1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
pub(super) mod single_keys;
mod single_keys_dispatch;
mod single_keys_inner;
mod single_keys_left;
mod single_keys_outer;
#[cfg(feature = "semi_anti_join")]
mod single_keys_semi_anti;
pub(super) mod sort_merge;
use arrow::array::ArrayRef;
use polars_core::utils::{_set_partition_size, split_ca};
use polars_core::POOL;
use polars_utils::index::ChunkId;
pub(super) use single_keys::*;
#[cfg(feature = "asof_join")]
pub(super) use single_keys_dispatch::prepare_bytes;
pub use single_keys_dispatch::SeriesJoin;
use single_keys_inner::*;
use single_keys_left::*;
use single_keys_outer::*;
#[cfg(feature = "semi_anti_join")]
use single_keys_semi_anti::*;
pub(crate) use sort_merge::*;

pub use super::*;
#[cfg(feature = "chunked_ids")]
use crate::chunked_array::gather::chunked::DfTake;

pub fn default_join_ids() -> ChunkJoinOptIds {
    #[cfg(feature = "chunked_ids")]
    {
        Either::Left(vec![])
    }
    #[cfg(not(feature = "chunked_ids"))]
    {
        vec![]
    }
}

macro_rules! det_hash_prone_order {
    ($self:expr, $other:expr) => {{
        // The shortest relation will be used to create a hash table.
        if $self.len() > $other.len() {
            ($self, $other, false)
        } else {
            ($other, $self, true)
        }
    }};
}

#[cfg(feature = "performant")]
use arrow::legacy::conversion::primitive_to_vec;
pub(super) use det_hash_prone_order;

pub trait JoinDispatch: IntoDf {
    /// # Safety
    /// Join tuples must be in bounds
    #[cfg(feature = "chunked_ids")]
    unsafe fn create_left_df_chunked(&self, chunk_ids: &[ChunkId], left_join: bool) -> DataFrame {
        let df_self = self.to_df();
        if left_join && chunk_ids.len() == df_self.height() {
            df_self.clone()
        } else {
            // left join keys are in ascending order
            let sorted = if left_join {
                IsSorted::Ascending
            } else {
                IsSorted::Not
            };
            df_self._take_chunked_unchecked(chunk_ids, sorted)
        }
    }

    /// # Safety
    /// Join tuples must be in bounds
    unsafe fn _create_left_df_from_slice(
        &self,
        join_tuples: &[IdxSize],
        left_join: bool,
        sorted_tuple_idx: bool,
    ) -> DataFrame {
        let df_self = self.to_df();
        if left_join && join_tuples.len() == df_self.height() {
            df_self.clone()
        } else {
            // left join tuples are always in ascending order
            let sorted = if left_join || sorted_tuple_idx {
                IsSorted::Ascending
            } else {
                IsSorted::Not
            };

            df_self._take_unchecked_slice_sorted(join_tuples, true, sorted)
        }
    }

    #[cfg(not(feature = "chunked_ids"))]
    fn _finish_left_join(
        &self,
        ids: LeftJoinIds,
        other: &DataFrame,
        args: JoinArgs,
    ) -> PolarsResult<DataFrame> {
        let ca_self = self.to_df();
        let (left_idx, right_idx) = ids;
        let materialize_left =
            || unsafe { ca_self._create_left_df_from_slice(&left_idx, true, true) };

        let materialize_right = || {
            let right_idx = &*right_idx;
            unsafe { IdxCa::with_nullable_idx(right_idx, |idx| other.take_unchecked(idx)) }
        };
        let (df_left, df_right) = POOL.join(materialize_left, materialize_right);

        _finish_join(df_left, df_right, args.suffix.as_deref())
    }

    #[cfg(feature = "chunked_ids")]
    fn _finish_left_join(
        &self,
        ids: LeftJoinIds,
        other: &DataFrame,
        args: JoinArgs,
    ) -> PolarsResult<DataFrame> {
        let ca_self = self.to_df();
        let suffix = &args.suffix;
        let (left_idx, right_idx) = ids;
        let materialize_left = || match left_idx {
            ChunkJoinIds::Left(left_idx) => unsafe {
                let mut left_idx = &*left_idx;
                if let Some((offset, len)) = args.slice {
                    left_idx = slice_slice(left_idx, offset, len);
                }
                ca_self._create_left_df_from_slice(left_idx, true, true)
            },
            ChunkJoinIds::Right(left_idx) => unsafe {
                let mut left_idx = &*left_idx;
                if let Some((offset, len)) = args.slice {
                    left_idx = slice_slice(left_idx, offset, len);
                }
                ca_self.create_left_df_chunked(left_idx, true)
            },
        };

        let materialize_right = || match right_idx {
            ChunkJoinOptIds::Left(right_idx) => unsafe {
                let mut right_idx = &*right_idx;
                if let Some((offset, len)) = args.slice {
                    right_idx = slice_slice(right_idx, offset, len);
                }
                IdxCa::with_nullable_idx(right_idx, |idx| other.take_unchecked(idx))
            },
            ChunkJoinOptIds::Right(right_idx) => unsafe {
                let mut right_idx = &*right_idx;
                if let Some((offset, len)) = args.slice {
                    right_idx = slice_slice(right_idx, offset, len);
                }
                other._take_opt_chunked_unchecked(right_idx)
            },
        };
        let (df_left, df_right) = POOL.join(materialize_left, materialize_right);

        _finish_join(df_left, df_right, suffix.as_deref())
    }

    fn _left_join_from_series(
        &self,
        other: &DataFrame,
        s_left: &Series,
        s_right: &Series,
        args: JoinArgs,
        verbose: bool,
        drop_names: Option<&[&str]>,
    ) -> PolarsResult<DataFrame> {
        let df_self = self.to_df();
        #[cfg(feature = "dtype-categorical")]
        _check_categorical_src(s_left.dtype(), s_right.dtype())?;

        let mut left = df_self.clone();
        let mut s_left = s_left.clone();
        // Eagerly limit left if possible.
        if let Some((offset, len)) = args.slice {
            if offset == 0 {
                left = left.slice(0, len);
                s_left = s_left.slice(0, len);
            }
        }

        // Ensure that the chunks are aligned otherwise we go OOB.
        let mut right = Cow::Borrowed(other);
        let mut s_right = s_right.clone();
        if left.should_rechunk() {
            left.as_single_chunk_par();
            s_left = s_left.rechunk();
        }
        if right.should_rechunk() {
            let mut other = other.clone();
            other.as_single_chunk_par();
            right = Cow::Owned(other);
            s_right = s_right.rechunk();
        }

        let ids = sort_or_hash_left(&s_left, &s_right, verbose, args.validation, args.join_nulls)?;
        let right = if let Some(drop_names) = drop_names {
            right.drop_many(drop_names)
        } else {
            right.drop(s_right.name()).unwrap()
        };
        left._finish_left_join(ids, &right, args)
    }

    #[cfg(feature = "semi_anti_join")]
    /// # Safety
    /// `idx` must be in bounds
    unsafe fn _finish_anti_semi_join(
        &self,
        mut idx: &[IdxSize],
        slice: Option<(i64, usize)>,
    ) -> DataFrame {
        let ca_self = self.to_df();
        if let Some((offset, len)) = slice {
            idx = slice_slice(idx, offset, len);
        }
        // idx from anti-semi join should always be sorted
        ca_self._take_unchecked_slice_sorted(idx, true, IsSorted::Ascending)
    }

    #[cfg(feature = "semi_anti_join")]
    fn _semi_anti_join_from_series(
        &self,
        s_left: &Series,
        s_right: &Series,
        slice: Option<(i64, usize)>,
        anti: bool,
        join_nulls: bool,
    ) -> PolarsResult<DataFrame> {
        let ca_self = self.to_df();
        #[cfg(feature = "dtype-categorical")]
        _check_categorical_src(s_left.dtype(), s_right.dtype())?;

        let idx = s_left.hash_join_semi_anti(s_right, anti, join_nulls);
        // SAFETY:
        // indices are in bounds
        Ok(unsafe { ca_self._finish_anti_semi_join(&idx, slice) })
    }
    fn _outer_join_from_series(
        &self,
        other: &DataFrame,
        s_left: &Series,
        s_right: &Series,
        args: JoinArgs,
    ) -> PolarsResult<DataFrame> {
        let df_self = self.to_df();
        #[cfg(feature = "dtype-categorical")]
        _check_categorical_src(s_left.dtype(), s_right.dtype())?;

        // Get the indexes of the joined relations
        let (mut join_idx_l, mut join_idx_r) =
            s_left.hash_join_outer(s_right, args.validation, args.join_nulls)?;

        if let Some((offset, len)) = args.slice {
            let (offset, len) = slice_offsets(offset, len, join_idx_l.len());
            join_idx_l.slice(offset, len);
            join_idx_r.slice(offset, len);
        }
        let idx_ca_l = IdxCa::with_chunk("", join_idx_l);
        let idx_ca_r = IdxCa::with_chunk("", join_idx_r);

        // Take the left and right dataframes by join tuples
        let (df_left, df_right) = POOL.join(
            || unsafe { df_self.take_unchecked(&idx_ca_l) },
            || unsafe { other.take_unchecked(&idx_ca_r) },
        );

        let JoinType::Outer { coalesce } = args.how else {
            unreachable!()
        };
        let out = _finish_join(df_left, df_right, args.suffix.as_deref());
        if coalesce {
            Ok(_coalesce_outer_join(
                out?,
                &[s_left.name()],
                &[s_right.name()],
                args.suffix.as_deref(),
                df_self,
            ))
        } else {
            out
        }
    }
}

impl JoinDispatch for DataFrame {}