polars_ops/frame/join/
general.rs

Help
1use polars_utils::format_pl_smallstr;
2
3use super::*;
4use crate::series::coalesce_columns;
5
6pub fn _join_suffix_name(name: &str, suffix: &str) -> PlSmallStr {
7    format_pl_smallstr!("{name}{suffix}")
8}
9
10fn get_suffix(suffix: Option<PlSmallStr>) -> PlSmallStr {
11    suffix.unwrap_or_else(|| PlSmallStr::from_static("_right"))
12}
13
14/// Renames the columns on the right to not clash with the left using a specified or otherwise default suffix
15/// and then merges the right dataframe into the left
16#[doc(hidden)]
17pub fn _finish_join(
18    mut df_left: DataFrame,
19    mut df_right: DataFrame,
20    suffix: Option<PlSmallStr>,
21) -> PolarsResult<DataFrame> {
22    let mut left_names = PlHashSet::with_capacity(df_left.width());
23
24    df_left.get_columns().iter().for_each(|series| {
25        left_names.insert(series.name());
26    });
27
28    let mut rename_strs = Vec::with_capacity(df_right.width());
29    let right_names = df_right.schema();
30
31    for name in right_names.iter_names() {
32        if left_names.contains(name) {
33            rename_strs.push(name.clone())
34        }
35    }
36
37    let suffix = get_suffix(suffix);
38
39    for name in rename_strs {
40        let new_name = _join_suffix_name(name.as_str(), suffix.as_str());
41        // Safety: IR resolving should guarantee this passes
42        df_right.rename(&name, new_name.clone()).unwrap();
43    }
44
45    drop(left_names);
46    // Safety: IR resolving should guarantee this passes
47    unsafe { df_left.hstack_mut_unchecked(df_right.get_columns()) };
48    Ok(df_left)
49}
50
51pub fn _coalesce_full_join(
52    mut df: DataFrame,
53    keys_left: &[PlSmallStr],
54    keys_right: &[PlSmallStr],
55    suffix: Option<PlSmallStr>,
56    df_left: &DataFrame,
57) -> DataFrame {
58    // No need to allocate the schema because we already
59    // know for certain that the column name for left is `name`
60    // and for right is `name + suffix`
61    let schema_left = if keys_left == keys_right {
62        Arc::new(Schema::default())
63    } else {
64        df_left.schema().clone()
65    };
66
67    let schema = df.schema().clone();
68    let mut to_remove = Vec::with_capacity(keys_right.len());
69
70    // SAFETY: we maintain invariants.
71    let columns = unsafe { df.get_columns_mut() };
72    let suffix = get_suffix(suffix);
73    for (l, r) in keys_left.iter().zip(keys_right.iter()) {
74        let pos_l = schema.get_full(l.as_str()).unwrap().0;
75
76        let r = if l == r || schema_left.contains(r.as_str()) {
77            _join_suffix_name(r.as_str(), suffix.as_str())
78        } else {
79            r.clone()
80        };
81        let pos_r = schema.get_full(&r).unwrap().0;
82
83        let l = columns[pos_l].clone();
84        let r = columns[pos_r].clone();
85
86        columns[pos_l] = coalesce_columns(&[l, r]).unwrap();
87        to_remove.push(pos_r);
88    }
89    // sort in reverse order, so the indexes remain correct if we remove.
90    to_remove.sort_by(|a, b| b.cmp(a));
91    for pos in to_remove {
92        let _ = columns.remove(pos);
93    }
94    df.clear_schema();
95    df
96}
97
98#[cfg(feature = "chunked_ids")]
99pub(crate) fn create_chunked_index_mapping(chunks: &[ArrayRef], len: usize) -> Vec<ChunkId> {
100    let mut vals = Vec::with_capacity(len);
101
102    for (chunk_i, chunk) in chunks.iter().enumerate() {
103        vals.extend(
104            (0..chunk.len()).map(|array_i| ChunkId::store(chunk_i as IdxSize, array_i as IdxSize)),
105        )
106    }
107
108    vals
109}
polars_ops/frame/join/general.rs

polars_ops/frame/join/
general.rs
