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
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
use avl::{Tree, Iter};
use std::{
    cmp::{PartialEq, Eq, PartialOrd, Ord, Ordering},
    fmt::{self, Debug, Formatter}, borrow::Borrow,
    ops::{Bound, Index}, default::Default,
    hash::{Hash, Hasher}, iter::FromIterator
};

/// This Map uses a similar strategy to BTreeMap to ensure cache
/// efficient performance on modern hardware while still providing
/// log(N) get, insert, and remove operations.
///
/// For good performance, it is very important to understand
/// that clone is a fundamental operation, it needs to be fast
/// for your key and data types, because it's going to be
/// called a lot whenever you change the map. If your key and
/// data types are cheap to clone (like e.g. Arc), and you
/// perform your update operations in largish batches, then it
/// is possible to get very good performance, even approaching
/// that of a standard HashMap.
///
/// # Why
///
/// Which begs the question, why would anyone ever want to use
/// a data structure where very careful structuring of key and
/// data type, and careful batching, MIGHT APPROACH the
/// performance of a plain old HashMap, it seems a silly thing
/// to work on. I know of two cases.
///
/// 1. Multiple threads can read this structure even while one
/// thread is updating it.
///
/// 2. You can take a snapshot and e.g. write it to disk, or
/// replicate it to another process without stopping reads or
/// writes.
///
/// There is some nuance to #1, because HashMap is generally
/// faster to read than a BTree. In a pure read application
/// it's the obvious choice when you don't require sorted
/// data. In a mixed read/write scenario at 4 reads for every
/// write HashMap is already the same speed as chunkmap for
/// reading a 10M entry map. That's a pretty write heavy
/// application, and wouldn't be news by itself. The real
/// killer of the mutable strucures is large operations, any
/// kind of housekeeping operation that's going to touch a
/// large number of keys can be death for availability,
/// holding onto a write lock for multiple hundreds of
/// milliseconds, even seconds, even longer. Sure it's
/// possible to amortize this in some cases by doing your
/// housekeeping in small batches, but that can be complex,
/// and it isn't always possible, and readers still pay a
/// price even if it's amortized.
///
/// That brings us to #2, which is really the mother of all
/// housekeeping operations. There is no way to amortize
/// taking a consistent snapshot, the best you can possibly do
/// is hold the write lock long enough to make a complete copy
/// of the data, if you even have the memory for that. God
/// help you if you miscalculate and start swapping while
/// you're making that copy, holding the write lock while your
/// disk or if you're lucky SSD churns away moving pages back
/// and forth between main memory, you may be holding that
/// lock for a long long time. Chunkmap gives you free
/// snapshots in exchange for slower writes, which, carefully
/// considered don't even need to be that much slower.
///
/// # Examples
/// ```
/// use std::string::String;
/// use self::immutable_chunkmap::map::Map;
///
/// let m =
///    Map::new()
///    .insert(String::from("1"), 1).0
///    .insert(String::from("2"), 2).0
///    .insert(String::from("3"), 3).0;
///
/// assert_eq!(m.get("1"), Option::Some(&1));
/// assert_eq!(m.get("2"), Option::Some(&2));
/// assert_eq!(m.get("3"), Option::Some(&3));
/// assert_eq!(m.get("4"), Option::None);
///
/// for (k, v) in &m {
///   println!("key {}, val: {}", k, v)
/// }
/// ```
#[derive(Clone)]
pub struct Map<K: Ord + Clone, V: Clone> {
    len: usize,
    root: Tree<K, V>
}

impl<K, V> Hash for Map<K, V>
where K: Hash + Ord + Clone, V: Hash + Clone {
    fn hash<H: Hasher>(&self, state: &mut H) {
        self.root.hash(state)
    }
}

impl<K, V> Default for Map<K, V>
where K: Ord + Clone, V: Clone {
    fn default() -> Map<K, V> { Map::new() }
}

impl<K, V> PartialEq for Map<K, V>
where K: PartialEq + Ord + Clone, V: PartialEq + Clone {
    fn eq(&self, other: &Map<K,V>) -> bool {
        self.len == other.len && self.root == other.root
    }
}

impl<K, V> Eq for Map<K, V>
where K: Eq + Ord + Clone, V: Eq + Clone {}

impl<K, V> PartialOrd for Map<K, V>
where K: Ord + Clone, V: PartialOrd + Clone {
    fn partial_cmp(&self, other: &Map<K, V>) -> Option<Ordering> {
        self.root.partial_cmp(&other.root)
    }
}

impl<K, V> Ord for Map<K, V>
where K: Ord + Clone, V: Ord + Clone {
    fn cmp(&self, other: &Map<K, V>) -> Ordering {
        self.root.cmp(&other.root)
    }
}

impl<K, V> Debug for Map<K, V>
where K: Debug + Ord + Clone, V: Debug + Clone {
    fn fmt(&self, f: &mut Formatter) -> fmt::Result { self.root.fmt(f) }
}

impl<'a, Q, K, V> Index<&'a Q> for Map<K, V>
where Q: Ord, K: Ord + Clone + Borrow<Q>, V: Clone {
    type Output = V;
    fn index(&self, k: &Q) -> &V {
        self.get(k).expect("element not found for key")
    }
}

impl<K, V> FromIterator<(K, V)> for Map<K, V>
where K: Ord + Clone, V: Clone {
    fn from_iter<T: IntoIterator<Item=(K, V)>>(iter: T) -> Self {
        Map::new().insert_many(iter)
    }
}

impl<'a, K, V> IntoIterator for &'a Map<K, V>
where
    K: 'a + Borrow<K> + Ord + Clone,
    V: 'a + Clone
{
    type Item = (&'a K, &'a V);
    type IntoIter = Iter<'a, K, K, V>;
    fn into_iter(self) -> Self::IntoIter { self.root.into_iter() }
}

impl<K, V> Map<K, V> where K: Ord + Clone, V: Clone {
    /// Create a new empty map
    pub fn new() -> Self { Map { len: 0, root: Tree::new() } }

    /// This will insert many elements at once, and is
    /// potentially a lot faster than inserting one by one,
    /// especially if the data is sorted. It is just a wrapper
    /// around the more general update_many method.
    ///
    /// #Examples
    ///```
    /// use self::immutable_chunkmap::map::Map;
    ///
    /// let mut v = vec![(1, 3), (10, 1), (-12, 2), (44, 0), (50, -1)];
    /// v.sort_unstable_by_key(|&(k, _)| k);
    ///
    /// let m = Map::new().insert_many(v.iter().map(|(k, v)| (*k, *v)));
    ///
    /// for (k, v) in &v {
    ///   assert_eq!(m.get(k), Option::Some(v))
    /// }
    /// ```
    pub fn insert_many<E: IntoIterator<Item=(K, V)>>(
        &self, elts: E
    ) -> Self {
        let (root, len) = self.root.insert_many(self.len, elts);
        Map { len, root }
    }

    /// This method updates multiple bindings in one call. Given an
    /// iterator of an arbitrary type (Q, D), where Q is any borrowed
    /// form of K, an update function taking Q, D, the current binding
    /// in the map, if any, and producing the new binding, if any,
    /// this method will produce a new map with updated bindings of
    /// many elements at once. It will skip intermediate node
    /// allocations where possible. If the data in elts is sorted, it
    /// will be able to skip many more intermediate allocations, and
    /// can produce a speedup of about 10x compared to
    /// inserting/updating one by one. In any case it should always be
    /// faster than inserting elements one by one, even with random
    /// unsorted keys.
    ///
    /// #Examples
    /// ```
    /// use self::immutable_chunkmap::map::Map;
    ///
    /// let m = Map::new().insert_many((0..4).map(|k| (k, k)));
    /// let m = m.update_many(
    ///     (0..4).map(|x| (x, ())),
    ///     &mut |k, (), cur| cur.map(|(_, c)| (k, c + 1))
    /// );
    /// assert_eq!(
    ///     m.into_iter().map(|(k, v)| (*k, *v)).collect::<Vec<_>>(),
    ///     vec![(0, 1), (1, 2), (2, 3), (3, 4)]
    /// );
    /// ```
    pub fn update_many<Q, D, E, F>(&self, elts: E, f: &mut F) -> Self
    where E: IntoIterator<Item=(Q, D)>, Q: Ord, K: Borrow<Q>,
          F: FnMut(Q, D, Option<(&K, &V)>) -> Option<(K, V)> {
        let (root, len) = self.root.update_many(self.len, elts, f);
        Map { len, root }
    }

    /// return a new map with (k, v) inserted into it. If k
    /// already exists in the old map, the old binding will be
    /// returned, and the new map will contain the new
    /// binding. In fact this method is just a wrapper around
    /// update.
    pub fn insert(&self, k: K, v: V) -> (Self, Option<V>) {
        let (root, len, prev) = self.root.insert(self.len, k, v);
        (Map {len, root}, prev)
    }

    /// return a new map with the binding for q, which can be any
    /// borrowed form of k, updated to the result of f. If f returns
    /// None, the binding will be removed from the new map, otherwise
    /// it will be inserted. This function is more efficient than
    /// calling `get` and then `insert`, since it makes only one tree
    /// traversal instead of two. This method runs in log(N) time and
    /// space where N is the size of the map.
    ///
    /// # Examples
    /// ```
    /// use self::immutable_chunkmap::map::Map;
    ///
    /// let (m, _) = Map::new().update(0, 0, &mut |k, d, _| Some((k, d)));
    /// let (m, _) = m.update(1, 1, &mut |k, d, _| Some((k, d)));
    /// let (m, _) = m.update(2, 2, &mut |k, d, _| Some((k, d)));
    /// assert_eq!(m.get(&0), Some(&0));
    /// assert_eq!(m.get(&1), Some(&1));
    /// assert_eq!(m.get(&2), Some(&2));
    ///
    /// let (m, _) = m.update(0, (), &mut |k, (), v| v.map(move |(_, v)| (k, v + 1)));
    /// assert_eq!(m.get(&0), Some(&1));
    /// assert_eq!(m.get(&1), Some(&1));
    /// assert_eq!(m.get(&2), Some(&2));
    ///
    /// let (m, _) = m.update(1, (), &mut |_, (), _| None);
    /// assert_eq!(m.get(&0), Some(&1));
    /// assert_eq!(m.get(&1), None);
    /// assert_eq!(m.get(&2), Some(&2));
    /// ```
    pub fn update<Q, D, F>(&self, q: Q, d: D, f: &mut F) -> (Self, Option<V>)
    where Q: Ord, K: Borrow<Q>, F: FnMut(Q, D, Option<(&K, &V)>) -> Option<(K, V)> {
        let (root, len, prev) = self.root.update(self.len, q, d, f);
        (Map {len, root}, prev)
    }

    /// lookup the mapping for k. If it doesn't exist return
    /// None. Runs in log(N) time and constant space. where N
    /// is the size of the map.
    pub fn get<'a, Q: ?Sized + Ord>(&'a self, k: &Q) -> Option<&'a V>
    where K: Borrow<Q>
    { self.root.get(k) }

    /// lookup the mapping for k. Return the key. If it doesn't exist
    /// return None. Runs in log(N) time and constant space. where N
    /// is the size of the map.
    pub fn get_key<'a, Q: ?Sized + Ord>(&'a self, k: &Q) -> Option<&'a K>
    where K: Borrow<Q>
    { self.root.get_key(k) }

    /// lookup the mapping for k. Return both the key and the
    /// value. If it doesn't exist return None. Runs in log(N) time
    /// and constant space. where N is the size of the map.
    pub fn get_full<'a, Q: ?Sized + Ord>(&'a self, k: &Q) -> Option<(&'a K, &'a V)>
    where K: Borrow<Q>
    { self.root.get_full(k) }

    /// return a new map with the mapping under k removed. If
    /// the binding existed in the old map return it. Runs in
    /// log(N) time and log(N) space, where N is the size of
    /// the map.
    pub fn remove<Q: Sized + Ord>(&self, k: &Q) -> (Self, Option<V>)
    where K: Borrow<Q>
    {
        let (t, len, prev) = self.root.remove(self.len, k);
        (Map {root: t, len: len}, prev)
    }

    /// get the number of elements in the map O(1) time and space
    pub fn len(&self) -> usize { self.len }

    /// return an iterator over the subset of elements in the
    /// map that are within the specified range.
    ///
    /// The returned iterator runs in O(log(N) + M) time, and
    /// constant space. N is the number of elements in the
    /// tree, and M is the number of elements you examine.
    ///
    /// if lbound >= ubound the returned iterator will be empty
    pub fn range<'a, Q>(
        &'a self, lbound: Bound<Q>, ubound: Bound<Q>
    ) -> Iter<'a, Q, K, V> where Q: Ord, K: Borrow<Q> {
        self.root.range(lbound, ubound)
    }
}

impl<K, V> Map<K, V> where K: Ord + Clone + Debug, V: Clone + Debug {
    #[allow(dead_code)]
    pub(crate) fn invariant(&self) -> () { self.root.invariant(self.len) }
}