed to create the union.

Which seems consistent 👍

However, it seems like UnionArray doesn't override is_null to take this into account. Filed #1625 to track

assert_eq!(slot.len(), 1);
let value = slot.value(0);
assert_eq!(1_i32, value);
}
1 => {
let slot = slot.as_any().downcast_ref::<Int64Array>().unwrap();
assert!(!union.is_null(i));
assert!(!slot.is_null(0));
assert_eq!(slot.len(), 1);
let value = slot.value(0);
assert_eq!(3_i64, value);
}
2 => {
let slot = slot.as_any().downcast_ref::<Int32Array>().unwrap();
assert!(!union.is_null(i));
assert!(!slot.is_null(0));
assert_eq!(slot.len(), 1);
let value = slot.value(0);
assert_eq!(10_i32, value);
}
3 => assert!(union.is_null(i)),
3 => assert!(slot.is_null(0)),
4 => {
let slot = slot.as_any().downcast_ref::<Int32Array>().unwrap();
assert!(!union.is_null(i));
assert!(!slot.is_null(0));
assert_eq!(slot.len(), 1);
let value = slot.value(0);
assert_eq!(6_i32, value);
Expand All @@ -560,7 +554,7 @@ mod tests {
builder.append::<Int32Type>("a", 1).unwrap();
builder.append::<Int64Type>("c", 3).unwrap();
builder.append::<Int32Type>("a", 10).unwrap();
builder.append_null().unwrap();
builder.append_null::<Int32Type>("a").unwrap();
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

it is strange to require a field name to append nulls -- I left a comment on how to improve things perhaps in the future

builder.append::<Int32Type>("a", 6).unwrap();
let union = builder.build().unwrap();

Expand All @@ -573,15 +567,15 @@ mod tests {
match i {
0 => {
let slot = slot.as_any().downcast_ref::<Int32Array>().unwrap();
assert!(!union.is_null(i));
assert!(!slot.is_null(0));
assert_eq!(slot.len(), 1);
let value = slot.value(0);
assert_eq!(10_i32, value);
}
1 => assert!(new_union.is_null(i)),
1 => assert!(slot.is_null(0)),
2 => {
let slot = slot.as_any().downcast_ref::<Int32Array>().unwrap();
assert!(!union.is_null(i));
assert!(!slot.is_null(0));
assert_eq!(slot.len(), 1);
let value = slot.value(0);
assert_eq!(6_i32, value);
Expand Down Expand Up @@ -614,13 +608,9 @@ mod tests {
Arc::new(float_array),
),
];
let array = UnionArray::try_new(
type_id_buffer,
Some(value_offsets_buffer),
children,
None,
)
.unwrap();
let array =
UnionArray::try_new(type_id_buffer, Some(value_offsets_buffer), children)
.unwrap();

// Check type ids
assert_eq!(Buffer::from_slice_ref(&type_ids), array.data().buffers()[0]);
Expand Down Expand Up @@ -800,7 +790,7 @@ mod tests {
fn test_sparse_mixed_with_nulls() {
let mut builder = UnionBuilder::new_sparse(5);
builder.append::<Int32Type>("a", 1).unwrap();
builder.append_null().unwrap();
builder.append_null::<Int32Type>("a").unwrap();
builder.append::<Float64Type>("c", 3.0).unwrap();
builder.append::<Int32Type>("a", 4).unwrap();
let union = builder.build().unwrap();
Expand All @@ -824,22 +814,22 @@ mod tests {
match i {
0 => {
let slot = slot.as_any().downcast_ref::<Int32Array>().unwrap();
assert!(!union.is_null(i));
assert!(!slot.is_null(0));
assert_eq!(slot.len(), 1);
let value = slot.value(0);
assert_eq!(1_i32, value);
}
1 => assert!(union.is_null(i)),
1 => assert!(slot.is_null(0)),
2 => {
let slot = slot.as_any().downcast_ref::<Float64Array>().unwrap();
assert!(!union.is_null(i));
assert!(!slot.is_null(0));
assert_eq!(slot.len(), 1);
let value = slot.value(0);
assert_eq!(value, 3_f64);
}
3 => {
let slot = slot.as_any().downcast_ref::<Int32Array>().unwrap();
assert!(!union.is_null(i));
assert!(!slot.is_null(0));
assert_eq!(slot.len(), 1);
let value = slot.value(0);
assert_eq!(4_i32, value);
Expand All @@ -853,9 +843,9 @@ mod tests {
fn test_sparse_mixed_with_nulls_and_offset() {
let mut builder = UnionBuilder::new_sparse(5);
builder.append::<Int32Type>("a", 1).unwrap();
builder.append_null().unwrap();
builder.append_null::<Int32Type>("a").unwrap();
builder.append::<Float64Type>("c", 3.0).unwrap();
builder.append_null().unwrap();
builder.append_null::<Float64Type>("c").unwrap();
builder.append::<Int32Type>("a", 4).unwrap();
let union = builder.build().unwrap();

Expand All @@ -866,18 +856,18 @@ mod tests {
for i in 0..new_union.len() {
let slot = new_union.value(i);
match i {
0 => assert!(new_union.is_null(i)),
0 => assert!(slot.is_null(0)),
1 => {
let slot = slot.as_any().downcast_ref::<Float64Array>().unwrap();
assert!(!new_union.is_null(i));
assert!(!slot.is_null(0));
assert_eq!(slot.len(), 1);
let value = slot.value(0);
assert_eq!(value, 3_f64);
}
2 => assert!(new_union.is_null(i)),
2 => assert!(slot.is_null(0)),
3 => {
let slot = slot.as_any().downcast_ref::<Int32Array>().unwrap();
assert!(!new_union.is_null(i));
assert!(!slot.is_null(0));
assert_eq!(slot.len(), 1);
let value = slot.value(0);
assert_eq!(4_i32, value);
Expand All @@ -886,4 +876,12 @@ mod tests {
}
}
}

#[test]
fn test_type_check() {
let mut builder = UnionBuilder::new_sparse(2);
builder.append::<Float32Type>("a", 1.0).unwrap();
let err = builder.append::<Int32Type>("a", 1).unwrap_err().to_string();
assert!(err.contains("Attempt to write col \"a\" with type Int32 doesn't match existing type Float32"), "{}", err);
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

👍

}
}
Loading