about summary refs log tree commit diff
path: root/src/libcore/send_map.rs
blob: 24af1d0aedfafa24a1a1c0312d925833809dd343 (plain)
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
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
/*!

Sendable hash maps.  Very much a work in progress.

*/


/**
 * A function that returns a hash of a value
 *
 * The hash should concentrate entropy in the lower bits.
 */
type hashfn<K> = pure fn~(K) -> uint;
type eqfn<K> = pure fn~(K, K) -> bool;

/// Open addressing with linear probing.
mod linear {
    export linear_map, linear_map_with_capacity, public_methods;

    const initial_capacity: uint = 32u; // 2^5
    type bucket<K,V> = {hash: uint, key: K, value: V};
    enum linear_map<K,V> {
        linear_map_({
            hashfn: pure fn~(x: &K) -> uint,
            eqfn: pure fn~(x: &K, y: &K) -> bool,
            resize_at: uint,
            size: uint,
            buckets: ~[option<bucket<K,V>>]})
    }

    // FIXME(#2979) -- with #2979 we could rewrite found_entry
    // to have type option<&bucket<K,V>> which would be nifty
    enum search_result {
        found_entry(uint), found_hole(uint), table_full
    }

    fn resize_at(capacity: uint) -> uint {
        ((capacity as float) * 3. / 4.) as uint
    }

    fn linear_map<K,V>(
        +hashfn: pure fn~(x: &K) -> uint,
        +eqfn: pure fn~(x: &K, y: &K) -> bool) -> linear_map<K,V> {

        linear_map_with_capacity(hashfn, eqfn, 32)
    }

    fn linear_map_with_capacity<K,V>(
        +hashfn: pure fn~(x: &K) -> uint,
        +eqfn: pure fn~(x: &K, y: &K) -> bool,
        initial_capacity: uint) -> linear_map<K,V> {

        linear_map_({
            hashfn: hashfn,
            eqfn: eqfn,
            resize_at: resize_at(initial_capacity),
            size: 0,
            buckets: vec::from_fn(initial_capacity, |_i| none)})
    }

    // FIXME(#2979) would allow us to use region type for k
    unsafe fn borrow<K>(&&k: K) -> &K {
        let p: *K = ptr::addr_of(k);
        p as &K
    }

    impl private_methods<K,V> for &const linear_map<K,V> {
        #[inline(always)]
        pure fn to_bucket(h: uint) -> uint {
            // FIXME(#3041) borrow a more sophisticated technique here from
            // Gecko, for example borrowing from Knuth, as Eich so
            // colorfully argues for here:
            // https://bugzilla.mozilla.org/show_bug.cgi?id=743107#c22
            h % self.buckets.len()
        }

        #[inline(always)]
        pure fn next_bucket(idx: uint, len_buckets: uint) -> uint {
            let n = (idx + 1) % len_buckets;
            unsafe{ // argh. log not considered pure.
                debug!{"next_bucket(%?, %?) = %?", idx, len_buckets, n};
            }
            ret n;
        }

        #[inline(always)]
        pure fn bucket_sequence(hash: uint, op: fn(uint) -> bool) -> uint {
            let start_idx = self.to_bucket(hash);
            let len_buckets = self.buckets.len();
            let mut idx = start_idx;
            loop {
                if !op(idx) {
                    ret idx;
                }
                idx = self.next_bucket(idx, len_buckets);
                if idx == start_idx {
                    ret start_idx;
                }
            }
        }

        #[inline(always)]
        pure fn bucket_for_key(
            buckets: &[option<bucket<K,V>>],
            k: &K) -> search_result {

            let hash = self.hashfn(k);
            self.bucket_for_key_with_hash(buckets, hash, k)
        }

        #[inline(always)]
        pure fn bucket_for_key_with_hash(
            buckets: &[option<bucket<K,V>>],
            hash: uint,
            k: &K) -> search_result {

            let _ = for self.bucket_sequence(hash) |i| {
                alt buckets[i] {
                  some(bkt) {
                    if bkt.hash == hash && self.eqfn(k, &bkt.key) {
                        ret found_entry(i);
                    }
                  }
                  none => {
                    ret found_hole(i);
                  }
                }
            };
            ret table_full;
        }
    }

    impl private_methods<K,V> for &mut linear_map<K,V> {
        /// Expands the capacity of the array and re-inserts each
        /// of the existing buckets.
        fn expand() {
            let old_capacity = self.buckets.len();
            let new_capacity = old_capacity * 2;
            self.resize_at = ((new_capacity as float) * 3.0 / 4.0) as uint;

            let mut old_buckets = vec::from_fn(new_capacity, |_i| none);
            self.buckets <-> old_buckets;

            for uint::range(0, old_capacity) |i| {
                let mut bucket = none;
                bucket <-> old_buckets[i];
                if bucket.is_some() {
                    self.insert_bucket(bucket);
                }
            }
        }

        fn insert_bucket(+bucket: option<bucket<K,V>>) {
            let {hash, key, value} <- option::unwrap(bucket);
            let _ = self.insert_internal(hash, key, value);
        }

        /// Inserts the key value pair into the buckets.
        /// Assumes that there will be a bucket.
        /// True if there was no previous entry with that key
        fn insert_internal(hash: uint, +k: K, +v: V) -> bool {
            alt self.bucket_for_key_with_hash(self.buckets, hash,
                                              unsafe{borrow(k)}) {
              table_full => {fail ~"Internal logic error";}
              found_hole(idx) {
                debug!{"insert fresh (%?->%?) at idx %?, hash %?",
                       k, v, idx, hash};
                self.buckets[idx] = some({hash: hash, key: k, value: v});
                self.size += 1;
                ret true;
              }
              found_entry(idx) => {
                debug!{"insert overwrite (%?->%?) at idx %?, hash %?",
                       k, v, idx, hash};
                self.buckets[idx] = some({hash: hash, key: k, value: v});
                ret false;
              }
            }
        }
    }

    impl public_methods<K,V> for &mut linear_map<K,V> {
        fn insert(+k: K, +v: V) -> bool {
            if self.size >= self.resize_at {
                // n.b.: We could also do this after searching, so
                // that we do not resize if this call to insert is
                // simply going to update a key in place.  My sense
                // though is that it's worse to have to search through
                // buckets to find the right spot twice than to just
                // resize in this corner case.
                self.expand();
            }

            let hash = self.hashfn(unsafe{borrow(k)});
            self.insert_internal(hash, k, v)
        }

        fn remove(k: &K) -> bool {
            // Removing from an open-addressed hashtable
            // is, well, painful.  The problem is that
            // the entry may lie on the probe path for other
            // entries, so removing it would make you think that
            // those probe paths are empty.
            //
            // To address this we basically have to keep walking,
            // re-inserting entries we find until we reach an empty
            // bucket.  We know we will eventually reach one because
            // we insert one ourselves at the beginning (the removed
            // entry).
            //
            // I found this explanation elucidating:
            // http://www.maths.lse.ac.uk/Courses/MA407/del-hash.pdf

            let mut idx = alt self.bucket_for_key(self.buckets, k) {
              table_full | found_hole(_) => {
                ret false;
              }
              found_entry(idx) => {
                idx
              }
            };

            let len_buckets = self.buckets.len();
            self.buckets[idx] = none;
            idx = self.next_bucket(idx, len_buckets);
            while self.buckets[idx].is_some() {
                let mut bucket = none;
                bucket <-> self.buckets[idx];
                self.insert_bucket(bucket);
                idx = self.next_bucket(idx, len_buckets);
            }
            ret true;
        }
    }

    impl private_methods<K,V> for &linear_map<K,V> {
        fn search(hash: uint, op: fn(x: &option<bucket<K,V>>) -> bool) {
            let _ = self.bucket_sequence(hash, |i| op(&self.buckets[i]));
        }
    }

    impl public_methods<K,V> for &const linear_map<K,V> {
        fn size() -> uint {
            self.size
        }

        fn contains_key(k: &K) -> bool {
            alt self.bucket_for_key(self.buckets, k) {
              found_entry(_) => {true}
              table_full | found_hole(_) => {false}
            }
        }
    }

    impl public_methods<K,V: copy> for &const linear_map<K,V> {
        fn find(k: &K) -> option<V> {
            alt self.bucket_for_key(self.buckets, k) {
              found_entry(idx) => {
                alt check self.buckets[idx] {
                  some(bkt) => {some(copy bkt.value)}
                }
              }
              table_full | found_hole(_) => {
                none
              }
            }
        }

        fn get(k: &K) -> V {
            let value = self.find(k);
            if value.is_none() {
                fail fmt!{"No entry found for key: %?", k};
            }
            option::unwrap(value)
        }

        fn [](k: &K) -> V {
            self.get(k)
        }
    }

    /*
    FIXME --- #2979 must be fixed to typecheck this
    impl imm_methods<K,V> for &linear_map<K,V> {
        fn find_ptr(k: K) -> option<&V> {
            //XXX this should not type check as written, but it should
            //be *possible* to typecheck it...
            self.with_ptr(k, |v| v)
        }
    }
    */
}

#[test]
mod test {

    import linear::linear_map;

    pure fn uint_hash(x: &uint) -> uint { *x }
    pure fn uint_eq(x: &uint, y: &uint) -> bool { *x == *y }

    fn int_linear_map<V>() -> linear_map<uint,V> {
        ret linear_map(uint_hash, uint_eq);
    }

    #[test]
    fn inserts() {
        let mut m = ~int_linear_map();
        assert m.insert(1, 2);
        assert m.insert(2, 4);
        assert m.get(&1) == 2;
        assert m.get(&2) == 4;
    }

    #[test]
    fn overwrite() {
        let mut m = ~int_linear_map();
        assert m.insert(1, 2);
        assert m.get(&1) == 2;
        assert !m.insert(1, 3);
        assert m.get(&1) == 3;
    }

    #[test]
    fn conflicts() {
        let mut m = ~linear::linear_map_with_capacity(uint_hash, uint_eq, 4);
        assert m.insert(1, 2);
        assert m.insert(5, 3);
        assert m.insert(9, 4);
        assert m.get(&9) == 4;
        assert m.get(&5) == 3;
        assert m.get(&1) == 2;
    }

    #[test]
    fn conflict_remove() {
        let mut m = ~linear::linear_map_with_capacity(uint_hash, uint_eq, 4);
        assert m.insert(1, 2);
        assert m.insert(5, 3);
        assert m.insert(9, 4);
        assert m.remove(&1);
        assert m.get(&9) == 4;
        assert m.get(&5) == 3;
    }
}