16577360936

Committed 28 Jul 2025 06:38PM UTC coverage: 81.792% (+0.01%) from 81.782%

Build # 16577360936

Build Type

Pull #4039

github

Committed by

web-flow

Commit Message

Merge f08faad54 into 1ae560509

Pull Request Pull Request #4039: docs: Document all canonical array types

Run Details

35 of 37 new or added lines in 1 file covered. (94.59%)

1 existing line in 1 file now uncovered.

43322 of 52966 relevant lines covered (81.79%)

170264.49 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

85.71

/vortex-array/src/arrays/struct_/mod.rs

// SPDX-License-Identifier: Apache-2.0
// SPDX-FileCopyrightText: Copyright the Vortex contributors

use std::fmt::Debug;
use std::iter::once;

use itertools::Itertools;
use vortex_dtype::{DType, FieldName, FieldNames, StructFields};
use vortex_error::{VortexExpect, VortexResult, vortex_bail, vortex_err};
use vortex_scalar::Scalar;

use crate::stats::{ArrayStats, StatsSetRef};
use crate::validity::Validity;
use crate::vtable::{
    ArrayVTable, CanonicalVTable, NotSupported, OperationsVTable, VTable, ValidityHelper,
    ValidityVTableFromValidityHelper,
};
use crate::{Array, ArrayRef, Canonical, EncodingId, EncodingRef, IntoArray, vtable};

mod compute;
mod serde;

vtable!(Struct);

impl VTable for StructVTable {
    type Array = StructArray;
    type Encoding = StructEncoding;

    type ArrayVTable = Self;
    type CanonicalVTable = Self;
    type OperationsVTable = Self;
    type ValidityVTable = ValidityVTableFromValidityHelper;
    type VisitorVTable = Self;
    type ComputeVTable = NotSupported;
    type EncodeVTable = NotSupported;
    type SerdeVTable = Self;

    fn id(_encoding: &Self::Encoding) -> EncodingId {
        EncodingId::new_ref("vortex.struct")
    }

    fn encoding(_array: &Self::Array) -> EncodingRef {
        EncodingRef::new_ref(StructEncoding.as_ref())
    }
}

/// A struct array that stores multiple named fields as columns, similar to a database row.
///
/// This mirrors the Apache Arrow Struct array encoding and provides a columnar representation
/// of structured data where each row contains multiple named fields of potentially different types.
///
/// ## Data Layout
///
/// The struct array uses a columnar layout where:
/// - Each field is stored as a separate child array
/// - All fields must have the same length (number of rows)
/// - Field names and types are defined in the struct's dtype
/// - An optional validity mask indicates which entire rows are null
///
/// ## Row-level nulls
///
/// The StructArray contains its own top-level nulls, which are superimposed on top of the
/// field-level validity values. This can be the case even if the fields themselves are non-nullable,
/// accessing a particular row can yield nulls even if all children are valid at that position.
///
/// ```
/// use vortex_array::arrays::{StructArray, BoolArray};
/// use vortex_array::validity::Validity;
/// use vortex_array::IntoArray;
/// use vortex_dtype::FieldNames;
/// use vortex_buffer::buffer;
///
/// // Create struct with all non-null fields but struct-level nulls
/// let struct_array = StructArray::try_new(
///     FieldNames::from(["a", "b", "c"]),
///     vec![
///         buffer![1i32, 2i32].into_array(),  // non-null field a
///         buffer![10i32, 20i32].into_array(), // non-null field b  
///         buffer![100i32, 200i32].into_array(), // non-null field c
///     ],
///     2,
///     Validity::Array(BoolArray::from_iter([true, false]).into_array()), // row 1 is null
/// ).unwrap();
///
/// // Row 0 is valid - returns a struct scalar with field values
/// let row0 = struct_array.scalar_at(0).unwrap();
/// assert!(!row0.is_null());
///
/// // Row 1 is null at struct level - returns null even though fields have values
/// let row1 = struct_array.scalar_at(1).unwrap();
/// assert!(row1.is_null());
/// ```
///
/// ## Name uniqueness
///
/// It is valid for a StructArray to have multiple child columns that have the same name. In this
/// case, any accessors that use column names will find the first column in sequence with the name.
///
/// ```
/// use vortex_array::arrays::StructArray;
/// use vortex_array::validity::Validity;
/// use vortex_array::IntoArray;
/// use vortex_dtype::FieldNames;
/// use vortex_buffer::buffer;
///
/// // Create struct with duplicate "data" field names
/// let struct_array = StructArray::try_new(
///     FieldNames::from(["data", "data"]),
///     vec![
///         buffer![1i32, 2i32].into_array(),   // first "data"
///         buffer![3i32, 4i32].into_array(),   // second "data"
///     ],
///     2,
///     Validity::NonNullable,
/// ).unwrap();
///
/// // field_by_name returns the FIRST "data" field
/// let first_data = struct_array.field_by_name("data").unwrap();
/// assert_eq!(first_data.scalar_at(0).unwrap(), 1i32.into());
/// ```
///
/// ## Field Operations
///
/// Struct arrays support efficient column operations:
/// - **Projection**: Select/reorder fields without copying data
/// - **Field access**: Get columns by name or index
/// - **Column addition**: Add new fields to create extended structs
/// - **Column removal**: Remove fields to create narrower structs
///
/// ## Validity Semantics
///
/// - Row-level nulls are tracked in the struct's validity child
/// - Individual field nulls are tracked in each field's own validity
/// - A null struct row means all fields in that row are conceptually null
/// - Field-level nulls can exist independently of struct-level nulls
///
/// # Examples
///
/// ```
/// use vortex_array::arrays::{StructArray, PrimitiveArray};
/// use vortex_array::validity::Validity;
/// use vortex_array::IntoArray;
/// use vortex_dtype::FieldNames;
/// use vortex_buffer::buffer;
///
/// // Create arrays for each field
/// let ids = PrimitiveArray::new(buffer![1i32, 2, 3], Validity::NonNullable);
/// let names = PrimitiveArray::new(buffer![100u64, 200, 300], Validity::NonNullable);
///
/// // Create struct array with named fields
/// let struct_array = StructArray::try_new(
///     FieldNames::from(["id", "score"]),
///     vec![ids.into_array(), names.into_array()],
///     3,
///     Validity::NonNullable,
/// ).unwrap();
///
/// assert_eq!(struct_array.len(), 3);
/// assert_eq!(struct_array.names().len(), 2);
///
/// // Access field by name
/// let id_field = struct_array.field_by_name("id").unwrap();
/// assert_eq!(id_field.len(), 3);
/// ```
#[derive(Clone, Debug)]
pub struct StructArray {
    len: usize,
    dtype: DType,
    fields: Vec<ArrayRef>,
    validity: Validity,
    stats_set: ArrayStats,
}

#[derive(Clone, Debug)]
pub struct StructEncoding;

impl StructArray {
    pub fn fields(&self) -> &[ArrayRef] {
        &self.fields
    }

    pub fn field_by_name(&self, name: impl AsRef<str>) -> VortexResult<&ArrayRef> {
        let name = name.as_ref();
        self.field_by_name_opt(name).ok_or_else(|| {
            vortex_err!(
                "Field {name} not found in struct array with names {:?}",
                self.names()
            )
        })
    }

    pub fn field_by_name_opt(&self, name: impl AsRef<str>) -> Option<&ArrayRef> {
        let name = name.as_ref();
        self.names()
            .iter()
            .position(|field_name| field_name.as_ref() == name)
            .map(|idx| &self.fields[idx])
    }

    pub fn names(&self) -> &FieldNames {
        self.struct_fields().names()
    }

    pub fn struct_fields(&self) -> &StructFields {
        let Some(struct_dtype) = &self.dtype.as_struct() else {
            unreachable!(
                "struct arrays must have be a DType::Struct, this is likely an internal bug."
            )
        };
        struct_dtype
    }

    /// Create a new `StructArray` with the given length, but without any fields.
    pub fn new_with_len(len: usize) -> Self {
        Self::try_new(
            FieldNames::default(),
            Vec::new(),
            len,
            Validity::NonNullable,
        )
        .vortex_expect("StructArray::new_with_len should not fail")
    }

    pub fn try_new(
        names: FieldNames,
        fields: Vec<ArrayRef>,
        length: usize,
        validity: Validity,
    ) -> VortexResult<Self> {
        let nullability = validity.nullability();

        if names.len() != fields.len() {
            vortex_bail!("Got {} names and {} fields", names.len(), fields.len());
        }

        for field in fields.iter() {
            if field.len() != length {
                vortex_bail!(
                    "Expected all struct fields to have length {length}, found {}",
                    fields.iter().map(|f| f.len()).format(","),
                );
            }
        }

        let field_dtypes: Vec<_> = fields.iter().map(|d| d.dtype()).cloned().collect();
        let dtype = DType::Struct(StructFields::new(names, field_dtypes), nullability);

        if length != validity.maybe_len().unwrap_or(length) {
            vortex_bail!(
                "array length {} and validity length must match {}",
                length,
                validity
                    .maybe_len()
                    .vortex_expect("can only fail if maybe is some")
            )
        }

        Ok(Self {
            len: length,
            dtype,
            fields,
            validity,
            stats_set: Default::default(),
        })
    }

    pub fn try_new_with_dtype(
        fields: Vec<ArrayRef>,
        dtype: StructFields,
        length: usize,
        validity: Validity,
    ) -> VortexResult<Self> {
        for (field, struct_dt) in fields.iter().zip(dtype.fields()) {
            if field.len() != length {
                vortex_bail!(
                    "Expected all struct fields to have length {length}, found {}",
                    field.len()
                );
            }

            if &struct_dt != field.dtype() {
                vortex_bail!(
                    "Expected all struct fields to have dtype {}, found {}",
                    struct_dt,
                    field.dtype()
                );
            }
        }

        Ok(Self {
            len: length,
            dtype: DType::Struct(dtype, validity.nullability()),
            fields,
            validity,
            stats_set: Default::default(),
        })
    }

    pub fn from_fields<N: AsRef<str>>(items: &[(N, ArrayRef)]) -> VortexResult<Self> {
        Self::try_from_iter(items.iter().map(|(a, b)| (a, b.to_array())))
    }

    pub fn try_from_iter_with_validity<
        N: AsRef<str>,
        A: IntoArray,
        T: IntoIterator<Item = (N, A)>,
    >(
        iter: T,
        validity: Validity,
    ) -> VortexResult<Self> {
        let (names, fields): (Vec<FieldName>, Vec<ArrayRef>) = iter
            .into_iter()
            .map(|(name, fields)| (FieldName::from(name.as_ref()), fields.into_array()))
            .unzip();
        let len = fields
            .first()
            .map(|f| f.len())
            .ok_or_else(|| vortex_err!("StructArray cannot be constructed from an empty slice of arrays because the length is unspecified"))?;

        Self::try_new(FieldNames::from_iter(names), fields, len, validity)
    }

    pub fn try_from_iter<N: AsRef<str>, A: IntoArray, T: IntoIterator<Item = (N, A)>>(
        iter: T,
    ) -> VortexResult<Self> {
        Self::try_from_iter_with_validity(iter, Validity::NonNullable)
    }

    // TODO(aduffy): Add equivalent function to support field masks for nested column access.
    /// Return a new StructArray with the given projection applied.
    ///
    /// Projection does not copy data arrays. Projection is defined by an ordinal array slice
    /// which specifies the new ordering of columns in the struct. The projection can be used to
    /// perform column re-ordering, deletion, or duplication at a logical level, without any data
    /// copying.
    #[allow(clippy::same_name_method)]
    pub fn project(&self, projection: &[FieldName]) -> VortexResult<Self> {
        let mut children = Vec::with_capacity(projection.len());
        let mut names = Vec::with_capacity(projection.len());

        for f_name in projection.iter() {
            let idx = self
                .names()
                .iter()
                .position(|name| name == f_name)
                .ok_or_else(|| vortex_err!("Unknown field {f_name}"))?;

            names.push(self.names()[idx].clone());
            children.push(self.fields()[idx].clone());
        }

        StructArray::try_new(
            FieldNames::from(names.as_slice()),
            children,
            self.len(),
            self.validity().clone(),
        )
    }

    /// Removes and returns a column from the struct array by name.
    /// If the column does not exist, returns `None`.
    pub fn remove_column(&mut self, name: impl Into<FieldName>) -> Option<ArrayRef> {
        let name = name.into();

        let struct_dtype = self.struct_fields().clone();

        let position = struct_dtype
            .names()
            .iter()
            .position(|field_name| field_name.as_ref() == name.as_ref())?;

        let field = self.fields.remove(position);

        let new_dtype = struct_dtype.without_field(position);
        self.dtype = DType::Struct(new_dtype, self.dtype.nullability());

        Some(field)
    }

    /// Create a new StructArray by appending a new column onto the existing array.
    pub fn with_column(&self, name: impl Into<FieldName>, array: ArrayRef) -> VortexResult<Self> {
        let name = name.into();
        let struct_dtype = self.struct_fields().clone();

        let names = struct_dtype.names().iter().cloned().chain(once(name));
        let types = struct_dtype.fields().chain(once(array.dtype().clone()));
        let new_fields = StructFields::new(names.collect(), types.collect());

        let mut children = self.fields.clone();
        children.push(array);

        Self::try_new_with_dtype(children, new_fields, self.len, self.validity.clone())
    }
}

impl ValidityHelper for StructArray {
    fn validity(&self) -> &Validity {
        &self.validity
    }
}

impl ArrayVTable<StructVTable> for StructVTable {
    fn len(array: &StructArray) -> usize {
        array.len
    }

    fn dtype(array: &StructArray) -> &DType {
        &array.dtype
    }

    fn stats(array: &StructArray) -> StatsSetRef<'_> {
        array.stats_set.to_ref(array.as_ref())
    }
}

impl CanonicalVTable<StructVTable> for StructVTable {
    fn canonicalize(array: &StructArray) -> VortexResult<Canonical> {
        Ok(Canonical::Struct(array.clone()))
    }
}

impl OperationsVTable<StructVTable> for StructVTable {
    fn slice(array: &StructArray, start: usize, stop: usize) -> VortexResult<ArrayRef> {
        let fields = array
            .fields()
            .iter()
            .map(|field| field.slice(start, stop))
            .try_collect()?;
        StructArray::try_new_with_dtype(
            fields,
            array.struct_fields().clone(),
            stop - start,
            array.validity().slice(start, stop)?,
        )
        .map(|a| a.into_array())
    }

    fn scalar_at(array: &StructArray, index: usize) -> VortexResult<Scalar> {
        match array.validity {
            Validity::AllValid | Validity::NonNullable => Ok(Scalar::struct_(
                array.dtype().clone(),
                array
                    .fields()
                    .iter()
                    .map(|field| field.scalar_at(index))
                    .try_collect()?,
            )),
            Validity::AllInvalid => Ok(Scalar::null(array.dtype().clone())),
            Validity::Array(_) => {
                if array.is_valid(index)? {
                    Ok(Scalar::struct_(
                        array.dtype().clone(),
                        array
                            .fields()
                            .iter()
                            .map(|field| field.scalar_at(index))
                            .try_collect()?,
                    ))
                } else {
                    Ok(Scalar::null(array.dtype().clone()))
                }
            }
        }
    }
}

#[cfg(test)]
mod test {
    use vortex_buffer::buffer;
    use vortex_dtype::{DType, FieldName, FieldNames, Nullability, PType};

    use crate::IntoArray;
    use crate::arrays::primitive::PrimitiveArray;
    use crate::arrays::struct_::StructArray;
    use crate::arrays::varbin::VarBinArray;
    use crate::arrays::{BoolArray, BoolVTable, PrimitiveVTable};
    use crate::validity::Validity;

    #[test]
    fn test_project() {
        let xs = PrimitiveArray::new(buffer![0i64, 1, 2, 3, 4], Validity::NonNullable);
        let ys = VarBinArray::from_vec(
            vec!["a", "b", "c", "d", "e"],
            DType::Utf8(Nullability::NonNullable),
        );
        let zs = BoolArray::from_iter([true, true, true, false, false]);

        let struct_a = StructArray::try_new(
            FieldNames::from(["xs", "ys", "zs"]),
            vec![xs.into_array(), ys.into_array(), zs.into_array()],
            5,
            Validity::NonNullable,
        )
        .unwrap();

        let struct_b = struct_a
            .project(&[FieldName::from("zs"), FieldName::from("xs")])
            .unwrap();
        assert_eq!(
            struct_b.names().as_ref(),
            [FieldName::from("zs"), FieldName::from("xs")],
        );

        assert_eq!(struct_b.len(), 5);

        let bools = &struct_b.fields[0];
        assert_eq!(
            bools
                .as_::<BoolVTable>()
                .boolean_buffer()
                .iter()
                .collect::<Vec<_>>(),
            vec![true, true, true, false, false]
        );

        let prims = &struct_b.fields[1];
        assert_eq!(
            prims.as_::<PrimitiveVTable>().as_slice::<i64>(),
            [0i64, 1, 2, 3, 4]
        );
    }

    #[test]
    fn test_remove_column() {
        let xs = PrimitiveArray::new(buffer![0i64, 1, 2, 3, 4], Validity::NonNullable);
        let ys = PrimitiveArray::new(buffer![4u64, 5, 6, 7, 8], Validity::NonNullable);

        let mut struct_a = StructArray::try_new(
            FieldNames::from(["xs", "ys"]),
            vec![xs.into_array(), ys.into_array()],
            5,
            Validity::NonNullable,
        )
        .unwrap();

        let removed = struct_a.remove_column("xs").unwrap();
        assert_eq!(
            removed.dtype(),
            &DType::Primitive(PType::I64, Nullability::NonNullable)
        );
        assert_eq!(
            removed.as_::<PrimitiveVTable>().as_slice::<i64>(),
            [0i64, 1, 2, 3, 4]
        );

        assert_eq!(struct_a.names(), &[FieldName::from("ys")].into());
        assert_eq!(struct_a.fields.len(), 1);
        assert_eq!(struct_a.len(), 5);
        assert_eq!(
            struct_a.fields[0].dtype(),
            &DType::Primitive(PType::U64, Nullability::NonNullable)
        );
        assert_eq!(
            struct_a.fields[0]
                .as_::<PrimitiveVTable>()
                .as_slice::<u64>(),
            [4u64, 5, 6, 7, 8]
        );

        let empty = struct_a.remove_column("non_existent");
        assert!(
            empty.is_none(),
            "Expected None when removing non-existent column"
        );
        assert_eq!(struct_a.names(), &[FieldName::from("ys")].into());
    }

    #[test]
    fn test_duplicate_field_names() {
        // Test that StructArray allows duplicate field names and returns the first match
        let field1 = buffer![1i32, 2, 3].into_array();
        let field2 = buffer![10i32, 20, 30].into_array();
        let field3 = buffer![100i32, 200, 300].into_array();

        // Create struct with duplicate field names - "value" appears twice
        let struct_array = StructArray::try_new(
            FieldNames::from(["value", "other", "value"]),
            vec![field1, field2, field3],
            3,
            Validity::NonNullable,
        )
        .unwrap();

        // field_by_name should return the first field with the matching name
        let first_value_field = struct_array.field_by_name("value").unwrap();
        assert_eq!(
            first_value_field.as_::<PrimitiveVTable>().as_slice::<i32>(),
            [1i32, 2, 3] // This is field1, not field3
        );

        // Verify field_by_name_opt also returns the first match
        let opt_field = struct_array.field_by_name_opt("value").unwrap();
        assert_eq!(
            opt_field.as_::<PrimitiveVTable>().as_slice::<i32>(),
            [1i32, 2, 3] // First "value" field
        );

        // Verify the third field (second "value") can be accessed by index
        let third_field = &struct_array.fields()[2];
        assert_eq!(
            third_field.as_::<PrimitiveVTable>().as_slice::<i32>(),
            [100i32, 200, 300]
        );
    }
}

1	// SPDX-License-Identifier: Apache-2.0
2	// SPDX-FileCopyrightText: Copyright the Vortex contributors
3
4	use std::fmt::Debug;
5	use std::iter::once;
6
7	use itertools::Itertools;
8	use vortex_dtype::{DType, FieldName, FieldNames, StructFields};
9	use vortex_error::{VortexExpect, VortexResult, vortex_bail, vortex_err};
10	use vortex_scalar::Scalar;
11
12	use crate::stats::{ArrayStats, StatsSetRef};
13	use crate::validity::Validity;
14	use crate::vtable::{
15	ArrayVTable, CanonicalVTable, NotSupported, OperationsVTable, VTable, ValidityHelper,
16	ValidityVTableFromValidityHelper,
17	};
18	use crate::{Array, ArrayRef, Canonical, EncodingId, EncodingRef, IntoArray, vtable};
19
20	mod compute;
21	mod serde;
22
23	vtable!(Struct);
24
25	impl VTable for StructVTable {
26	type Array = StructArray;
27	type Encoding = StructEncoding;
28
29	type ArrayVTable = Self;
30	type CanonicalVTable = Self;
31	type OperationsVTable = Self;
32	type ValidityVTable = ValidityVTableFromValidityHelper;
33	type VisitorVTable = Self;
34	type ComputeVTable = NotSupported;
35	type EncodeVTable = NotSupported;
36	type SerdeVTable = Self;
37
38	fn id(_encoding: &Self::Encoding) -> EncodingId {	75,301✔
39	EncodingId::new_ref("vortex.struct")	75,301✔
40	}	75,301✔
41
42	fn encoding(_array: &Self::Array) -> EncodingRef {	9,830✔
43	EncodingRef::new_ref(StructEncoding.as_ref())	9,830✔
44	}	9,830✔
45	}
46
47	/// A struct array that stores multiple named fields as columns, similar to a database row.
48	///
49	/// This mirrors the Apache Arrow Struct array encoding and provides a columnar representation
50	/// of structured data where each row contains multiple named fields of potentially different types.
51	///
52	/// ## Data Layout
53	///
54	/// The struct array uses a columnar layout where:
55	/// - Each field is stored as a separate child array
56	/// - All fields must have the same length (number of rows)
57	/// - Field names and types are defined in the struct's dtype
58	/// - An optional validity mask indicates which entire rows are null
59	///
60	/// ## Row-level nulls
61	///
62	/// The StructArray contains its own top-level nulls, which are superimposed on top of the
63	/// field-level validity values. This can be the case even if the fields themselves are non-nullable,
64	/// accessing a particular row can yield nulls even if all children are valid at that position.
65	///
66	/// ```
67	/// use vortex_array::arrays::{StructArray, BoolArray};
68	/// use vortex_array::validity::Validity;
69	/// use vortex_array::IntoArray;
70	/// use vortex_dtype::FieldNames;
71	/// use vortex_buffer::buffer;
72	///
73	/// // Create struct with all non-null fields but struct-level nulls
74	/// let struct_array = StructArray::try_new(
75	/// FieldNames::from(["a", "b", "c"]),
76	/// vec![
77	/// buffer![1i32, 2i32].into_array(), // non-null field a
78	/// buffer![10i32, 20i32].into_array(), // non-null field b
79	/// buffer![100i32, 200i32].into_array(), // non-null field c
80	/// ],
81	/// 2,
82	/// Validity::Array(BoolArray::from_iter([true, false]).into_array()), // row 1 is null
83	/// ).unwrap();
84	///
85	/// // Row 0 is valid - returns a struct scalar with field values
86	/// let row0 = struct_array.scalar_at(0).unwrap();
87	/// assert!(!row0.is_null());
88	///
89	/// // Row 1 is null at struct level - returns null even though fields have values
90	/// let row1 = struct_array.scalar_at(1).unwrap();
91	/// assert!(row1.is_null());
92	/// ```
93	///
94	/// ## Name uniqueness
95	///
96	/// It is valid for a StructArray to have multiple child columns that have the same name. In this
97	/// case, any accessors that use column names will find the first column in sequence with the name.
98	///
99	/// ```
100	/// use vortex_array::arrays::StructArray;
101	/// use vortex_array::validity::Validity;
102	/// use vortex_array::IntoArray;
103	/// use vortex_dtype::FieldNames;
104	/// use vortex_buffer::buffer;
105	///
106	/// // Create struct with duplicate "data" field names
107	/// let struct_array = StructArray::try_new(
108	/// FieldNames::from(["data", "data"]),
109	/// vec![
110	/// buffer![1i32, 2i32].into_array(), // first "data"
111	/// buffer![3i32, 4i32].into_array(), // second "data"
112	/// ],
113	/// 2,
114	/// Validity::NonNullable,
115	/// ).unwrap();
116	///
117	/// // field_by_name returns the FIRST "data" field
118	/// let first_data = struct_array.field_by_name("data").unwrap();
119	/// assert_eq!(first_data.scalar_at(0).unwrap(), 1i32.into());
120	/// ```
121	///
122	/// ## Field Operations
123	///
124	/// Struct arrays support efficient column operations:
125	/// - Projection: Select/reorder fields without copying data
126	/// - Field access: Get columns by name or index
127	/// - Column addition: Add new fields to create extended structs
128	/// - Column removal: Remove fields to create narrower structs
129	///
130	/// ## Validity Semantics
131	///
132	/// - Row-level nulls are tracked in the struct's validity child
133	/// - Individual field nulls are tracked in each field's own validity
134	/// - A null struct row means all fields in that row are conceptually null
135	/// - Field-level nulls can exist independently of struct-level nulls
136	///
137	/// # Examples
138	///
139	/// ```
140	/// use vortex_array::arrays::{StructArray, PrimitiveArray};
141	/// use vortex_array::validity::Validity;
142	/// use vortex_array::IntoArray;
143	/// use vortex_dtype::FieldNames;
144	/// use vortex_buffer::buffer;
145	///
146	/// // Create arrays for each field
147	/// let ids = PrimitiveArray::new(buffer![1i32, 2, 3], Validity::NonNullable);
148	/// let names = PrimitiveArray::new(buffer![100u64, 200, 300], Validity::NonNullable);
149	///
150	/// // Create struct array with named fields
151	/// let struct_array = StructArray::try_new(
152	/// FieldNames::from(["id", "score"]),
153	/// vec![ids.into_array(), names.into_array()],
154	/// 3,
155	/// Validity::NonNullable,
156	/// ).unwrap();
157	///
158	/// assert_eq!(struct_array.len(), 3);
159	/// assert_eq!(struct_array.names().len(), 2);
160	///
161	/// // Access field by name
162	/// let id_field = struct_array.field_by_name("id").unwrap();
163	/// assert_eq!(id_field.len(), 3);
164	/// ```
165	#[derive(Clone, Debug)]
166	pub struct StructArray {
167	len: usize,
168	dtype: DType,
169	fields: Vec<ArrayRef>,
170	validity: Validity,
171	stats_set: ArrayStats,
172	}
173
174	#[derive(Clone, Debug)]
175	pub struct StructEncoding;
176
177	impl StructArray {
178	pub fn fields(&self) -> &[ArrayRef] {	69,134✔
179	&self.fields	69,134✔
180	}	69,134✔
181
182	pub fn field_by_name(&self, name: impl AsRef<str>) -> VortexResult<&ArrayRef> {	15,346✔
183	let name = name.as_ref();	15,346✔
184	self.field_by_name_opt(name).ok_or_else(\|\| {	15,346✔
185	vortex_err!(	21✔
186	"Field {name} not found in struct array with names {:?}",	21✔
187	self.names()	21✔
188	)
189	})	21✔
190	}	15,346✔
191
192	pub fn field_by_name_opt(&self, name: impl AsRef<str>) -> Option<&ArrayRef> {	23,097✔
193	let name = name.as_ref();	23,097✔
194	self.names()	23,097✔
195	.iter()	23,097✔
196	.position(\|field_name\| field_name.as_ref() == name)	55,896✔
197	.map(\|idx\| &self.fields[idx])	23,097✔
198	}	23,097✔
199
200	pub fn names(&self) -> &FieldNames {	46,562✔
201	self.struct_fields().names()	46,562✔
202	}	46,562✔
203
204	pub fn struct_fields(&self) -> &StructFields {	48,184✔
205	let Some(struct_dtype) = &self.dtype.as_struct() else {	48,184✔
206	unreachable!(	×
207	"struct arrays must have be a DType::Struct, this is likely an internal bug."
208	)
209	};
210	struct_dtype	48,184✔
211	}	48,184✔
212
213	/// Create a new `StructArray` with the given length, but without any fields.
214	pub fn new_with_len(len: usize) -> Self {	38✔
215	Self::try_new(	38✔
216	FieldNames::default(),	38✔
217	Vec::new(),	38✔
218	len,	38✔
219	Validity::NonNullable,	38✔
220	)
221	.vortex_expect("StructArray::new_with_len should not fail")	38✔
222	}	38✔
223
224	pub fn try_new(	23,021✔
225	names: FieldNames,	23,021✔
226	fields: Vec<ArrayRef>,	23,021✔
227	length: usize,	23,021✔
228	validity: Validity,	23,021✔
229	) -> VortexResult<Self> {	23,021✔
230	let nullability = validity.nullability();	23,021✔
231
232	if names.len() != fields.len() {	23,021✔
233	vortex_bail!("Got {} names and {} fields", names.len(), fields.len());	×
234	}	23,021✔
235
236	for field in fields.iter() {	68,637✔
237	if field.len() != length {	68,637✔
238	vortex_bail!(	×
239	"Expected all struct fields to have length {length}, found {}",	×
240	fields.iter().map(\|f\| f.len()).format(","),	×
241	);
242	}	68,637✔
243	}
244
245	let field_dtypes: Vec<_> = fields.iter().map(\|d\| d.dtype()).cloned().collect();	68,637✔
246	let dtype = DType::Struct(StructFields::new(names, field_dtypes), nullability);	23,021✔
247
248	if length != validity.maybe_len().unwrap_or(length) {	23,021✔
249	vortex_bail!(	×
250	"array length {} and validity length must match {}",	×
251	length,
252	validity	×
253	.maybe_len()	×
254	.vortex_expect("can only fail if maybe is some")	×
255	)
256	}	23,021✔
257
258	Ok(Self {	23,021✔
259	len: length,	23,021✔
260	dtype,	23,021✔
261	fields,	23,021✔
262	validity,	23,021✔
263	stats_set: Default::default(),	23,021✔
264	})	23,021✔
265	}	23,021✔
266
267	pub fn try_new_with_dtype(	1,257✔
268	fields: Vec<ArrayRef>,	1,257✔
269	dtype: StructFields,	1,257✔
270	length: usize,	1,257✔
271	validity: Validity,	1,257✔
272	) -> VortexResult<Self> {	1,257✔
273	for (field, struct_dt) in fields.iter().zip(dtype.fields()) {	4,799✔
274	if field.len() != length {	4,799✔
275	vortex_bail!(	×
276	"Expected all struct fields to have length {length}, found {}",	×
277	field.len()	×
278	);
279	}	4,799✔
280
281	if &struct_dt != field.dtype() {	4,799✔
282	vortex_bail!(	×
283	"Expected all struct fields to have dtype {}, found {}",	×
284	struct_dt,
285	field.dtype()	×
286	);
287	}	4,799✔
288	}
289
290	Ok(Self {	1,257✔
291	len: length,	1,257✔
292	dtype: DType::Struct(dtype, validity.nullability()),	1,257✔
293	fields,	1,257✔
294	validity,	1,257✔
295	stats_set: Default::default(),	1,257✔
296	})	1,257✔
297	}	1,257✔
298
299	pub fn from_fields<N: AsRef<str>>(items: &[(N, ArrayRef)]) -> VortexResult<Self> {	257✔
300	Self::try_from_iter(items.iter().map(\|(a, b)\| (a, b.to_array())))	611✔
301	}	257✔
302
303	pub fn try_from_iter_with_validity<	270✔
304	N: AsRef<str>,	270✔
305	A: IntoArray,	270✔
306	T: IntoIterator<Item = (N, A)>,	270✔
307	>(	270✔
308	iter: T,	270✔
309	validity: Validity,	270✔
310	) -> VortexResult<Self> {	270✔
311	let (names, fields): (Vec<FieldName>, Vec<ArrayRef>) = iter	270✔
312	.into_iter()	270✔
313	.map(\|(name, fields)\| (FieldName::from(name.as_ref()), fields.into_array()))	629✔
314	.unzip();	270✔
315	let len = fields	270✔
316	.first()	270✔
317	.map(\|f\| f.len())	270✔
318	.ok_or_else(\|\| vortex_err!("StructArray cannot be constructed from an empty slice of arrays because the length is unspecified"))?;	270✔
319
320	Self::try_new(FieldNames::from_iter(names), fields, len, validity)	270✔
321	}	270✔
322
323	pub fn try_from_iter<N: AsRef<str>, A: IntoArray, T: IntoIterator<Item = (N, A)>>(	267✔
324	iter: T,	267✔
325	) -> VortexResult<Self> {	267✔
326	Self::try_from_iter_with_validity(iter, Validity::NonNullable)	267✔
327	}	267✔
328
329	// TODO(aduffy): Add equivalent function to support field masks for nested column access.
330	/// Return a new StructArray with the given projection applied.
331	///
332	/// Projection does not copy data arrays. Projection is defined by an ordinal array slice
333	/// which specifies the new ordering of columns in the struct. The projection can be used to
334	/// perform column re-ordering, deletion, or duplication at a logical level, without any data
335	/// copying.
336	#[allow(clippy::same_name_method)]
337	pub fn project(&self, projection: &[FieldName]) -> VortexResult<Self> {	75✔
338	let mut children = Vec::with_capacity(projection.len());	75✔
339	let mut names = Vec::with_capacity(projection.len());	75✔
340
341	for f_name in projection.iter() {	76✔
342	let idx = self	76✔
343	.names()	76✔
344	.iter()	76✔
345	.position(\|name\| name == f_name)	115✔
346	.ok_or_else(\|\| vortex_err!("Unknown field {f_name}"))?;	76✔
347
348	names.push(self.names()[idx].clone());	76✔
349	children.push(self.fields()[idx].clone());	76✔
350	}
351
352	StructArray::try_new(	75✔
353	FieldNames::from(names.as_slice()),	75✔
354	children,	75✔
355	self.len(),	75✔
356	self.validity().clone(),	75✔
357	)
358	}	75✔
359
360	/// Removes and returns a column from the struct array by name.
361	/// If the column does not exist, returns `None`.
362	pub fn remove_column(&mut self, name: impl Into<FieldName>) -> Option<ArrayRef> {	2✔
363	let name = name.into();	2✔
364
365	let struct_dtype = self.struct_fields().clone();	2✔
366
367	let position = struct_dtype	2✔
368	.names()	2✔
369	.iter()	2✔
370	.position(\|field_name\| field_name.as_ref() == name.as_ref())?;	2✔
371
372	let field = self.fields.remove(position);	1✔
373
374	let new_dtype = struct_dtype.without_field(position);	1✔
375	self.dtype = DType::Struct(new_dtype, self.dtype.nullability());	1✔
376
377	Some(field)	1✔
378	}	2✔
379
380	/// Create a new StructArray by appending a new column onto the existing array.
381	pub fn with_column(&self, name: impl Into<FieldName>, array: ArrayRef) -> VortexResult<Self> {	×
382	let name = name.into();	×
383	let struct_dtype = self.struct_fields().clone();	×
384
385	let names = struct_dtype.names().iter().cloned().chain(once(name));	×
386	let types = struct_dtype.fields().chain(once(array.dtype().clone()));	×
387	let new_fields = StructFields::new(names.collect(), types.collect());	×
388
389	let mut children = self.fields.clone();	×
390	children.push(array);	×
391
392	Self::try_new_with_dtype(children, new_fields, self.len, self.validity.clone())	×
393	}	×
394	}
395
396	impl ValidityHelper for StructArray {
397	fn validity(&self) -> &Validity {	20,058✔
398	&self.validity	20,058✔
399	}	20,058✔
400	}
401
402	impl ArrayVTable<StructVTable> for StructVTable {
403	fn len(array: &StructArray) -> usize {	132,550✔
404	array.len	132,550✔
405	}	132,550✔
406
407	fn dtype(array: &StructArray) -> &DType {	165,955✔
408	&array.dtype	165,955✔
409	}	165,955✔
410
411	fn stats(array: &StructArray) -> StatsSetRef<'_> {	93,130✔
412	array.stats_set.to_ref(array.as_ref())	93,130✔
413	}	93,130✔
414	}
415
416	impl CanonicalVTable<StructVTable> for StructVTable {
417	fn canonicalize(array: &StructArray) -> VortexResult<Canonical> {	32,214✔
418	Ok(Canonical::Struct(array.clone()))	32,214✔
419	}	32,214✔
420	}
421
422	impl OperationsVTable<StructVTable> for StructVTable {
423	fn slice(array: &StructArray, start: usize, stop: usize) -> VortexResult<ArrayRef> {	×
424	let fields = array	×
425	.fields()	×
426	.iter()	×
427	.map(\|field\| field.slice(start, stop))	×
428	.try_collect()?;	×
429	StructArray::try_new_with_dtype(	×
430	fields,	×
431	array.struct_fields().clone(),	×
432	stop - start,	×
433	array.validity().slice(start, stop)?,	×
434	)
435	.map(\|a\| a.into_array())	×
436	}	×
437
438	fn scalar_at(array: &StructArray, index: usize) -> VortexResult<Scalar> {	654✔
439	match array.validity {	654✔
440	Validity::AllValid \| Validity::NonNullable => Ok(Scalar::struct_(	641✔
441	array.dtype().clone(),	641✔
442	array	641✔
443	.fields()	641✔
444	.iter()	641✔
445	.map(\|field\| field.scalar_at(index))	1,263✔
446	.try_collect()?,	641✔
447	)),
NEW 448	Validity::AllInvalid => Ok(Scalar::null(array.dtype().clone())),	×
449	Validity::Array(_) => {
450	if array.is_valid(index)? {	13✔
451	Ok(Scalar::struct_(	13✔
452	array.dtype().clone(),	13✔
453	array	13✔
454	.fields()	13✔
455	.iter()	13✔
456	.map(\|field\| field.scalar_at(index))	13✔
457	.try_collect()?,	13✔
458	))
459	} else {
NEW UNCOV 460	Ok(Scalar::null(array.dtype().clone()))	×
461	}
462	}
463	}
464	}	654✔
465	}
466
467	#[cfg(test)]
468	mod test {
469	use vortex_buffer::buffer;
470	use vortex_dtype::{DType, FieldName, FieldNames, Nullability, PType};
471
472	use crate::IntoArray;
473	use crate::arrays::primitive::PrimitiveArray;
474	use crate::arrays::struct_::StructArray;
475	use crate::arrays::varbin::VarBinArray;
476	use crate::arrays::{BoolArray, BoolVTable, PrimitiveVTable};
477	use crate::validity::Validity;
478
479	#[test]
480	fn test_project() {	1✔
481	let xs = PrimitiveArray::new(buffer![0i64, 1, 2, 3, 4], Validity::NonNullable);	1✔
482	let ys = VarBinArray::from_vec(	1✔
483	vec!["a", "b", "c", "d", "e"],	1✔
484	DType::Utf8(Nullability::NonNullable),	1✔
485	);
486	let zs = BoolArray::from_iter([true, true, true, false, false]);	1✔
487
488	let struct_a = StructArray::try_new(	1✔
489	FieldNames::from(["xs", "ys", "zs"]),	1✔
490	vec![xs.into_array(), ys.into_array(), zs.into_array()],	1✔
491	5,
492	Validity::NonNullable,	1✔
493	)
494	.unwrap();	1✔
495
496	let struct_b = struct_a	1✔
497	.project(&[FieldName::from("zs"), FieldName::from("xs")])	1✔
498	.unwrap();	1✔
499	assert_eq!(	1✔
500	struct_b.names().as_ref(),	1✔
501	[FieldName::from("zs"), FieldName::from("xs")],	1✔
502	);
503
504	assert_eq!(struct_b.len(), 5);	1✔
505
506	let bools = &struct_b.fields[0];	1✔
507	assert_eq!(	1✔
508	bools	1✔
509	.as_::<BoolVTable>()	1✔
510	.boolean_buffer()	1✔
511	.iter()	1✔
512	.collect::<Vec<_>>(),	1✔
513	vec![true, true, true, false, false]	1✔
514	);
515
516	let prims = &struct_b.fields[1];	1✔
517	assert_eq!(	1✔
518	prims.as_::<PrimitiveVTable>().as_slice::<i64>(),	1✔
519	[0i64, 1, 2, 3, 4]
520	);
521	}	1✔
522
523	#[test]
524	fn test_remove_column() {	1✔
525	let xs = PrimitiveArray::new(buffer![0i64, 1, 2, 3, 4], Validity::NonNullable);	1✔
526	let ys = PrimitiveArray::new(buffer![4u64, 5, 6, 7, 8], Validity::NonNullable);	1✔
527
528	let mut struct_a = StructArray::try_new(	1✔
529	FieldNames::from(["xs", "ys"]),	1✔
530	vec![xs.into_array(), ys.into_array()],	1✔
531	5,
532	Validity::NonNullable,	1✔
533	)
534	.unwrap();	1✔
535
536	let removed = struct_a.remove_column("xs").unwrap();	1✔
537	assert_eq!(	1✔
538	removed.dtype(),	1✔
539	&DType::Primitive(PType::I64, Nullability::NonNullable)
540	);
541	assert_eq!(	1✔
542	removed.as_::<PrimitiveVTable>().as_slice::<i64>(),	1✔
543	[0i64, 1, 2, 3, 4]
544	);
545
546	assert_eq!(struct_a.names(), &[FieldName::from("ys")].into());	1✔
547	assert_eq!(struct_a.fields.len(), 1);	1✔
548	assert_eq!(struct_a.len(), 5);	1✔
549	assert_eq!(	1✔
550	struct_a.fields[0].dtype(),	1✔
551	&DType::Primitive(PType::U64, Nullability::NonNullable)
552	);
553	assert_eq!(	1✔
554	struct_a.fields[0]	1✔
555	.as_::<PrimitiveVTable>()	1✔
556	.as_slice::<u64>(),	1✔
557	[4u64, 5, 6, 7, 8]
558	);
559
560	let empty = struct_a.remove_column("non_existent");	1✔
561	assert!(	1✔
562	empty.is_none(),	1✔
563	"Expected None when removing non-existent column"
564	);
565	assert_eq!(struct_a.names(), &[FieldName::from("ys")].into());	1✔
566	}	1✔
567
568	#[test]
569	fn test_duplicate_field_names() {	1✔
570	// Test that StructArray allows duplicate field names and returns the first match
571	let field1 = buffer![1i32, 2, 3].into_array();	1✔
572	let field2 = buffer![10i32, 20, 30].into_array();	1✔
573	let field3 = buffer![100i32, 200, 300].into_array();	1✔
574
575	// Create struct with duplicate field names - "value" appears twice
576	let struct_array = StructArray::try_new(	1✔
577	FieldNames::from(["value", "other", "value"]),	1✔
578	vec![field1, field2, field3],	1✔
579	3,
580	Validity::NonNullable,	1✔
581	)
582	.unwrap();	1✔
583
584	// field_by_name should return the first field with the matching name
585	let first_value_field = struct_array.field_by_name("value").unwrap();	1✔
586	assert_eq!(	1✔
587	first_value_field.as_::<PrimitiveVTable>().as_slice::<i32>(),	1✔
588	[1i32, 2, 3] // This is field1, not field3
589	);
590
591	// Verify field_by_name_opt also returns the first match
592	let opt_field = struct_array.field_by_name_opt("value").unwrap();	1✔
593	assert_eq!(	1✔
594	opt_field.as_::<PrimitiveVTable>().as_slice::<i32>(),	1✔
595	[1i32, 2, 3] // First "value" field
596	);
597
598	// Verify the third field (second "value") can be accessed by index
599	let third_field = &struct_array.fields()[2];	1✔
600	assert_eq!(	1✔
601	third_field.as_::<PrimitiveVTable>().as_slice::<i32>(),	1✔
602	[100i32, 200, 300]
603	);
604	}	1✔
605	}

vortex-data / vortex / 16577360936

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous