Fix null_count for Null type

Author: pitrou

arrow-data/src/ffi.rs

@@ -168,6 +168,12 @@ impl FFI_ArrowArray {
             .collect::<Box<_>>();
         let n_children = children.len() as i64;
 
+        // As in the IPC format, emit null_count = length for Null type
+        let null_count = match data.data_type() {
+            DataType::Null => data.len(),
+            _ => data.null_count(),
+        };
+
         // create the private data owning everything.
         // any other data must be added here, e.g. via a struct, to track lifetime.
         let mut private_data = Box::new(ArrayPrivateData {
@@ -179,7 +185,7 @@ impl FFI_ArrowArray {
 
         Self {
             length: data.len() as i64,
-            null_count: data.null_count() as i64,
+            null_count: null_count as i64,
             offset: data.offset() as i64,
             n_buffers,
             n_children,

arrow-integration-testing/src/lib.rs

@@ -247,6 +247,7 @@ fn cdata_integration_import_batch_and_compare_to_json(
     let imported_array = unsafe { std::ptr::replace(c_array, FFI_ArrowArray::empty()) };
     let imported_array =
         from_ffi_and_data_type(imported_array, DataType::Struct(schema.fields.clone()))?;
+    imported_array.validate_full()?;
     let imported_batch = RecordBatch::from(StructArray::from(imported_array));
 
     compare_batches(&json_batch, &imported_batch)

arrow/src/ffi.rs

@@ -271,7 +271,10 @@ impl<'a> ImportedArrowArray<'a> {
     fn consume(self) -> Result<ArrayData> {
         let len = self.array.len();
         let offset = self.array.offset();
-        let null_count = self.array.null_count();
+        let null_count = match &self.data_type {
+            DataType::Null => 0,
+            _ => self.array.null_count(),
+        };
 
         let data_layout = layout(&self.data_type);
         let buffers = self.buffers(data_layout.can_contain_null_mask)?;