gnitz-zset 0.1.3

The Z-set kernel of the gnitz database: schema, columnar batches, cursors and operators
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

use super::*;
use crate::repr::{Batch, BatchBuilder};
use crate::schema::{ColumnTable, SchemaColumn, TypeCode};
use crate::test_support::{arb_fold_case, cell, fold_batch, fold_schemas, le_cell, opk_pk, pk_only_schema};
use proptest::prelude::*;

// ---------------------------------------------------------------------------
// Group key
// ---------------------------------------------------------------------------

/// A single natural column keys by its OPK image — at its own width as the
/// output PK — whether it is a PK or a payload column.
#[test]
fn single_natural_col_keys_by_its_opk_from_either_side() {
    // `(identity, out_pk)` of the value `le` held in column 1 of `[U64 pk, <tc>]`,
    // with column 1 either a second PK column or the sole payload column.
    let key = |tc: TypeCode, le: &[u8], col1_is_pk: bool| -> (u128, Vec<u8>) {
        let cols = [SchemaColumn::new(TypeCode::U64, false), SchemaColumn::new(tc, false)];
        let schema = SchemaDescriptor::new(&cols, if col1_is_pk { &[0, 1] } else { &[0] });
        let mut b = BatchBuilder::new(&schema);
        match col1_is_pk {
            true => b.begin_row_natives(&[0, le_cell(le)], 1),
            false => {
                b.begin_row(0, 1);
                b.put_int(le_cell(le));
            }
        }
        b.end_row();
        let b = b.finish();
        let (key, _) = GroupOutKey::new(&schema, &[1], []).unwrap();
        let mb = b.as_mem_batch();
        (key.identity(&mb, 0), key.out_pk(&mb, 0).bytes().to_vec())
    };

    for (tc, vals) in [
        (TypeCode::I32, vec![1i128, -1, 100, i32::MIN as i128, i32::MAX as i128]),
        (TypeCode::I64, vec![0, -1, i64::MIN as i128, i64::MAX as i128]),
        (TypeCode::U16, vec![0, 1, 0xBEEF, u16::MAX as i128]),
        (TypeCode::U64, vec![0, 1, u64::MAX as i128]),
    ] {
        let width = SchemaColumn::new(tc, false).size() as usize;
        for v in vals {
            let le = &(v as u128).to_le_bytes()[..width];
            let want_pk = opk_pk(&pk_only_schema(&[tc]), &[v as u128]);
            let want = (gnitz_wire::widen_pk_be(&want_pk), want_pk);
            assert_eq!(key(tc, le, true), want, "PK-column key for {tc} v={v}");
            assert_eq!(key(tc, le, false), want, "payload-column key for {tc} v={v}");
        }
    }
}

// ---------------------------------------------------------------------------
// Group runs and ordinals
// ---------------------------------------------------------------------------

/// Sorting numbers the groups in ascending key order, and a group's first row is
/// its first in the batch: the whole `(key, index)` pair is the sort key, so the
/// index breaks every tie.
#[test]
fn sorted_ordinals_number_groups_by_key_and_keep_the_first_row() {
    // Three keys, descending, four rows each, interleaved.
    let groups = GroupOrdinals::sorted(12, |i| 2 - i % 3);
    assert_eq!(groups.ord, [2, 1, 0, 2, 1, 0, 2, 1, 0, 2, 1, 0]);
    assert_eq!(groups.first, [2, 1, 0]);
    assert_eq!(groups.by_pk, [0, 1, 2]);
    // Every row one key: one group.
    let groups = GroupOrdinals::sorted(5, |_| 0u64);
    assert_eq!((groups.ord, groups.first, groups.by_pk), (vec![0; 5], vec![0], vec![0]));
}

/// Under every key form — the empty set, every leading run of the PK list, each
/// single column, each pair of columns and all of them — over `raw` and its
/// consolidation: `ordinals` gives two rows one ordinal iff they share an
/// `out_pk`, names each group's first row, and lists the groups in strictly
/// ascending `out_pk` order; two rows share an `out_pk` iff they share their
/// group columns' values, and iff they share an identity. `runs`, where it
/// answers, cuts the rows at exactly the ordinals' boundaries, and it answers
/// for a consolidated batch grouped by leading PK columns whose bytes are its
/// output PK.
fn assert_groups_follow_out_pk(raw: &Batch) -> Result<(), TestCaseError> {
    let schema = *raw.schema();
    let consolidated = Batch::clone(raw).into_consolidated();
    let pk: Vec<u32> = schema.pk_cols().to_vec();
    let all: Vec<u32> = (0..schema.num_columns() as u32).collect();
    let n = schema.num_columns() as u32;
    let mut forms = vec![vec![], all];
    forms.extend((1..=pk.len()).map(|lead| pk[..lead].to_vec()));
    forms.extend((0..n).map(|c| vec![c]));
    forms.extend((0..n).flat_map(|a| (0..n).filter(move |&b| b != a).map(move |b| vec![a, b])));
    for cols in &forms {
        let (key, _) = GroupOutKey::new(&schema, cols, []).unwrap();
        for batch in [raw, &consolidated] {
            let mb = batch.as_mem_batch();
            let out_pk = |row: usize| key.out_pk(&mb, row).bytes().to_vec();
            let groups = key.ordinals(batch);
            prop_assert_eq!(groups.ord.len(), batch.count, "{:?}", cols);
            for (row, &g) in groups.ord.iter().enumerate() {
                let first = groups.first[g as usize] as usize;
                prop_assert!(first <= row, "{:?}: row {} precedes its group's first", cols, row);
                prop_assert_eq!(out_pk(first), out_pk(row), "{:?}: a group mixes keys", cols);
            }
            let mut seen = groups.by_pk.clone();
            seen.sort_unstable();
            prop_assert_eq!(seen, (0..groups.len() as u32).collect::<Vec<_>>(), "{:?}", cols);
            let pks: Vec<Vec<u8>> = groups
                .by_pk
                .iter()
                .map(|&g| out_pk(groups.first[g as usize] as usize))
                .collect();
            prop_assert!(pks.windows(2).all(|w| w[0] < w[1]), "{:?}: groups out of order", cols);

            let runs = key.runs(batch);
            if let Some(runs) = &runs {
                prop_assert_eq!(runs.len(), groups.len(), "{:?}", cols);
                for (g, run) in runs.iter().enumerate() {
                    prop_assert!(
                        run.clone().all(|row| groups.ord[row] == g as u32),
                        "{:?}: run {} is not group {}",
                        cols,
                        g,
                        g
                    );
                }
            }
            // A longer proper prefix is a fold, which no PK order groups.
            let bytes: usize = cols.iter().map(|&c| schema.columns[c as usize].size() as usize).sum();
            let in_place = *cols == pk || bytes <= GROUP_PK_BYTES;
            if std::ptr::eq(batch, &consolidated) && !cols.is_empty() && pk.starts_with(cols) && in_place {
                prop_assert!(runs.is_some(), "{:?}: a PK prefix of a consolidated batch", cols);
            }

            let group =
                |r: usize| -> Vec<_> { cols.iter().map(|&c| cell(&mb, schema.locate(c as usize), r)).collect() };
            for a in 0..batch.count {
                for b in a + 1..batch.count {
                    let same_out_pk = out_pk(a) == out_pk(b);
                    prop_assert_eq!(group(a) == group(b), same_out_pk, "{:?}: rows {} and {}", cols, a, b);
                    prop_assert_eq!(
                        groups.ord[a] == groups.ord[b],
                        same_out_pk,
                        "{:?}: rows {} and {}",
                        cols,
                        a,
                        b
                    );
                    prop_assert_eq!(
                        key.identity(&mb, a) == key.identity(&mb, b),
                        same_out_pk,
                        "{:?}: rows {} and {}",
                        cols,
                        a,
                        b
                    );
                }
            }
        }
    }
    Ok(())
}

proptest! {
    /// [`assert_groups_follow_out_pk`] over every PK width arm.
    #[test]
    fn groups_follow_out_pk_at_every_pk_width((si, rows) in arb_fold_case()) {
        assert_groups_follow_out_pk(&fold_batch(&fold_schemas()[si], &rows))?;
    }
}

/// [`assert_groups_follow_out_pk`] over a compound signed PK and signed, wide
/// and nullable payload columns, each holding repeated values: 64 rows, so both
/// the hashed and the sorted numbering are taken, by group count.
#[test]
fn groups_follow_out_pk_over_signed_wide_and_nullable_columns() {
    // `[U32 pk0, I32 pk1, I64 v, U128 w, I64 n NULL]`, PK `(pk0, pk1)`.
    let schema = SchemaDescriptor::new(
        &[
            SchemaColumn::new(TypeCode::U32, false),
            SchemaColumn::new(TypeCode::I32, false),
            SchemaColumn::new(TypeCode::I64, false),
            SchemaColumn::new(TypeCode::U128, false),
            SchemaColumn::new(TypeCode::I64, true),
        ],
        &[0, 1],
    );
    let mut bb = BatchBuilder::new(&schema);
    for i in 0..64u64 {
        let m = i.wrapping_mul(0x9E37_79B9_7F4A_7C15) >> 40;
        bb.begin_row_natives(&[(m % 5) as u128, (m % 7) as i32 as i64 as u128], 1);
        bb.put_int(((m % 9) as i64 - 4) as u128);
        bb.put_int(u128::from(m % 3) << 100);
        match m % 4 {
            0 => bb.put_null(),
            k => bb.put_int((k as i64 - 2) as u128),
        }
        bb.end_row();
    }
    assert_groups_follow_out_pk(&bb.finish()).unwrap();
}