docs

Author: wjones127

datafusion/optimizer/src/simplify_expressions/expr_simplifier.rs

@@ -152,7 +152,58 @@ impl<S: SimplifyInfo> ExprSimplifier<S> {
         expr.rewrite(&mut expr_rewrite)
     }
 
-    /// Add guarantees
+    /// Input guarantees and simplify the expression.
+    ///
+    /// The guarantees can simplify expressions. For example, if a column is
+    /// guaranteed to always be a certain value, it's references in the expression
+    /// can be replaced with that literal.
+    ///
+    /// ```rust
+    /// use arrow::datatypes::{DataType, Field, Schema};
+    /// use datafusion_expr::{col, lit, Expr};
+    /// use datafusion_common::{Result, ScalarValue, ToDFSchema};
+    /// use datafusion_physical_expr::execution_props::ExecutionProps;
+    /// use datafusion_optimizer::simplify_expressions::{
+    ///     ExprSimplifier, SimplifyContext,
+    ///     guarantees::{Guarantee, GuaranteeBound, NullStatus}};
+    ///
+    /// let schema = Schema::new(vec![
+    ///   Field::new("x", DataType::Int64, false),
+    ///   Field::new("y", DataType::UInt32, false),
+    ///   Field::new("z", DataType::Int64, false),
+    ///   ])
+    ///   .to_dfschema_ref().unwrap();
+    ///
+    /// // Create the simplifier
+    /// let props = ExecutionProps::new();
+    /// let context = SimplifyContext::new(&props)
+    ///    .with_schema(schema);
+    /// let simplifier = ExprSimplifier::new(context);
+    ///
+    /// // Expression: (x >= 3) AND (y + 2 < 10) AND (z > 5)
+    /// let expr_x = col("x").gt_eq(lit(3_i64));
+    /// let expr_y = (col("y") + lit(2_u32)).lt(lit(10_u32));
+    /// let expr_z = col("z").gt(lit(5_i64));
+    /// let expr = expr_x.and(expr_y).and(expr_z.clone());
+    ///
+    /// let guarantees = vec![
+    ///    // x is guaranteed to be between 3 and 5
+    ///    (
+    ///        col("x"),
+    ///        Guarantee::new(
+    ///            Some(GuaranteeBound::new(ScalarValue::Int64(Some(3)), false)),
+    ///            Some(GuaranteeBound::new(ScalarValue::Int64(Some(5)), false)),
+    ///            NullStatus::NeverNull,
+    ///        )
+    ///    ),
+    ///    // y is guaranteed to be 3
+    ///    (col("y"), Guarantee::from(&ScalarValue::UInt32(Some(3)))),
+    /// ];
+    /// let output = simplifier.simplify_with_guarantees(expr, &guarantees).unwrap();
+    /// // Expression becomes: true AND true AND (z > 5), which simplifies to
+    /// // z > 5.
+    /// assert_eq!(output, expr_z);
+    /// ```
     pub fn simplify_with_guarantees<'a>(
         &self,
         expr: Expr,

datafusion/optimizer/src/simplify_expressions/guarantees.rs

@@ -15,16 +15,34 @@
 // specific language governing permissions and limitations
 // under the License.
 
-//! Logic to inject guarantees with expressions.
+//! Guarantees which can be used with [ExprSimplifier::simplify_with_guarantees()][crate::simplify_expressions::expr_simplifier::ExprSimplifier::simplify_with_guarantees].
 //!
+//! Guarantees can represent single values or possible ranges of values.
+//!
+//! ```
+//! use datafusion_optimizer::simplify_expressions::guarantees::{
+//!     Guarantee, GuaranteeBound, NullStatus};
+//!
+//! // Guarantee that value is always 1_i32
+//! Guarantee::from(&ScalarValue::Int32(Some(1)));
+//! // Guarantee that value is always NULL
+//! Guarantee::from(&ScalarValue::Null);
+//! // Guarantee that value is always between 1_i32 and 10_i32 (inclusive)
+//! // and never null.
+//! Guarantee::new(
+//!    Some(GuaranteeBound::new(ScalarValue::Int32(Some(1)), false)),
+//!    Some(GuaranteeBound::new(ScalarValue::Int32(Some(10)), false)),
+//!    NullStatus::NeverNull,
+//! );
+//! ```
 use datafusion_common::{tree_node::TreeNodeRewriter, Result, ScalarValue};
 use datafusion_expr::{expr::InList, lit, Between, BinaryExpr, Expr, Operator};
 use std::collections::HashMap;
 
 /// A bound on the value of an expression.
 #[derive(Debug, Clone, PartialEq)]
 pub struct GuaranteeBound {
-    /// The value of the bound.
+    /// The value of the bound. If the bound is null, then there is no bound.
     pub bound: ScalarValue,
     /// If true, the bound is exclusive. If false, the bound is inclusive.
     /// In terms of inequalities, this means the bound is `<` or `>` rather than
@@ -40,6 +58,7 @@ impl GuaranteeBound {
 }
 
 impl Default for GuaranteeBound {
+    /// Default value is a closed bound at null.
     fn default() -> Self {
         Self {
             bound: ScalarValue::Null,
@@ -70,9 +89,11 @@ pub enum NullStatus {
 /// nulls.
 #[derive(Debug, Clone, PartialEq)]
 pub struct Guarantee {
-    /// The min values that the expression can take on. If `min.bound` is
+    /// The min values that the expression can take on. If the min is null, then
+    /// there is no known min.
     pub min: GuaranteeBound,
-    /// The max values that the expression can take on.
+    /// The max values that the expression can take on. If the max is null,
+    /// then there is no known max.
     pub max: GuaranteeBound,
     /// Whether the expression is expected to be either always null or never null.
     pub null_status: NullStatus,
@@ -97,14 +118,19 @@ impl Guarantee {
         self.min.bound > *value || (self.min.bound == *value && self.min.open)
     }
 
+    /// Whether values are guaranteed to be greater than or equal to the given
+    /// value.
     fn greater_than_or_eq(&self, value: &ScalarValue) -> bool {
         self.min.bound >= *value
     }
 
+    /// Whether values are guaranteed to be less than the given value.
     fn less_than(&self, value: &ScalarValue) -> bool {
         self.max.bound < *value || (self.max.bound == *value && self.max.open)
     }
 
+    /// Whether values are guaranteed to be less than or equal to the given
+    /// value.
     fn less_than_or_eq(&self, value: &ScalarValue) -> bool {
         self.max.bound <= *value
     }
@@ -136,8 +162,6 @@ impl From<&ScalarValue> for Guarantee {
 }
 
 /// Rewrite expressions to incorporate guarantees.
-///
-///
 pub(crate) struct GuaranteeRewriter<'a> {
     guarantees: HashMap<&'a Expr, &'a Guarantee>,
 }