Merge branch 'main' into feat-vector-hnsw

Author: b41sh

.github/actions/publish_binary/action.yml

@@ -41,6 +41,7 @@ runs:
 
     - name: Sync normal release to R2
       shell: bash
+      continue-on-error: true
       if: inputs.category == 'default'
       run: |
         aws s3 cp ${{ steps.name.outputs.name }}.tar.gz s3://repo/databend/${{ inputs.version }}/${{ steps.name.outputs.name }}.tar.gz --no-progress --checksum-algorithm=CRC32

.github/actions/publish_deb/action.yml

@@ -56,6 +56,7 @@ runs:
 
     - name: Publish to S3
       shell: bash
+      continue-on-error: true
       working-directory: scripts/distribution/deb
       run: |
         aws s3 sync dists s3://repo/deb/dists --delete --no-progress --checksum-algorithm=CRC32

src/query/expression/src/types/compute_view.rs

@@ -18,9 +18,13 @@ use std::marker::PhantomData;
 use std::ops::Range;
 
 use databend_common_column::buffer::Buffer;
+use num_traits::AsPrimitive;
 
 use super::simple_type::SimpleType;
 use super::AccessType;
+use crate::types::CoreNumber;
+use crate::types::Number;
+use crate::types::SimpleDomain;
 use crate::Column;
 use crate::Domain;
 use crate::ScalarRef;
@@ -140,3 +144,25 @@ where
         T::less_than_equal(&left, &right)
     }
 }
+
+/// For number convert
+pub type NumberConvertView<F, T> = ComputeView<NumberConvert<F, T>, CoreNumber<F>, CoreNumber<T>>;
+
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct NumberConvert<F, T>(std::marker::PhantomData<(F, T)>);
+
+impl<F, T> Compute<CoreNumber<F>, CoreNumber<T>> for NumberConvert<F, T>
+where
+    F: Number + AsPrimitive<T>,
+    T: Number,
+{
+    fn compute(value: &F) -> T {
+        value.as_()
+    }
+
+    fn compute_domain(domain: &SimpleDomain<F>) -> SimpleDomain<T> {
+        let min = domain.min.as_();
+        let max = domain.max.as_();
+        SimpleDomain { min, max }
+    }
+}

src/query/expression/src/types/decimal.rs

@@ -57,7 +57,9 @@ use super::SimpleDomain;
 use super::SimpleType;
 use super::SimpleValueType;
 use super::ValueType;
+use crate::types::CoreNumber;
 use crate::types::DataType;
+use crate::types::F64;
 use crate::utils::arrow::buffer_into_mut;
 use crate::with_decimal_mapped_type;
 use crate::with_decimal_type;
@@ -71,11 +73,15 @@ use crate::Value;
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct CoreDecimal<T: Decimal>(PhantomData<T>);
 
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct CoreScalarDecimal<T: Decimal>(PhantomData<T>);
+
 pub type Decimal64Type = DecimalType<i64>;
 pub type Decimal128Type = DecimalType<i128>;
 pub type Decimal256Type = DecimalType<i256>;
 
 pub type DecimalType<T> = SimpleValueType<CoreDecimal<T>>;
+pub type DecimalScalarType<T> = SimpleValueType<CoreScalarDecimal<T>>;
 
 impl<Num: Decimal> SimpleType for CoreDecimal<Num> {
     type Scalar = Num;
@@ -149,6 +155,95 @@ impl<Num: Decimal> SimpleType for CoreDecimal<Num> {
     }
 }
 
+impl<Num: Decimal> SimpleType for CoreScalarDecimal<Num> {
+    type Scalar = (Num, DecimalSize);
+    type Domain = SimpleDomain<(Num, DecimalSize)>;
+
+    fn downcast_scalar(scalar: &ScalarRef) -> Option<Self::Scalar> {
+        let scalar = scalar.as_decimal()?;
+        Num::try_downcast_scalar(scalar).map(|v| (v, scalar.size()))
+    }
+
+    fn downcast_column(col: &Column) -> Option<Buffer<Self::Scalar>> {
+        Num::try_downcast_column(col).map(|(col, x)| col.into_iter().map(|v| (v, x)).collect())
+    }
+
+    fn downcast_domain(domain: &Domain) -> Option<Self::Domain> {
+        let size = domain.as_decimal()?.decimal_size();
+        let domain = Num::try_downcast_domain(domain.as_decimal()?)?;
+        Some(SimpleDomain {
+            min: (domain.min, size),
+            max: (domain.max, size),
+        })
+    }
+
+    // It's not allowed to call downcast_builder temporarily
+    fn downcast_builder(_builder: &mut ColumnBuilder) -> Option<&mut Vec<Self::Scalar>> {
+        None
+    }
+
+    fn downcast_owned_builder(builder: ColumnBuilder) -> Option<Vec<Self::Scalar>> {
+        let size = builder.as_decimal()?.decimal_size();
+        let b = Num::try_downcast_owned_builder(builder);
+        b.map(|v| v.into_iter().map(|v| (v, size)).collect())
+    }
+
+    fn upcast_column_builder(
+        builder: Vec<Self::Scalar>,
+        data_type: &DataType,
+    ) -> Option<ColumnBuilder> {
+        Some(ColumnBuilder::Decimal(Num::upcast_builder(
+            builder.into_iter().map(|(v, _)| v).collect(),
+            *data_type.as_decimal().unwrap(),
+        )))
+    }
+
+    fn upcast_scalar(scalar: Self::Scalar, data_type: &DataType) -> Scalar {
+        let size = *data_type.as_decimal().unwrap();
+        Num::upcast_scalar(scalar.0, size)
+    }
+
+    fn upcast_column(col: Buffer<Self::Scalar>, data_type: &DataType) -> Column {
+        let col = col.into_iter().map(|(v, _)| v).collect();
+        let size = *data_type.as_decimal().unwrap();
+        Num::upcast_column(col, size)
+    }
+
+    fn upcast_domain(domain: Self::Domain, data_type: &DataType) -> Domain {
+        let domain = SimpleDomain {
+            min: domain.min.0,
+            max: domain.max.0,
+        };
+        let size = *data_type.as_decimal().unwrap();
+        Num::upcast_domain(domain, size)
+    }
+
+    #[inline(always)]
+    fn compare(lhs: &Self::Scalar, rhs: &Self::Scalar) -> Ordering {
+        lhs.cmp(rhs)
+    }
+
+    #[inline(always)]
+    fn greater_than(left: &Self::Scalar, right: &Self::Scalar) -> bool {
+        left > right
+    }
+
+    #[inline(always)]
+    fn greater_than_equal(left: &Self::Scalar, right: &Self::Scalar) -> bool {
+        left >= right
+    }
+
+    #[inline(always)]
+    fn less_than(left: &Self::Scalar, right: &Self::Scalar) -> bool {
+        left < right
+    }
+
+    #[inline(always)]
+    fn less_than_equal(left: &Self::Scalar, right: &Self::Scalar) -> bool {
+        left <= right
+    }
+}
+
 impl<Num: Decimal> DecimalType<Num> {
     pub fn full_domain(size: &DecimalSize) -> SimpleDomain<Num> {
         SimpleDomain {
@@ -200,6 +295,12 @@ impl DecimalScalar {
         })
     }
 
+    pub fn scale(&self) -> u8 {
+        with_decimal_type!(|DECIMAL| match self {
+            DecimalScalar::DECIMAL(_, size) => size.scale,
+        })
+    }
+
     pub fn as_decimal<D: Decimal>(&self) -> D {
         with_decimal_type!(|DECIMAL| match self {
             DecimalScalar::DECIMAL(value, _) => value.as_decimal(),
@@ -249,6 +350,8 @@ impl DecimalDomain {
     Copy,
     PartialEq,
     Eq,
+    PartialOrd,
+    Ord,
     Hash,
     Serialize,
     Deserialize,
@@ -260,6 +363,15 @@ pub struct DecimalSize {
     scale: u8,
 }
 
+impl Default for DecimalSize {
+    fn default() -> Self {
+        DecimalSize {
+            precision: 15,
+            scale: 2,
+        }
+    }
+}
+
 impl DecimalSize {
     pub fn new_unchecked(precision: u8, scale: u8) -> DecimalSize {
         DecimalSize { precision, scale }
@@ -2807,6 +2919,8 @@ impl Ord for i256 {
 }
 
 pub type DecimalView<F, T> = ComputeView<DecimalConvert<F, T>, CoreDecimal<F>, CoreDecimal<T>>;
+pub type DecimalF64View<F> =
+    ComputeView<DecimalConvert<F, F64>, CoreScalarDecimal<F>, CoreNumber<F64>>;
 
 #[derive(Debug, Clone, PartialEq, Eq, Default)]
 pub struct DecimalConvert<F, T>(std::marker::PhantomData<(F, T)>);
@@ -2829,10 +2943,32 @@ where
     }
 }
 
+impl<F> Compute<CoreScalarDecimal<F>, CoreNumber<F64>> for DecimalConvert<F, F64>
+where F: Decimal
+{
+    #[inline]
+    fn compute(value: &(F, DecimalSize)) -> F64 {
+        value.0.to_float64(value.1.scale()).into()
+    }
+
+    fn compute_domain(domain: &SimpleDomain<(F, DecimalSize)>) -> SimpleDomain<F64> {
+        let min = domain.min.0.to_float64(domain.min.1.scale());
+        let max = domain.max.0.to_float64(domain.max.1.scale());
+
+        SimpleDomain {
+            min: min.into(),
+            max: max.into(),
+        }
+    }
+}
+
+pub type Decimal64AsF64Type = DecimalF64View<i64>;
+pub type Decimal128AsF64Type = DecimalF64View<i128>;
+pub type Decimal256AsF64Type = DecimalF64View<i256>;
+
 macro_rules! decimal_convert_type {
     ($from:ty, $to:ty, $alias:ident, $compat:ident) => {
-        pub type $alias =
-            ComputeView<DecimalConvert<$from, $to>, CoreDecimal<$from>, CoreDecimal<$to>>;
+        pub type $alias = DecimalView<$from, $to>;
         pub type $compat = DecimalConvert<$from, $to>;
     };
 }

src/query/functions/src/aggregates/aggregate_kurtosis.rs

@@ -16,8 +16,10 @@ use borsh::BorshDeserialize;
 use borsh::BorshSerialize;
 use databend_common_exception::ErrorCode;
 use databend_common_exception::Result;
+use databend_common_expression::types::compute_view::NumberConvertView;
 use databend_common_expression::types::number::*;
 use databend_common_expression::types::*;
+use databend_common_expression::with_decimal_mapped_type;
 use databend_common_expression::with_number_mapped_type;
 use databend_common_expression::Scalar;
 use num_traits::AsPrimitive;
@@ -41,7 +43,7 @@ struct KurtosisState {
 
 impl<T> UnaryState<T, Float64Type> for KurtosisState
 where
-    T: ValueType + Sync + Send,
+    T: AccessType + Sync + Send,
     T::Scalar: AsPrimitive<f64>,
 {
     fn add(
@@ -119,16 +121,29 @@ pub fn try_create_aggregate_kurtosis_function(
     _sort_descs: Vec<AggregateFunctionSortDesc>,
 ) -> Result<AggregateFunctionRef> {
     assert_unary_arguments(display_name, arguments.len())?;
+    let return_type = DataType::Number(NumberDataType::Float64);
 
     with_number_mapped_type!(|NUM_TYPE| match &arguments[0] {
         DataType::Number(NumberDataType::NUM_TYPE) => {
-            let return_type = DataType::Number(NumberDataType::Float64);
             AggregateUnaryFunction::<
                 KurtosisState,
-                NumberType<NUM_TYPE>,
+                NumberConvertView<NUM_TYPE, F64>,
                 Float64Type,
             >::try_create_unary(display_name, return_type, params, arguments[0].clone())
         }
+        DataType::Decimal(s) => {
+            with_decimal_mapped_type!(|DECIMAL| match s.data_kind() {
+                DecimalDataKind::DECIMAL => {
+                    AggregateUnaryFunction::<
+                        KurtosisState,
+                        DecimalF64View<DECIMAL>,
+                        Float64Type,
+                    >::try_create_unary(
+                        display_name, return_type, params, arguments[0].clone()
+                    )
+                }
+            })
+        }
 
         _ => Err(ErrorCode::BadDataValueType(format!(
             "{} does not support type '{:?}'",