test: add regression tests for initial hash table allocation tracking

Verify that ArrowBytesMap::size() and ArrowBytesViewMap::size() account
for the hash table memory pre-allocated by HashTable::with_capacity().
Both tests fail without the fix (map_size: 0) and pass with it.

Author: yashrb24

datafusion/physical-expr-common/src/binary_map.rs

@@ -876,6 +876,47 @@ mod tests {
         assert!(size_after_values2 > total_strings1_len + total_strings2_len);
     }
 
+    /// Verify that `size()` accounts for the initial hash table allocation
+    /// from `HashTable::with_capacity(INITIAL_MAP_CAPACITY)`.
+    ///
+    /// Regression test: `map_size` was previously initialized to 0 despite
+    /// the hash table pre-allocating memory, causing `size()` to undercount
+    /// until the first hash table resize.
+    #[test]
+    fn test_size_accounts_for_initial_hash_table_allocation() {
+        let set = ArrowBytesSet::<i32>::new(OutputType::Utf8);
+        let initial_size = set.size();
+
+        // Compute the exact allocation for a HashTable with the same
+        // capacity and entry type used by ArrowBytesMap
+        let expected_ht_alloc =
+            hashbrown::hash_table::HashTable::<Entry<i32, ()>>::with_capacity(
+                INITIAL_MAP_CAPACITY,
+            )
+            .allocation_size();
+
+        // The hash table must allocate non-trivial memory for 128 entries
+        assert!(
+            expected_ht_alloc > 0,
+            "hash table should allocate memory for {INITIAL_MAP_CAPACITY} entries"
+        );
+
+        // size() = map_size + buffer_capacity + offsets + hashes_buffer
+        // At creation the non-map components are:
+        //   buffer:  INITIAL_BUFFER_CAPACITY bytes
+        //   offsets: vec with 1 element
+        //   hashes:  empty vec (0 bytes)
+        //
+        // Before the fix (map_size=0), size() ≈ INITIAL_BUFFER_CAPACITY + sizeof(i32),
+        // missing the hash table allocation entirely.
+        // After the fix, size() includes expected_ht_alloc as well.
+        assert!(
+            initial_size >= expected_ht_alloc + INITIAL_BUFFER_CAPACITY,
+            "size() ({initial_size}) should include both hash table allocation \
+             ({expected_ht_alloc}) and buffer capacity ({INITIAL_BUFFER_CAPACITY})"
+        );
+    }
+
     #[test]
     fn test_map() {
         let input = vec![

datafusion/physical-expr-common/src/binary_view_map.rs

@@ -680,6 +680,43 @@ mod tests {
         assert_eq!(set.len(), 10);
     }
 
+    /// Verify that `size()` accounts for the initial hash table allocation
+    /// from `HashTable::with_capacity(INITIAL_MAP_CAPACITY)`.
+    ///
+    /// Regression test: `map_size` was previously initialized to 0 despite
+    /// the hash table pre-allocating memory, causing `size()` to undercount
+    /// until the first hash table resize.
+    #[test]
+    fn test_size_accounts_for_initial_hash_table_allocation() {
+        let set = ArrowBytesViewSet::new(OutputType::Utf8View);
+        let initial_size = set.size();
+
+        // Compute the exact allocation for a HashTable with the same
+        // capacity and entry type used by ArrowBytesViewMap
+        let expected_ht_alloc =
+            hashbrown::hash_table::HashTable::<Entry<()>>::with_capacity(
+                INITIAL_MAP_CAPACITY,
+            )
+            .allocation_size();
+
+        assert!(
+            expected_ht_alloc > 0,
+            "hash table should allocate memory for {INITIAL_MAP_CAPACITY} entries"
+        );
+
+        // For ArrowBytesViewMap, all non-map components (views, in_progress,
+        // completed, nulls, hashes_buffer) start empty, so:
+        //   size() = map_size + 0
+        //
+        // Before the fix (map_size=0), size() was 0 at creation,
+        // missing the hash table allocation entirely.
+        // After the fix, size() == expected_ht_alloc.
+        assert!(
+            initial_size >= expected_ht_alloc,
+            "size() ({initial_size}) should include hash table allocation ({expected_ht_alloc})"
+        );
+    }
+
     #[derive(Debug, PartialEq, Eq, Default, Clone, Copy)]
     struct TestPayload {
         // store the string value to check against input