Exposing empty column builder. (#1959)

columnar/src/column/mod.rs

@@ -16,7 +16,7 @@ pub use serialize::{
 use crate::column_index::ColumnIndex;
 use crate::column_values::monotonic_mapping::StrictlyMonotonicMappingToInternal;
 use crate::column_values::{monotonic_map_column, ColumnValues};
-use crate::{Cardinality, DocId, MonotonicallyMappableToU64, RowId};
+use crate::{Cardinality, DocId, EmptyColumnValues, MonotonicallyMappableToU64, RowId};
 
 #[derive(Clone)]
 pub struct Column<T = u64> {
@@ -24,6 +24,15 @@ pub struct Column<T = u64> {
     pub values: Arc<dyn ColumnValues<T>>,
 }
 
+impl<T: PartialOrd + Default> Column<T> {
+    pub fn build_empty_column(num_docs: u32) -> Column<T> {
+        Column {
+            idx: ColumnIndex::Empty { num_docs },
+            values: Arc::new(EmptyColumnValues),
+        }
+    }
+}
+
 impl<T: MonotonicallyMappableToU64> Column<T> {
     pub fn to_u64_monotonic(self) -> Column<u64> {
         let values = Arc::new(monotonic_map_column(

columnar/src/column_values/mod.rs

@@ -110,20 +110,26 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
         }
     }
 
-    /// Returns the minimum value for this fast field.
+    /// Returns a lower bound for this column of values.
     ///
-    /// This min_value may not be exact.
-    /// For instance, the min value does not take in account of possible
-    /// deleted document. All values are however guaranteed to be higher than
-    /// `.min_value()`.
+    /// All values are guaranteed to be higher than `.min_value()`
+    /// but this value is not necessary the best boundary value.
+    ///
+    /// We have
+    /// ∀i < self.num_vals(), self.get_val(i) >= self.min_value()
+    /// But we don't have necessarily
+    /// ∃i < self.num_vals(), self.get_val(i) == self.min_value()
     fn min_value(&self) -> T;
 
-    /// Returns the maximum value for this fast field.
+    /// Returns an upper bound for this column of values.
     ///
-    /// This max_value may not be exact.
-    /// For instance, the max value does not take in account of possible
-    /// deleted document. All values are however guaranteed to be higher than
-    /// `.max_value()`.
+    /// All values are guaranteed to be lower than `.max_value()`
+    /// but this value is not necessary the best boundary value.
+    ///
+    /// We have
+    /// ∀i < self.num_vals(), self.get_val(i) <= self.max_value()
+    /// But we don't have necessarily
+    /// ∃i < self.num_vals(), self.get_val(i) == self.max_value()
     fn max_value(&self) -> T;
 
     /// The number of values in the column.
@@ -135,6 +141,27 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
     }
 }
 
+/// Empty column of values.
+pub struct EmptyColumnValues;
+
+impl<T: PartialOrd + Default> ColumnValues<T> for EmptyColumnValues {
+    fn get_val(&self, _idx: u32) -> T {
+        panic!("Internal Error: Called get_val of empty column.")
+    }
+
+    fn min_value(&self) -> T {
+        T::default()
+    }
+
+    fn max_value(&self) -> T {
+        T::default()
+    }
+
+    fn num_vals(&self) -> u32 {
+        0
+    }
+}
+
 impl<T: Copy + PartialOrd + Debug> ColumnValues<T> for Arc<dyn ColumnValues<T>> {
     #[inline(always)]
     fn get_val(&self, idx: u32) -> T {

columnar/src/lib.rs

@@ -23,7 +23,9 @@ mod value;
 pub use block_accessor::ColumnBlockAccessor;
 pub use column::{BytesColumn, Column, StrColumn};
 pub use column_index::ColumnIndex;
-pub use column_values::{ColumnValues, MonotonicallyMappableToU128, MonotonicallyMappableToU64};
+pub use column_values::{
+    ColumnValues, EmptyColumnValues, MonotonicallyMappableToU128, MonotonicallyMappableToU64,
+};
 pub use columnar::{
     merge_columnar, ColumnType, ColumnarReader, ColumnarWriter, HasAssociatedColumnType,
     MergeRowOrder, ShuffleMergeOrder, StackMergeOrder,