duckdb: define virtual columns on c++ side (#7571)

Slightly reduce complexity. Virtual columns aren't dependent on
anything, so
we don't need to expose extra stuff

Signed-off-by: Mikhail Kot <to@myrrc.dev>

Author: myrrc

vortex-duckdb/cpp/include/duckdb_vx/table_function.h

@@ -38,15 +38,6 @@ void duckdb_vx_tfunc_bind_result_add_column(duckdb_vx_tfunc_bind_result ffi_resu
                                             size_t name_len,
                                             duckdb_logical_type ffi_type);
 
-// Opaque type for the result of get_virtual_columns
-typedef struct duckdb_vx_tfunc_virtual_cols_result_ *duckdb_vx_tfunc_virtual_cols_result;
-// Push a column into the get_virtual_columns result.
-void duckdb_vx_tfunc_virtual_columns_push(duckdb_vx_tfunc_virtual_cols_result ffi_result,
-                                          idx_t column_idx,
-                                          const char *name_str,
-                                          size_t name_len,
-                                          duckdb_logical_type ffi_type);
-
 // String map for to_string result
 typedef struct duckdb_vx_string_map_ *duckdb_vx_string_map;
 
@@ -161,8 +152,6 @@ typedef struct {
 
     bool (*pushdown_complex_filter)(void *bind_data, duckdb_vx_expr expr, duckdb_vx_error *error_out);
 
-    void (*get_virtual_columns)(void *bind_data, duckdb_vx_tfunc_virtual_cols_result result_out);
-
     void *pushdown_expression;
     duckdb_vx_string_map (*to_string)(void *bind_data);
     // void *dynamic_to_string;
@@ -179,7 +168,6 @@ typedef struct {
     // void *supports_pushdown_type;
     // void *get_partition_info;
     // void *get_partition_stats;
-    // void *get_virtual_columns;
     // void *get_row_id_columns;
 
     bool projection_pushdown;

vortex-duckdb/cpp/table_function.cpp

@@ -377,33 +377,6 @@ extern "C" void duckdb_vx_tfunc_bind_result_add_column(duckdb_vx_tfunc_bind_resu
     result->return_types.emplace_back(*logical_type);
 }
 
-virtual_column_map_t c_get_virtual_columns(ClientContext & /*context*/,
-                                           optional_ptr<FunctionData> bind_data) {
-    auto &bind = bind_data->Cast<CTableBindData>();
-
-    auto result = virtual_column_map_t();
-    bind.info->vtab.get_virtual_columns(bind_data->Cast<CTableBindData>().ffi_data->DataPtr(),
-                                        reinterpret_cast<duckdb_vx_tfunc_virtual_cols_result>(&result));
-    return result;
-}
-
-extern "C" void duckdb_vx_tfunc_virtual_columns_push(duckdb_vx_tfunc_virtual_cols_result ffi_result,
-                                                     idx_t column_idx,
-                                                     const char *name_str,
-                                                     size_t name_len,
-                                                     duckdb_logical_type ffi_type) {
-    if (!ffi_result || !name_str || !ffi_type) {
-        return;
-    }
-
-    auto result = reinterpret_cast<virtual_column_map_t *>(ffi_result);
-    const auto logical_type = reinterpret_cast<LogicalType *>(ffi_type);
-    const auto name = string(name_str, name_len);
-
-    auto table_col = TableColumn(std::move(name), *logical_type);
-    result->emplace(column_idx, std::move(table_col));
-}
-
 OperatorPartitionData c_get_partition_data(ClientContext & /*context*/,
                                            TableFunctionGetPartitionInput &input) {
     if (input.partition_info.RequiresPartitionColumns()) {
@@ -463,11 +436,14 @@ extern "C" duckdb_state duckdb_vx_tfunc_register(duckdb_database ffi_db, const d
     tf.late_materialization = vtab->late_materialization;
     tf.cardinality = c_cardinality;
     tf.get_partition_data = c_get_partition_data;
-    tf.get_virtual_columns = c_get_virtual_columns;
     tf.to_string = c_to_string;
     tf.table_scan_progress = c_table_scan_progress;
     tf.statistics = c_statistics;
 
+    tf.get_virtual_columns = [](auto &, auto) -> virtual_column_map_t {
+        return {{COLUMN_IDENTIFIER_EMPTY, TableColumn("", LogicalTypeId::BOOLEAN)}};
+    };
+
     // Set up the parameters
     tf.arguments.reserve(vtab->parameter_count);
     for (size_t i = 0; i < vtab->parameter_count; i++) {

vortex-duckdb/src/datasource.rs

@@ -5,7 +5,7 @@
 //!
 //! Table functions that resolve to a [`DataSourceRef`] can implement [`DataSourceTableFunction`]
 //! to get a blanket [`TableFunction`] implementation covering init, scan, progress, filter
-//! pushdown, cardinality, partitioning, and virtual columns.
+//! pushdown, cardinality, and partitioning.
 
 use std::ffi::CString;
 use std::fmt::Debug;
@@ -69,7 +69,6 @@ use crate::duckdb::LogicalType;
 use crate::duckdb::TableFilterSetRef;
 use crate::duckdb::TableFunction;
 use crate::duckdb::TableInitInput;
-use crate::duckdb::VirtualColumnsResultRef;
 use crate::exporter::ArrayExporter;
 use crate::exporter::ConversionCache;
 
@@ -83,17 +82,15 @@ use crate::exporter::ConversionCache;
 /// If you define COLUMN_IDENTIFIER_EMPTY, planner takes it, otherwise the
 /// first column. As we don't want to fill the output chunk and we can leave
 /// it uninitialized in this case, we define COLUMN_IDENTIFIER_EMPTY as a
-/// virtual column in our table function vtab's get_virtual_columns.
-/// See vortex-duckdb/cpp/include/duckdb_vx/table_function.h
-/// See virtual_columns in this file
+/// virtual column.
+/// See vortex-duckdb/cpp/table_function.cpp
 static EMPTY_COLUMN_IDX: u64 = 18446744073709551614;
-static EMPTY_COLUMN_NAME: &str = "";
 
 /// A trait for table functions that resolve to a [`DataSourceRef`].
 ///
 /// Implementors only need to define how parameters are declared and how binding produces a
 /// data source. All other [`TableFunction`] methods (init, scan, progress, filter pushdown,
-/// cardinality, partitioning, virtual columns) are provided by a blanket implementation.
+/// cardinality, partitioning) are provided by a blanket implementation.
 pub(crate) trait DataSourceTableFunction: Sized + Debug {
     /// Returns the positional parameters of the table function.
     fn parameters() -> Vec<LogicalType> {
@@ -555,10 +552,6 @@ impl<T: DataSourceTableFunction> TableFunction for T {
 
         Some(result)
     }
-
-    fn virtual_columns(_bind_data: &Self::BindData, result: &mut VirtualColumnsResultRef) {
-        result.register(EMPTY_COLUMN_IDX, EMPTY_COLUMN_NAME, &LogicalType::bool());
-    }
 }
 
 // ---------------------------------------------------------------------------

vortex-duckdb/src/duckdb/table_function/mod.rs

@@ -16,12 +16,9 @@ mod partition;
 mod pushdown_complex_filter;
 mod statistics;
 mod table_scan_progress;
-mod virtual_columns;
 
 pub use bind::*;
 pub use init::*;
-pub use virtual_columns::VirtualColumnsResult;
-pub use virtual_columns::VirtualColumnsResultRef;
 
 use crate::cpp;
 use crate::cpp::duckdb_client_context;
@@ -38,7 +35,6 @@ use crate::duckdb::table_function::partition::get_partition_data_callback;
 use crate::duckdb::table_function::pushdown_complex_filter::pushdown_complex_filter_callback;
 use crate::duckdb::table_function::statistics::statistics;
 use crate::duckdb::table_function::table_scan_progress::table_scan_progress_callback;
-use crate::duckdb::table_function::virtual_columns::get_virtual_columns_callback;
 use crate::duckdb_try;
 
 #[derive(Debug, Default)]
@@ -157,9 +153,6 @@ pub trait TableFunction: Sized + Debug {
         _local_init_data: &mut Self::LocalState,
     ) -> VortexResult<u64>;
 
-    /// Returns the virtual columns of the table function.
-    fn virtual_columns(_bind_data: &Self::BindData, _result: &mut VirtualColumnsResultRef) {}
-
     /// Returns a vector of key-value pairs for EXPLAIN output
     fn to_string(_bind_data: &Self::BindData) -> Option<Vec<(String, String)>> {
         None
@@ -209,7 +202,6 @@ impl DatabaseRef {
             cardinality: Some(cardinality_callback::<T>),
             pushdown_complex_filter: Some(pushdown_complex_filter_callback::<T>),
             pushdown_expression: ptr::null_mut::<c_void>(),
-            get_virtual_columns: Some(get_virtual_columns_callback::<T>),
             to_string: Some(to_string_callback::<T>),
             table_scan_progress: Some(table_scan_progress_callback::<T>),
             get_partition_data: Some(get_partition_data_callback::<T>),