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
346
347
348
349
350
351
352
353
354
355
356
357
358
#![deny(missing_docs)]

//! A crate for things that are
//! 1) Lazily initialized
//! 2) Expensive to create
//! 3) Immutable after creation
//! 4) Used on multiple threads
//!
//! `Lazy<T>` is better than `Mutex<Option<T>>` because after creation accessing
//! `T` does not require any locking, just a single boolean load with
//! `Ordering::Acquire` (which on x86 is just a compiler barrier, not an actual
//! memory barrier).

use std::cell::UnsafeCell;
use std::fmt;
use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::Mutex;

#[derive(Clone)]
enum ThisOrThat<T, U> {
    This(T),
    That(U),
}

/// `LazyTransform<T, U>` is a synchronized holder type, that holds a value of
/// type T until it is lazily converted into a value of type U.
pub struct LazyTransform<T, U> {
    initialized: AtomicBool,
    lock: Mutex<()>,
    value: UnsafeCell<Option<ThisOrThat<T, U>>>,
}

// Implementation details.
impl<T, U> LazyTransform<T, U> {
    fn extract(&self) -> Option<&U> {
        // Make sure we're initialized first!
        match unsafe { (*self.value.get()).as_ref() } {
            None => None,
            Some(&ThisOrThat::This(_)) => panic!(), // Should already be initialized!
            Some(&ThisOrThat::That(ref that)) => Some(that),
        }
    }
}

// Public API.
impl<T, U> LazyTransform<T, U> {
    /// Construct a new, untransformed `LazyTransform<T, U>` with an argument of
    /// type T.
    pub fn new(t: T) -> LazyTransform<T, U> {
        LazyTransform {
            initialized: AtomicBool::new(false),
            lock: Mutex::new(()),
            value: UnsafeCell::new(Some(ThisOrThat::This(t))),
        }
    }

    /// Unwrap the contained value, returning `Ok(U)` if the `LazyTransform<T, U>` has been
    /// transformed or `Err(T)` if it has not.
    pub fn into_inner(self) -> Result<U, T> {
        // We don't need to inspect `self.initialized` since `self` is owned
        // so it is guaranteed that no other threads are accessing its data.
        match self.value.into_inner().unwrap() {
            ThisOrThat::This(t) => Err(t),
            ThisOrThat::That(u) => Ok(u),
        }
    }
}

// Public API.
impl<T, U> LazyTransform<T, U> {
    /// Get a reference to the transformed value, invoking `f` to transform it
    /// if the `LazyTransform<T, U>` has yet to be transformed.  It is
    /// guaranteed that if multiple calls to `get_or_create` race, only one
    /// will invoke its closure, and every call will receive a reference to the
    /// newly transformed value.
    ///
    /// The closure can only ever be called once, so think carefully about what
    /// transformation you want to apply!
    pub fn get_or_create<F>(&self, f: F) -> &U
    where
        F: FnOnce(T) -> U,
    {
        // In addition to being correct, this pattern is vouched for by Hans Boehm
        // (http://schd.ws/hosted_files/cppcon2016/74/HansWeakAtomics.pdf Page 27)
        if !self.initialized.load(Ordering::Acquire) {
            // We *may* not be initialized. We have to block to be certain.
            let _lock = self.lock.lock().unwrap();
            if !self.initialized.load(Ordering::Relaxed) {
                // Ok, we're definitely uninitialized.
                // Safe to fiddle with the UnsafeCell now, because we're locked,
                // and there can't be any outstanding references.
                let value = unsafe { &mut *self.value.get() };
                let this = match value.take().unwrap() {
                    ThisOrThat::This(t) => t,
                    ThisOrThat::That(_) => panic!(), // Can't already be initialized!
                };
                *value = Some(ThisOrThat::That(f(this)));
                self.initialized.store(true, Ordering::Release);
            } else {
                // We raced, and someone else initialized us. We can fall
                // through now.
            }
        }

        // We're initialized, our value is immutable, no synchronization needed.
        self.extract().unwrap()
    }

    /// Get a reference to the transformed value, returning `Some(&U)` if the
    /// `LazyTransform<T, U>` has been transformed or `None` if it has not.  It
    /// is guaranteed that if a reference is returned it is to the transformed
    /// value inside the the `LazyTransform<T>`.
    pub fn get(&self) -> Option<&U> {
        if self.initialized.load(Ordering::Acquire) {
            // We're initialized, our value is immutable, no synchronization needed.
            self.extract()
        } else {
            None
        }
    }
}

// As `T` is only ever accessed when locked, it's enough if it's `Send` for `Self` to be `Sync`.
unsafe impl<T, U> Sync for LazyTransform<T, U>
where
    T: Send,
    U: Send + Sync,
{
}

impl<T, U> Clone for LazyTransform<T, U>
where
    T: Clone,
    U: Clone,
{
    fn clone(&self) -> Self {
        // Overall, this method is very similar to `get_or_create` and uses the same
        // soundness reasoning.

        if self.initialized.load(Ordering::Acquire) {
            Self {
                initialized: true.into(),
                lock: Mutex::default(),
                value: UnsafeCell::new(unsafe {
                    // SAFETY:
                    // Everything is initialized and immutable here, so lockless cloning is safe.
                    (&*self.value.get()).clone()
                }),
            }
        } else {
            // We *may* not be initialized. We have to block here before accessing `value`,
            // which also synchronises the `initialized` load.
            let _lock = self.lock.lock().unwrap();
            Self {
                initialized: self.initialized.load(Ordering::Relaxed).into(),
                lock: Mutex::default(),
                value: UnsafeCell::new(unsafe {
                    // SAFETY:
                    // Exclusive access while `_lock` is held.
                    (&*self.value.get()).clone()
                }),
            }
        }
    }

    fn clone_from(&mut self, source: &Self) {
        // Overall, this method is very similar to `get_or_create` and uses the same
        // soundness reasoning. It's implemented explicitly here to avoid a `Mutex` drop/new.

        if self.initialized.load(Ordering::Acquire) {
            unsafe {
                // SAFETY:
                // Everything is initialized and immutable here, so lockless cloning is safe.
                // It's still important to store `initialized` with correct ordering, though.
                *self.value.get() = (&*source.value.get()).clone();
                self.initialized.store(true, Ordering::Release);
            }
        } else {
            // `source` *may* not be initialized. We have to block here before accessing `value`,
            // which also synchronises the `initialized` load (and incidentally also the `initialized`
            // store due to the exclusive reference to `self`, so that can be `Relaxed` here too).
            let _lock = source.lock.lock().unwrap();
            unsafe {
                // SAFETY:
                // Exclusive access to `source` while `_lock` is held.
                *self.value.get() = (&*source.value.get()).clone();
                self.initialized.store(
                    source.initialized.load(Ordering::Relaxed),
                    Ordering::Relaxed,
                );
            }
        }
    }
}

impl<T, U> Default for LazyTransform<T, U>
where
    T: Default,
{
    fn default() -> Self {
        LazyTransform::new(T::default())
    }
}

/// `Lazy<T>` is a lazily initialized synchronized holder type.  You can think
/// of it as a `LazyTransform` where the initial type doesn't exist.
#[derive(Clone)]
pub struct Lazy<T> {
    inner: LazyTransform<(), T>,
}

impl<T> Lazy<T> {
    /// Construct a new, uninitialized `Lazy<T>`.
    pub fn new() -> Lazy<T> {
        Self::default()
    }

    /// Unwrap the contained value, returning `Some` if the `Lazy<T>` has been initialized
    /// or `None` if it has not.
    pub fn into_inner(self) -> Option<T> {
        self.inner.into_inner().ok()
    }
}

impl<T> Lazy<T> {
    /// Get a reference to the contained value, invoking `f` to create it
    /// if the `Lazy<T>` is uninitialized.  It is guaranteed that if multiple
    /// calls to `get_or_create` race, only one will invoke its closure, and
    /// every call will receive a reference to the newly created value.
    ///
    /// The value stored in the `Lazy<T>` is immutable after the closure returns
    /// it, so think carefully about what you want to put inside!
    pub fn get_or_create<F>(&self, f: F) -> &T
    where
        F: FnOnce() -> T,
    {
        self.inner.get_or_create(|_| f())
    }

    /// Get a reference to the contained value, returning `Some(ref)` if the
    /// `Lazy<T>` has been initialized or `None` if it has not.  It is
    /// guaranteed that if a reference is returned it is to the value inside
    /// the `Lazy<T>`.
    pub fn get(&self) -> Option<&T> {
        self.inner.get()
    }
}

// `#[derive(Default)]` automatically adds `T: Default` trait bound, but that
// is too restrictive, because `Lazy<T>` always has a default value for any `T`.
impl<T> Default for Lazy<T> {
    fn default() -> Self {
        Lazy {
            inner: LazyTransform::new(()),
        }
    }
}

impl<T> fmt::Debug for Lazy<T>
where
    T: fmt::Debug,
{
    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
        if let Some(v) = self.get() {
            f.write_fmt(format_args!("Lazy({:?})", v))
        } else {
            f.write_str("Lazy(<uninitialized>)")
        }
    }
}

#[cfg(test)]
extern crate rayon;

#[cfg(test)]
mod tests {

    use super::{Lazy, LazyTransform};
    use rayon::ThreadPoolBuilder;
    use std::sync::atomic::{AtomicUsize, Ordering};
    use std::{thread, time};

    #[test]
    fn test_lazy() {
        let lazy_value: Lazy<u8> = Lazy::new();

        assert_eq!(lazy_value.get(), None);

        let n = AtomicUsize::new(0);

        let pool = ThreadPoolBuilder::new().num_threads(100).build().unwrap();
        pool.scope(|scope| {
            for _ in 0..100 {
                let lazy_ref = &lazy_value;
                let n_ref = &n;
                scope.spawn(move |_| {
                    let ten_millis = time::Duration::from_millis(10);
                    thread::sleep(ten_millis);

                    let value = *lazy_ref.get_or_create(|| {
                        // Make everybody else wait on me, because I'm a jerk.
                        thread::sleep(ten_millis);

                        // Make this relaxed so it doesn't interfere with
                        // Lazy internals at all.
                        n_ref.fetch_add(1, Ordering::Relaxed);

                        42
                    });
                    assert_eq!(value, 42);

                    let value = lazy_ref.get();
                    assert_eq!(value, Some(&42));
                });
            }
        });

        assert_eq!(n.load(Ordering::SeqCst), 1);
    }

    #[test]
    fn test_lazy_transform() {
        let lazy_value: LazyTransform<u8, u8> = LazyTransform::new(21);

        assert_eq!(lazy_value.get(), None);

        let n = AtomicUsize::new(0);

        let pool = ThreadPoolBuilder::new().num_threads(100).build().unwrap();
        pool.scope(|scope| {
            for _ in 0..100 {
                let lazy_ref = &lazy_value;
                let n_ref = &n;
                scope.spawn(move |_| {
                    let ten_millis = time::Duration::from_millis(10);
                    thread::sleep(ten_millis);

                    let value = *lazy_ref.get_or_create(|v| {
                        // Make everybody else wait on me, because I'm a jerk.
                        thread::sleep(ten_millis);

                        // Make this relaxed so it doesn't interfere with
                        // Lazy internals at all.
                        n_ref.fetch_add(1, Ordering::Relaxed);

                        v * 2
                    });
                    assert_eq!(value, 42);

                    let value = lazy_ref.get();
                    assert_eq!(value, Some(&42));
                });
            }
        });

        assert_eq!(n.load(Ordering::SeqCst), 1);
    }
}