16398874017

Committed 20 Jul 2025 10:24AM UTC coverage: 81.523% (+0.01%) from 81.512%

Build # 16398874017

Build Type

push

github

Committed by

web-flow

Commit Message

Add cardinality callback to DuckDB TableFunction (#3941)

This helps DuckDB to produce far better joins

Signed-off-by: Nicholas Gates <nick@nickgates.com>

Run Details

24 of 28 new or added lines in 3 files covered. (85.71%)

2 existing lines in 1 file now uncovered.

42047 of 51577 relevant lines covered (81.52%)

171533.82 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

76.47

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

// SPDX-License-Identifier: Apache-2.0
// SPDX-FileCopyrightText: Copyright the Vortex contributors

use std::ffi::{CStr, CString, c_void};
use std::fmt::Debug;
use std::ptr;

use vortex::error::{VortexExpect, VortexResult};
mod bind;
mod cardinality;
mod init;
mod pushdown_complex_filter;

pub use bind::*;
pub use init::*;

use crate::duckdb::LogicalType;
use crate::duckdb::connection::Connection;
use crate::duckdb::data_chunk::DataChunk;
use crate::duckdb::expr::Expression;
use crate::duckdb::table_function::cardinality::cardinality_callback;
use crate::duckdb::table_function::pushdown_complex_filter::pushdown_complex_filter_callback;
use crate::{cpp, duckdb_try};

/// A trait that defines the supported operations for a table function in DuckDB.
///
/// This trait does not yet cover the full C++ API, see table_function.hpp.
pub trait TableFunction: Sized + Debug {
    type BindData: Send + Clone;
    type GlobalState: Send + Sync;
    type LocalState;

    /// Whether the table function supports projection pushdown.
    /// If not supported a projection will be added that filters out unused columns.
    const PROJECTION_PUSHDOWN: bool = false;

    /// Whether the table function supports filter pushdown.
    /// If not supported a filter will be added that applies the table filter directly.
    const FILTER_PUSHDOWN: bool = false;

    /// Whether the table function can immediately prune out filter columns that are unused
    /// in the remainder of the query plan.
    /// e.g. "SELECT i FROM tbl WHERE j = 42;"
    ///   - j does not need to leave the table function at all.
    const FILTER_PRUNE: bool = false;

    /// Returns the parameters of the table function.
    fn parameters() -> Vec<LogicalType> {
        // By default, we don't have any parameters.
        vec![]
    }

    /// Returns the named parameters of the table function, if any.
    fn named_parameters() -> Vec<(CString, LogicalType)> {
        // By default, we don't have any named parameters.
        vec![]
    }

    /// This function is used for determining the schema of a table producing function and
    /// returning bind data.
    fn bind(input: &BindInput, result: &mut BindResult) -> VortexResult<Self::BindData>;

    /// The function is called during query execution and is responsible for producing the output
    fn scan(
        bind_data: &Self::BindData,
        init_local: &mut Self::LocalState,
        init_global: &mut Self::GlobalState,
        chunk: &mut DataChunk,
    ) -> VortexResult<()>;

    /// Initialize the global operator state of the function.
    ///
    /// The global operator state is used to keep track of the progress in the table function and
    /// is shared between all threads working on the table function.
    fn init_global(input: &TableInitInput<Self>) -> VortexResult<Self::GlobalState>;

    /// Initialize the local operator state of the function.
    ///
    /// The local operator state is used to keep track of the progress in the table function and
    /// is thread-local.
    fn init_local(
        init: &TableInitInput<Self>,
        global: &mut Self::GlobalState,
    ) -> VortexResult<Self::LocalState>;

    /// Pushes down a filter expression to the table function.
    ///
    /// Returns `true` if the filter was successfully pushed down (and stored on the bind data),
    /// or `false` if the filter could not be pushed down. In which case, the filter will be
    /// applied later in the query plan.
    fn pushdown_complex_filter(
        _bind_data: &mut Self::BindData,
        _expr: &Expression,
    ) -> VortexResult<bool> {
        Ok(false)
    }

    /// Returns the cardinality estimate of the table function.
    fn cardinality(_bind_data: &Self::BindData) -> Cardinality {
        Cardinality::Unknown
    }

    // TODO(ngates): there are many more callbacks that can be configured.
}

pub enum Cardinality {
    /// Completely unknown cardinality.
    Unknown,
    /// An estimate of the number of rows that will be returned by the table function.
    Estimate(u64),
    /// Will not return more than this number of rows.
    Maximum(u64),
}

impl Connection {
    pub fn register_table_function<T: TableFunction>(&self, name: &CStr) -> VortexResult<()> {
        // Set up the parameters.
        let parameters = T::parameters();
        let parameter_ptrs = parameters
            .iter()
            .map(|logical_type| logical_type.as_ptr())
            .collect::<Vec<_>>();

        let param_names = T::named_parameters();
        let (param_names_ptrs, param_types_ptr) = param_names
            .into_iter()
            .map(|(name, logical_type)| (name.as_ptr(), logical_type.as_ptr()))
            .unzip::<_, _, Vec<_>, Vec<_>>();

        let vtab = cpp::duckdb_vx_tfunc_vtab_t {
            name: name.as_ptr(),
            parameters: parameter_ptrs.as_ptr(),
            parameter_count: parameters.len() as _,
            named_parameter_names: param_names_ptrs.as_ptr(),
            named_parameter_types: param_types_ptr.as_ptr(),
            named_parameter_count: param_names_ptrs.len() as _,
            bind: Some(bind_callback::<T>),
            bind_data_clone: Some(bind_data_clone_callback::<T>),
            init_global: Some(init_global_callback::<T>),
            init_local: Some(init_local_callback::<T>),
            function: Some(function::<T>),
            statistics: ptr::null_mut::<c_void>(),
            cardinality: Some(cardinality_callback::<T>),
            pushdown_complex_filter: Some(pushdown_complex_filter_callback::<T>),
            pushdown_expression: ptr::null_mut::<c_void>(),
            table_scan_progress: ptr::null_mut::<c_void>(),
            projection_pushdown: T::PROJECTION_PUSHDOWN,
            filter_pushdown: T::FILTER_PUSHDOWN,
            filter_prune: T::FILTER_PRUNE,
            sampling_pushdown: false,
            late_materialization: false,
        };

        duckdb_try!(
            unsafe { cpp::duckdb_vx_tfunc_register(self.as_ptr(), &raw const vtab) },
            "Failed to register table function '{}'",
            name.to_string_lossy()
        );

        Ok(())
    }
}

/// The native function callback for a table function.
unsafe extern "C" fn function<T: TableFunction>(
    bind_data: *const c_void,
    global_init_data: *mut c_void,
    local_init_data: *mut c_void,
    output: cpp::duckdb_data_chunk,
    error_out: *mut cpp::duckdb_vx_error,
) {
    let bind_data = unsafe { &*(bind_data as *const T::BindData) };
    let global_init_data = unsafe { global_init_data.cast::<T::GlobalState>().as_mut() }
        .vortex_expect("global_init_data null pointer");
    let local_init_data = unsafe { local_init_data.cast::<T::LocalState>().as_mut() }
        .vortex_expect("local_init_data null pointer");
    let mut data_chunk = unsafe { DataChunk::borrow(output) };

    match T::scan(
        bind_data,
        local_init_data,
        global_init_data,
        &mut data_chunk,
    ) {
        Ok(()) => {
            // The data chunk is already filled by the function.
            // No need to do anything here.
        }
        Err(e) => unsafe {
            error_out.write(cpp::duckdb_vx_error_create(
                e.to_string().as_ptr().cast(),
                e.to_string().len(),
            ));
        },
    }
}

1	// SPDX-License-Identifier: Apache-2.0
2	// SPDX-FileCopyrightText: Copyright the Vortex contributors
3
4	use std::ffi::{CStr, CString, c_void};
5	use std::fmt::Debug;
6	use std::ptr;
7
8	use vortex::error::{VortexExpect, VortexResult};
9	mod bind;
10	mod cardinality;
11	mod init;
12	mod pushdown_complex_filter;
13
14	pub use bind::*;
15	pub use init::*;
16
17	use crate::duckdb::LogicalType;
18	use crate::duckdb::connection::Connection;
19	use crate::duckdb::data_chunk::DataChunk;
20	use crate::duckdb::expr::Expression;
21	use crate::duckdb::table_function::cardinality::cardinality_callback;
22	use crate::duckdb::table_function::pushdown_complex_filter::pushdown_complex_filter_callback;
23	use crate::{cpp, duckdb_try};
24
25	/// A trait that defines the supported operations for a table function in DuckDB.
26	///
27	/// This trait does not yet cover the full C++ API, see table_function.hpp.
28	pub trait TableFunction: Sized + Debug {
29	type BindData: Send + Clone;
30	type GlobalState: Send + Sync;
31	type LocalState;
32
33	/// Whether the table function supports projection pushdown.
34	/// If not supported a projection will be added that filters out unused columns.
35	const PROJECTION_PUSHDOWN: bool = false;
36
37	/// Whether the table function supports filter pushdown.
38	/// If not supported a filter will be added that applies the table filter directly.
39	const FILTER_PUSHDOWN: bool = false;
40
41	/// Whether the table function can immediately prune out filter columns that are unused
42	/// in the remainder of the query plan.
43	/// e.g. "SELECT i FROM tbl WHERE j = 42;"
44	/// - j does not need to leave the table function at all.
45	const FILTER_PRUNE: bool = false;
46
47	/// Returns the parameters of the table function.
48	fn parameters() -> Vec<LogicalType> {	×
49	// By default, we don't have any parameters.
50	vec![]	×
51	}	×
52
53	/// Returns the named parameters of the table function, if any.
54	fn named_parameters() -> Vec<(CString, LogicalType)> {	36✔
55	// By default, we don't have any named parameters.
56	vec![]	36✔
57	}	36✔
58
59	/// This function is used for determining the schema of a table producing function and
60	/// returning bind data.
61	fn bind(input: &BindInput, result: &mut BindResult) -> VortexResult<Self::BindData>;
62
63	/// The function is called during query execution and is responsible for producing the output
64	fn scan(
65	bind_data: &Self::BindData,
66	init_local: &mut Self::LocalState,
67	init_global: &mut Self::GlobalState,
68	chunk: &mut DataChunk,
69	) -> VortexResult<()>;
70
71	/// Initialize the global operator state of the function.
72	///
73	/// The global operator state is used to keep track of the progress in the table function and
74	/// is shared between all threads working on the table function.
75	fn init_global(input: &TableInitInput<Self>) -> VortexResult<Self::GlobalState>;
76
77	/// Initialize the local operator state of the function.
78	///
79	/// The local operator state is used to keep track of the progress in the table function and
80	/// is thread-local.
81	fn init_local(
82	init: &TableInitInput<Self>,
83	global: &mut Self::GlobalState,
84	) -> VortexResult<Self::LocalState>;
85
86	/// Pushes down a filter expression to the table function.
87	///
88	/// Returns `true` if the filter was successfully pushed down (and stored on the bind data),
89	/// or `false` if the filter could not be pushed down. In which case, the filter will be
90	/// applied later in the query plan.
91	fn pushdown_complex_filter(	×
92	_bind_data: &mut Self::BindData,	×
93	_expr: &Expression,	×
94	) -> VortexResult<bool> {	×
95	Ok(false)	×
96	}	×
97
98	/// Returns the cardinality estimate of the table function.
NEW 99	fn cardinality(_bind_data: &Self::BindData) -> Cardinality {	×
NEW 100	Cardinality::Unknown	×
NEW 101	}	×
102
103	// TODO(ngates): there are many more callbacks that can be configured.
104	}
105
106	pub enum Cardinality {
107	/// Completely unknown cardinality.
108	Unknown,
109	/// An estimate of the number of rows that will be returned by the table function.
110	Estimate(u64),
111	/// Will not return more than this number of rows.
112	Maximum(u64),
113	}
114
115	impl Connection {
116	pub fn register_table_function<T: TableFunction>(&self, name: &CStr) -> VortexResult<()> {	36✔
117	// Set up the parameters.
118	let parameters = T::parameters();	36✔
119	let parameter_ptrs = parameters	36✔
120	.iter()	36✔
121	.map(\|logical_type\| logical_type.as_ptr())	36✔
122	.collect::<Vec<_>>();	36✔
123
124	let param_names = T::named_parameters();	36✔
125	let (param_names_ptrs, param_types_ptr) = param_names	36✔
126	.into_iter()	36✔
127	.map(\|(name, logical_type)\| (name.as_ptr(), logical_type.as_ptr()))	36✔
128	.unzip::<_, _, Vec<_>, Vec<_>>();	36✔
129
130	let vtab = cpp::duckdb_vx_tfunc_vtab_t {	36✔
131	name: name.as_ptr(),	36✔
132	parameters: parameter_ptrs.as_ptr(),	36✔
133	parameter_count: parameters.len() as _,	36✔
134	named_parameter_names: param_names_ptrs.as_ptr(),	36✔
135	named_parameter_types: param_types_ptr.as_ptr(),	36✔
136	named_parameter_count: param_names_ptrs.len() as _,	36✔
137	bind: Some(bind_callback::<T>),	36✔
138	bind_data_clone: Some(bind_data_clone_callback::<T>),	36✔
139	init_global: Some(init_global_callback::<T>),	36✔
140	init_local: Some(init_local_callback::<T>),	36✔
141	function: Some(function::<T>),	36✔
142	statistics: ptr::null_mut::<c_void>(),	36✔
143	cardinality: Some(cardinality_callback::<T>),	36✔
144	pushdown_complex_filter: Some(pushdown_complex_filter_callback::<T>),	36✔
145	pushdown_expression: ptr::null_mut::<c_void>(),	36✔
146	table_scan_progress: ptr::null_mut::<c_void>(),	36✔
147	projection_pushdown: T::PROJECTION_PUSHDOWN,	36✔
148	filter_pushdown: T::FILTER_PUSHDOWN,	36✔
149	filter_prune: T::FILTER_PRUNE,	36✔
150	sampling_pushdown: false,	36✔
151	late_materialization: false,	36✔
152	};	36✔
153
154	duckdb_try!(	36✔
155	unsafe { cpp::duckdb_vx_tfunc_register(self.as_ptr(), &raw const vtab) },	36✔
156	"Failed to register table function '{}'",	×
157	name.to_string_lossy()	×
158	);
159
160	Ok(())	36✔
161	}	36✔
162	}
163
164	/// The native function callback for a table function.
165	unsafe extern "C" fn function<T: TableFunction>(	10,666✔
166	bind_data: *const c_void,	10,666✔
167	global_init_data: *mut c_void,	10,666✔
168	local_init_data: *mut c_void,	10,666✔
169	output: cpp::duckdb_data_chunk,	10,666✔
170	error_out: *mut cpp::duckdb_vx_error,	10,666✔
171	) {	10,666✔
172	let bind_data = unsafe { &(bind_data as const T::BindData) };	10,666✔
173	let global_init_data = unsafe { global_init_data.cast::<T::GlobalState>().as_mut() }	10,666✔
174	.vortex_expect("global_init_data null pointer");	10,666✔
175	let local_init_data = unsafe { local_init_data.cast::<T::LocalState>().as_mut() }	10,666✔
176	.vortex_expect("local_init_data null pointer");	10,666✔
177	let mut data_chunk = unsafe { DataChunk::borrow(output) };	10,666✔
178
179	match T::scan(	10,666✔
180	bind_data,	10,666✔
181	local_init_data,	10,666✔
182	global_init_data,	10,666✔
183	&mut data_chunk,	10,666✔
184	) {	10,666✔
185	Ok(()) => {	10,666✔
186	// The data chunk is already filled by the function.	10,666✔
187	// No need to do anything here.	10,666✔
188	}	10,666✔
189	Err(e) => unsafe {	×
190	error_out.write(cpp::duckdb_vx_error_create(	×
191	e.to_string().as_ptr().cast(),	×
192	e.to_string().len(),	×
193	));	×
194	},	×
195	}
196	}	10,666✔

vortex-data / vortex / 16398874017

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous