vortex_pco/compute/
cast.rs

1// SPDX-License-Identifier: Apache-2.0
2// SPDX-FileCopyrightText: Copyright the Vortex contributors
3
4use vortex_array::ArrayRef;
5use vortex_array::IntoArray;
6use vortex_array::dtype::DType;
7use vortex_array::scalar_fn::fns::cast::CastReduce;
8use vortex_error::VortexResult;
9
10use crate::PcoArray;
11use crate::PcoVTable;
12
13impl CastReduce for PcoVTable {
14    fn cast(array: &PcoArray, dtype: &DType) -> VortexResult<Option<ArrayRef>> {
15        if !dtype.is_nullable() || !array.all_valid()? {
16            // TODO(joe): fixme
17            // We cannot cast to non-nullable since the validity containing nulls is used to decode
18            // the PCO array, this would require rewriting tables.
19            return Ok(None);
20        }
21        // PCO (Pcodec) is a compression encoding that stores data in a compressed format.
22        // It can efficiently handle nullability changes without decompression, but type changes
23        // require decompression since the compression algorithm is type-specific.
24        // PCO supports: F16, F32, F64, I16, I32, I64, U16, U32, U64
25        if array.dtype().eq_ignore_nullability(dtype) {
26            // Create a new validity with the target nullability
27            let new_validity = array
28                .unsliced_validity
29                .clone()
30                .cast_nullability(dtype.nullability(), array.len())?;
31
32            return Ok(Some(
33                PcoArray::new(
34                    array.chunk_metas.clone(),
35                    array.pages.clone(),
36                    dtype.clone(),
37                    array.metadata.clone(),
38                    array.unsliced_n_rows(),
39                    new_validity,
40                )
41                ._slice(array.slice_start(), array.slice_stop())
42                .into_array(),
43            ));
44        }
45
46        // For other casts (e.g., numeric type changes), decode to canonical and let PrimitiveArray handle it
47        Ok(None)
48    }
49}
50
51#[cfg(test)]
52mod tests {
53    use rstest::rstest;
54    use vortex_array::IntoArray;
55    use vortex_array::arrays::PrimitiveArray;
56    use vortex_array::assert_arrays_eq;
57    use vortex_array::builtins::ArrayBuiltins;
58    use vortex_array::compute::conformance::cast::test_cast_conformance;
59    use vortex_array::dtype::DType;
60    use vortex_array::dtype::Nullability;
61    use vortex_array::dtype::PType;
62    use vortex_array::validity::Validity;
63    use vortex_buffer::buffer;
64
65    use crate::PcoArray;
66
67    #[test]
68    fn test_cast_pco_f32_to_f64() {
69        let values = PrimitiveArray::from_iter([1.0f32, 2.0, 3.0, 4.0, 5.0]);
70        let pco = PcoArray::from_primitive(&values, 0, 128).unwrap();
71
72        let casted = pco
73            .into_array()
74            .cast(DType::Primitive(PType::F64, Nullability::NonNullable))
75            .unwrap();
76        assert_eq!(
77            casted.dtype(),
78            &DType::Primitive(PType::F64, Nullability::NonNullable)
79        );
80
81        assert_arrays_eq!(
82            casted,
83            PrimitiveArray::from_iter([1.0f64, 2.0, 3.0, 4.0, 5.0])
84        );
85    }
86
87    #[test]
88    fn test_cast_pco_nullability_change() {
89        // Test casting from NonNullable to Nullable
90        let values = PrimitiveArray::from_iter([10u32, 20, 30, 40]);
91        let pco = PcoArray::from_primitive(&values, 0, 128).unwrap();
92
93        let casted = pco
94            .into_array()
95            .cast(DType::Primitive(PType::U32, Nullability::Nullable))
96            .unwrap();
97        assert_arrays_eq!(
98            casted,
99            PrimitiveArray::new(buffer![10u32, 20, 30, 40], Validity::AllValid,)
100        );
101    }
102
103    #[test]
104    fn test_cast_sliced_pco_nullable_to_nonnullable() {
105        let values = PrimitiveArray::new(
106            buffer![10u32, 20, 30, 40, 50, 60],
107            Validity::from_iter([true, true, true, true, true, true]),
108        );
109        let pco = PcoArray::from_primitive(&values, 0, 128).unwrap();
110        let sliced = pco.slice(1..5).unwrap();
111        let casted = sliced
112            .cast(DType::Primitive(PType::U32, Nullability::NonNullable))
113            .unwrap();
114        assert_eq!(
115            casted.dtype(),
116            &DType::Primitive(PType::U32, Nullability::NonNullable)
117        );
118        // Verify the values are correct
119        assert_arrays_eq!(casted, PrimitiveArray::from_iter([20u32, 30, 40, 50]));
120    }
121
122    #[test]
123    fn test_cast_sliced_pco_part_valid_to_nonnullable() {
124        let values = PrimitiveArray::from_option_iter([
125            None,
126            Some(20u32),
127            Some(30),
128            Some(40),
129            Some(50),
130            Some(60),
131        ]);
132        let pco = PcoArray::from_primitive(&values, 0, 128).unwrap();
133        let sliced = pco.slice(1..5).unwrap();
134        let casted = sliced
135            .cast(DType::Primitive(PType::U32, Nullability::NonNullable))
136            .unwrap();
137        assert_eq!(
138            casted.dtype(),
139            &DType::Primitive(PType::U32, Nullability::NonNullable)
140        );
141        assert_arrays_eq!(casted, PrimitiveArray::from_iter([20u32, 30, 40, 50]));
142    }
143
144    #[rstest]
145    #[case::f32(PrimitiveArray::new(
146        buffer![1.23f32, 4.56, 7.89, 10.11, 12.13],
147        Validity::NonNullable,
148    ))]
149    #[case::f64(PrimitiveArray::new(
150        buffer![100.1f64, 200.2, 300.3, 400.4, 500.5],
151        Validity::NonNullable,
152    ))]
153    #[case::i32(PrimitiveArray::new(
154        buffer![100i32, 200, 300, 400, 500],
155        Validity::NonNullable,
156    ))]
157    #[case::u64(PrimitiveArray::new(
158        buffer![1000u64, 2000, 3000, 4000],
159        Validity::NonNullable,
160    ))]
161    #[case::single(PrimitiveArray::new(
162        buffer![42.42f64],
163        Validity::NonNullable,
164    ))]
165    fn test_cast_pco_conformance(#[case] values: PrimitiveArray) {
166        let pco = PcoArray::from_primitive(&values, 0, 128).unwrap();
167        test_cast_conformance(&pco.into_array());
168    }
169}
vortex_pco/compute/cast.rs

vortex_pco/compute/
cast.rs
