eyros 2.0.0

multi-dimensional interval database
Documentation 
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

use crate::{data::DataBatch,point::Point,Value,pivots};
use crate::order::{order,order_len};
use std::cmp::Ordering;
use std::mem::size_of;
use std::rc::Rc;
use std::cell::RefCell;
use failure::{Error,bail,format_err};
use desert::ToBytes;

#[derive(Clone)]
pub enum Node<D,P,V> where D: DataBatch<P,V>, P: Point, V: Value {
  Empty,
  Branch(Branch<D,P,V>),
  Data(u64)
}

#[derive(Clone)]
pub struct Data<P,V> where P: Point, V: Value {
  pub offset: u64,
  bucket: Vec<usize>,
  rows: Rc<Vec<((P,V),u64)>>
}

#[derive(Clone)]
pub struct Branch<D,P,V> where D: DataBatch<P,V>, P: Point, V: Value {
  pub offset: u64,
  pub level: usize,
  pub index: usize,
  branch_factor: usize,
  max_data_size: usize,
  data_batch: Rc<RefCell<D>>,
  bucket: Vec<usize>,
  buckets: Vec<Vec<usize>>,
  rows: Rc<Vec<((P,V),u64)>>,
  pivots: Vec<P>,
  sorted: Vec<usize>,
  intersecting: Vec<Vec<usize>>,
  matched: Vec<bool>
}

impl<D,P,V> Branch<D,P,V> where D: DataBatch<P,V>, P: Point, V: Value {
  pub fn new (level: usize, index: usize, max_data_size: usize, bf: usize,
  data_batch: Rc<RefCell<D>>, bucket: Vec<usize>, rows: Rc<Vec<((P,V),u64)>>)
  -> Result<Self,Error> {
    let n = order_len(bf);
    let mut sorted: Vec<usize> = (0..bucket.len()).collect();
    sorted.sort_unstable_by(|a,b| {
      (rows[bucket[*a]].0).0.cmp_at(&(rows[bucket[*b]].0).0, level)
    });
    let mut pivots: Vec<P> =
      if sorted.len() == 2 {
        let a = &rows[bucket[sorted[0]]].0;
        let b = &rows[bucket[sorted[1]]].0;
        vec![a.0.midpoint_upper(&b.0)]
      } else {
        let z = n.min(sorted.len()-2);
        (0..z).map(|k| {
          let m = (k+1) * sorted.len() / (z+1);
          let a = &rows[bucket[sorted[m+0]]].0;
          let b = &rows[bucket[sorted[m+1]]].0;
          a.0.midpoint_upper(&b.0)
        }).collect()
      };
    // sometimes the sorted intervals overlap.
    // sort again to make sure the pivots are always in ascending order
    pivots.sort_unstable_by(|a,b| {
      a.cmp_at(b, level)
    });
    if pivots.is_empty() {
      bail!["empty set of pivots"]
    } else if pivots.len() == 1 {
      pivots = vec![pivots[0],pivots[0]];
      /*
      bail!["not enough data to pad pivots. need at least 2, found {}",
        pivots.len()];
      */
    } else {
      // remove duplicate pivots in already sorted vec
      let mut i = 0;
      while i < pivots.len()-1 {
        while i < pivots.len()-1
        && pivots[i].cmp_at(&pivots[i+1], level) == Ordering::Equal {
          pivots.remove(i+1);
        }
        i += 1;
      }
    }
    // pad out pivots so there's exactly n elements
    if pivots.len() < n {
      pivots = pivots::pad(&pivots, n);
    }
    let blen = bucket.len();
    Ok(Self {
      offset: 0,
      max_data_size,
      index,
      level,
      branch_factor: bf,
      data_batch,
      bucket,
      buckets: vec![vec![];bf],
      rows,
      pivots,
      sorted,
      intersecting: vec![vec![];n],
      matched: vec![false;blen]
    })
  }
  pub fn alloc (&mut self, alloc: &mut dyn FnMut (usize) -> u64) -> () {
    self.offset = alloc(self.bytes());
  }
  fn bytes (&self) -> usize {
    let mut pivot_size = 0;
    let bf = self.branch_factor;
    let n = order_len(bf);
    for p in self.pivots.iter() {
      pivot_size += p.pivot_bytes_at(self.level % P::dim());
    }
    let bitfield_size = (n + bf + 7) / 8;
    let intersect_size = n*size_of::<u64>();
    let bucket_size = bf*size_of::<u64>();
    4 + pivot_size + bitfield_size + intersect_size + bucket_size
  }
  pub fn build (&mut self, alloc: &mut dyn FnMut (usize) -> u64)
  -> Result<(Vec<u8>,Vec<Node<D,P,V>>),Error> {
    let n = order_len(self.branch_factor);
    let bf = self.branch_factor;
    for k in 0..n {
      let i = order(bf, k);
      let pivot = self.pivots[i];
      for j in self.sorted.iter() {
        let row = &self.rows[self.bucket[*j]];
        if self.matched[*j] { continue }
        if (row.0).0.cmp_at(&pivot, self.level) == Ordering::Equal {
          self.matched[*j] = true;
          self.intersecting[i].push(self.bucket[*j]);
        }
      }
    }
    for i in self.sorted.iter() {
      if self.matched[*i] { continue }
      let row = &self.rows[self.bucket[*i]];
      let mut j = 0;
      while j < bf-1 {
        let pivot = self.pivots[j*2];
        match (row.0).0.cmp_at(&pivot, self.level) {
          Ordering::Less => { break },
          Ordering::Greater => j += 1,
          Ordering::Equal => bail!["bucket interval intersects pivot"]
        }
      }
      self.buckets[j].push(self.bucket[*i]);
    }
    let mut nodes = Vec::with_capacity(bf + n);
    let mut bitfield: Vec<bool> = vec![];

    ensure_eq!(self.intersecting.len(), n, "unexpected intersecting length");
    ensure_eq!(self.buckets.len(), bf, "unexpected bucket length");
    for ref buckets in [&self.intersecting,&self.buckets].iter() {
      for bucket in buckets.iter() {
        let mut size = 0u64;
        for b in bucket.iter() { size += self.rows[*b].1 }
        if bucket.is_empty() {
          nodes.push(Node::Empty);
          bitfield.push(false);
        } else if size as usize <= self.max_data_size {
          let mut dstore = self.data_batch.try_borrow_mut()?;
          let offset = dstore.batch(&bucket.iter().map(|b| {
            &self.rows[*b].0
          }).collect())?;
          nodes.push(Node::Data(offset));
          bitfield.push(true);
        } else {
          let mut b = Branch::new(
            self.level+1,
            self.index,
            self.max_data_size,
            self.branch_factor,
            Rc::clone(&self.data_batch),
            bucket.clone(), Rc::clone(&self.rows)
          )?;
          b.alloc(alloc);
          nodes.push(Node::Branch(b));
          bitfield.push(false);
        }
      }
    }
    ensure_eq!(nodes.len(), n+bf, "incorrect number of nodes");
    ensure_eq!(self.pivots.len(), n, "incorrect number of pivots");

    let bitfield_len = (n+bf+7)/8; // in bytes
    let node_len = (n+bf) * 8; // in bytes
    let mut len = 4 + bitfield_len + node_len;
    for pivot in self.pivots.iter() {
      len += pivot.pivot_bytes_at(self.level);
    }
    let mut data = vec![0u8;len];
    let mut offset = 0;
    // length
    offset += (len as u32).write_bytes(&mut data[offset..])?;
    // pivots
    for pivot in self.pivots.iter() {
      offset += pivot.serialize_at(self.level, &mut data[offset..])?;
    }
    // data bitfield
    for i in 0..bitfield_len {
      let mut byte = 0u8;
      for j in 0..8.min(n+bf-i*8) {
        byte += (1 << j) * (bitfield[i*8+j] as u8);
      }
      data[offset] = byte;
      offset += 1;
    }
    // intersecting + buckets
    for node in nodes.iter() {
      offset += match node {
        Node::Branch(b) => b.offset+1,
        Node::Data(d) => *d+1,
        Node::Empty => 0u64
      }.write_bytes(&mut data[offset..])?;
    }
    Ok((data,nodes))
  }
}