src/librand/lib.rs

   1 // Copyright 2013-2014 The Rust Project Developers. See the COPYRIGHT
   2 // file at the top-level directory of this distribution and at
   3 // http://rust-lang.org/COPYRIGHT.
   4 //
   5 // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
   6 // http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
   7 // <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
   8 // option. This file may not be copied, modified, or distributed
   9 // except according to those terms.
  10
  11 //! Interface to random number generators in Rust.
  12 //!
  13 //! This is an experimental library which lives underneath the standard library
  14 //! in its dependency chain. This library is intended to define the interface
  15 //! for random number generation and also provide utilities around doing so. It
  16 //! is not recommended to use this library directly, but rather the official
  17 //! interface through `std::rand`.
  18
  19 // Do not remove on snapshot creation. Needed for bootstrap. (Issue #22364)
  20 #![cfg_attr(stage0, feature(custom_attribute))]
  21 #![crate_name = "rand"]
  22 #![crate_type = "rlib"]
  23 #![doc(html_logo_url = "http://www.rust-lang.org/logos/rust-logo-128x128-blk.png",
  24        html_favicon_url = "https://doc.rust-lang.org/favicon.ico",
  25        html_root_url = "http://doc.rust-lang.org/nightly/",
  26        html_playground_url = "http://play.rust-lang.org/")]
  27 #![no_std]
  28 #![staged_api]
  29 #![unstable(feature = "rand",
  30             reason = "use `rand` from crates.io")]
  31 #![feature(core)]
  32 #![feature(core_float)]
  33 #![feature(core_prelude)]
  34 #![feature(core_slice_ext)]
  35 #![feature(no_std)]
  36 #![feature(num_bits_bytes)]
  37 #![feature(staged_api)]
  38 #![feature(step_by)]
  39
  40 #![cfg_attr(test, feature(test, rand, rustc_private, iter_order))]
  41
  42 #![allow(deprecated)]
  43
  44 #[macro_use]
  45 extern crate core;
  46
  47 #[cfg(test)] #[macro_use] extern crate std;
  48 #[cfg(test)] #[macro_use] extern crate log;
  49
  50 use core::prelude::*;
  51 use core::marker::PhantomData;
  52
  53 pub use isaac::{IsaacRng, Isaac64Rng};
  54 pub use chacha::ChaChaRng;
  55
  56 use distributions::{Range, IndependentSample};
  57 use distributions::range::SampleRange;
  58
  59 #[cfg(test)]
  60 const RAND_BENCH_N: u64 = 100;
  61
  62 pub mod distributions;
  63 pub mod isaac;
  64 pub mod chacha;
  65 pub mod reseeding;
  66 mod rand_impls;
  67
  68 /// A type that can be randomly generated using an `Rng`.
  69 pub trait Rand : Sized {
  70     /// Generates a random instance of this type using the specified source of
  71     /// randomness.
  72     fn rand<R: Rng>(rng: &mut R) -> Self;
  73 }
  74
  75 /// A random number generator.
  76 pub trait Rng : Sized {
  77     /// Return the next random u32.
  78     ///
  79     /// This rarely needs to be called directly, prefer `r.gen()` to
  80     /// `r.next_u32()`.
  81     // FIXME #7771: Should be implemented in terms of next_u64
  82     fn next_u32(&mut self) -> u32;
  83
  84     /// Return the next random u64.
  85     ///
  86     /// By default this is implemented in terms of `next_u32`. An
  87     /// implementation of this trait must provide at least one of
  88     /// these two methods. Similarly to `next_u32`, this rarely needs
  89     /// to be called directly, prefer `r.gen()` to `r.next_u64()`.
  90     fn next_u64(&mut self) -> u64 {
  91         ((self.next_u32() as u64) << 32) | (self.next_u32() as u64)
  92     }
  93
  94     /// Return the next random f32 selected from the half-open
  95     /// interval `[0, 1)`.
  96     ///
  97     /// By default this is implemented in terms of `next_u32`, but a
  98     /// random number generator which can generate numbers satisfying
  99     /// the requirements directly can overload this for performance.
 100     /// It is required that the return value lies in `[0, 1)`.
 101     ///
 102     /// See `Closed01` for the closed interval `[0,1]`, and
 103     /// `Open01` for the open interval `(0,1)`.
 104     fn next_f32(&mut self) -> f32 {
 105         const MANTISSA_BITS: usize = 24;
 106         const IGNORED_BITS: usize = 8;
 107         const SCALE: f32 = (1u64 << MANTISSA_BITS) as f32;
 108
 109         // using any more than `MANTISSA_BITS` bits will
 110         // cause (e.g.) 0xffff_ffff to correspond to 1
 111         // exactly, so we need to drop some (8 for f32, 11
 112         // for f64) to guarantee the open end.
 113         (self.next_u32() >> IGNORED_BITS) as f32 / SCALE
 114     }
 115
 116     /// Return the next random f64 selected from the half-open
 117     /// interval `[0, 1)`.
 118     ///
 119     /// By default this is implemented in terms of `next_u64`, but a
 120     /// random number generator which can generate numbers satisfying
 121     /// the requirements directly can overload this for performance.
 122     /// It is required that the return value lies in `[0, 1)`.
 123     ///
 124     /// See `Closed01` for the closed interval `[0,1]`, and
 125     /// `Open01` for the open interval `(0,1)`.
 126     fn next_f64(&mut self) -> f64 {
 127         const MANTISSA_BITS: usize = 53;
 128         const IGNORED_BITS: usize = 11;
 129         const SCALE: f64 = (1u64 << MANTISSA_BITS) as f64;
 130
 131         (self.next_u64() >> IGNORED_BITS) as f64 / SCALE
 132     }
 133
 134     /// Fill `dest` with random data.
 135     ///
 136     /// This has a default implementation in terms of `next_u64` and
 137     /// `next_u32`, but should be overridden by implementations that
 138     /// offer a more efficient solution than just calling those
 139     /// methods repeatedly.
 140     ///
 141     /// This method does *not* have a requirement to bear any fixed
 142     /// relationship to the other methods, for example, it does *not*
 143     /// have to result in the same output as progressively filling
 144     /// `dest` with `self.gen::<u8>()`, and any such behaviour should
 145     /// not be relied upon.
 146     ///
 147     /// This method should guarantee that `dest` is entirely filled
 148     /// with new data, and may panic if this is impossible
 149     /// (e.g. reading past the end of a file that is being used as the
 150     /// source of randomness).
 151     fn fill_bytes(&mut self, dest: &mut [u8]) {
 152         // this could, in theory, be done by transmuting dest to a
 153         // [u64], but this is (1) likely to be undefined behaviour for
 154         // LLVM, (2) has to be very careful about alignment concerns,
 155         // (3) adds more `unsafe` that needs to be checked, (4)
 156         // probably doesn't give much performance gain if
 157         // optimisations are on.
 158         let mut count = 0;
 159         let mut num = 0;
 160         for byte in dest {
 161             if count == 0 {
 162                 // we could micro-optimise here by generating a u32 if
 163                 // we only need a few more bytes to fill the vector
 164                 // (i.e. at most 4).
 165                 num = self.next_u64();
 166                 count = 8;
 167             }
 168
 169             *byte = (num & 0xff) as u8;
 170             num >>= 8;
 171             count -= 1;
 172         }
 173     }
 174
 175     /// Return a random value of a `Rand` type.
 176     #[inline(always)]
 177     fn gen<T: Rand>(&mut self) -> T {
 178         Rand::rand(self)
 179     }
 180
 181     /// Return an iterator that will yield an infinite number of randomly
 182     /// generated items.
 183     fn gen_iter<'a, T: Rand>(&'a mut self) -> Generator<'a, T, Self> {
 184         Generator { rng: self, _marker: PhantomData }
 185     }
 186
 187     /// Generate a random value in the range [`low`, `high`).
 188     ///
 189     /// This is a convenience wrapper around
 190     /// `distributions::Range`. If this function will be called
 191     /// repeatedly with the same arguments, one should use `Range`, as
 192     /// that will amortize the computations that allow for perfect
 193     /// uniformity, as they only happen on initialization.
 194     ///
 195     /// # Panics
 196     ///
 197     /// Panics if `low >= high`.
 198     fn gen_range<T: PartialOrd + SampleRange>(&mut self, low: T, high: T) -> T {
 199         assert!(low < high, "Rng.gen_range called with low >= high");
 200         Range::new(low, high).ind_sample(self)
 201     }
 202
 203     /// Return a bool with a 1 in n chance of true
 204     fn gen_weighted_bool(&mut self, n: usize) -> bool {
 205         n <= 1 || self.gen_range(0, n) == 0
 206     }
 207
 208     /// Return an iterator of random characters from the set A-Z,a-z,0-9.
 209     fn gen_ascii_chars<'a>(&'a mut self) -> AsciiGenerator<'a, Self> {
 210         AsciiGenerator { rng: self }
 211     }
 212
 213     /// Return a random element from `values`.
 214     ///
 215     /// Return `None` if `values` is empty.
 216     fn choose<'a, T>(&mut self, values: &'a [T]) -> Option<&'a T> {
 217         if values.is_empty() {
 218             None
 219         } else {
 220             Some(&values[self.gen_range(0, values.len())])
 221         }
 222     }
 223
 224     /// Shuffle a mutable slice in place.
 225     fn shuffle<T>(&mut self, values: &mut [T]) {
 226         let mut i = values.len();
 227         while i >= 2 {
 228             // invariant: elements with index >= i have been locked in place.
 229             i -= 1;
 230             // lock element i in place.
 231             values.swap(i, self.gen_range(0, i + 1));
 232         }
 233     }
 234 }
 235
 236 /// Iterator which will generate a stream of random items.
 237 ///
 238 /// This iterator is created via the `gen_iter` method on `Rng`.
 239 pub struct Generator<'a, T, R:'a> {
 240     rng: &'a mut R,
 241     _marker: PhantomData<T>
 242 }
 243
 244 impl<'a, T: Rand, R: Rng> Iterator for Generator<'a, T, R> {
 245     type Item = T;
 246
 247     fn next(&mut self) -> Option<T> {
 248         Some(self.rng.gen())
 249     }
 250 }
 251
 252 /// Iterator which will continuously generate random ascii characters.
 253 ///
 254 /// This iterator is created via the `gen_ascii_chars` method on `Rng`.
 255 pub struct AsciiGenerator<'a, R:'a> {
 256     rng: &'a mut R,
 257 }
 258
 259 impl<'a, R: Rng> Iterator for AsciiGenerator<'a, R> {
 260     type Item = char;
 261
 262     fn next(&mut self) -> Option<char> {
 263         const GEN_ASCII_STR_CHARSET: &'static [u8] =
 264             b"ABCDEFGHIJKLMNOPQRSTUVWXYZ\
 265               abcdefghijklmnopqrstuvwxyz\
 266               0123456789";
 267         Some(*self.rng.choose(GEN_ASCII_STR_CHARSET).unwrap() as char)
 268     }
 269 }
 270
 271 /// A random number generator that can be explicitly seeded to produce
 272 /// the same stream of randomness multiple times.
 273 pub trait SeedableRng<Seed>: Rng {
 274     /// Reseed an RNG with the given seed.
 275     fn reseed(&mut self, Seed);
 276
 277     /// Create a new RNG with the given seed.
 278     fn from_seed(seed: Seed) -> Self;
 279 }
 280
 281 /// An Xorshift[1] random number
 282 /// generator.
 283 ///
 284 /// The Xorshift algorithm is not suitable for cryptographic purposes
 285 /// but is very fast. If you do not know for sure that it fits your
 286 /// requirements, use a more secure one such as `IsaacRng` or `OsRng`.
 287 ///
 288 /// [1]: Marsaglia, George (July 2003). ["Xorshift
 289 /// RNGs"](http://www.jstatsoft.org/v08/i14/paper). *Journal of
 290 /// Statistical Software*. Vol. 8 (Issue 14).
 291 #[derive(Clone)]
 292 pub struct XorShiftRng {
 293     x: u32,
 294     y: u32,
 295     z: u32,
 296     w: u32,
 297 }
 298
 299 impl XorShiftRng {
 300     /// Creates a new XorShiftRng instance which is not seeded.
 301     ///
 302     /// The initial values of this RNG are constants, so all generators created
 303     /// by this function will yield the same stream of random numbers. It is
 304     /// highly recommended that this is created through `SeedableRng` instead of
 305     /// this function
 306     pub fn new_unseeded() -> XorShiftRng {
 307         XorShiftRng {
 308             x: 0x193a6754,
 309             y: 0xa8a7d469,
 310             z: 0x97830e05,
 311             w: 0x113ba7bb,
 312         }
 313     }
 314 }
 315
 316 impl Rng for XorShiftRng {
 317     #[inline]
 318     fn next_u32(&mut self) -> u32 {
 319         let x = self.x;
 320         let t = x ^ (x << 11);
 321         self.x = self.y;
 322         self.y = self.z;
 323         self.z = self.w;
 324         let w = self.w;
 325         self.w = w ^ (w >> 19) ^ (t ^ (t >> 8));
 326         self.w
 327     }
 328 }
 329
 330 impl SeedableRng<[u32; 4]> for XorShiftRng {
 331     /// Reseed an XorShiftRng. This will panic if `seed` is entirely 0.
 332     fn reseed(&mut self, seed: [u32; 4]) {
 333         assert!(!seed.iter().all(|&x| x == 0),
 334                 "XorShiftRng.reseed called with an all zero seed.");
 335
 336         self.x = seed[0];
 337         self.y = seed[1];
 338         self.z = seed[2];
 339         self.w = seed[3];
 340     }
 341
 342     /// Create a new XorShiftRng. This will panic if `seed` is entirely 0.
 343     fn from_seed(seed: [u32; 4]) -> XorShiftRng {
 344         assert!(!seed.iter().all(|&x| x == 0),
 345                 "XorShiftRng::from_seed called with an all zero seed.");
 346
 347         XorShiftRng {
 348             x: seed[0],
 349             y: seed[1],
 350             z: seed[2],
 351             w: seed[3]
 352         }
 353     }
 354 }
 355
 356 impl Rand for XorShiftRng {
 357     fn rand<R: Rng>(rng: &mut R) -> XorShiftRng {
 358         let mut tuple: (u32, u32, u32, u32) = rng.gen();
 359         while tuple == (0, 0, 0, 0) {
 360             tuple = rng.gen();
 361         }
 362         let (x, y, z, w) = tuple;
 363         XorShiftRng { x: x, y: y, z: z, w: w }
 364     }
 365 }
 366
 367 /// A wrapper for generating floating point numbers uniformly in the
 368 /// open interval `(0,1)` (not including either endpoint).
 369 ///
 370 /// Use `Closed01` for the closed interval `[0,1]`, and the default
 371 /// `Rand` implementation for `f32` and `f64` for the half-open
 372 /// `[0,1)`.
 373 pub struct Open01<F>(pub F);
 374
 375 /// A wrapper for generating floating point numbers uniformly in the
 376 /// closed interval `[0,1]` (including both endpoints).
 377 ///
 378 /// Use `Open01` for the closed interval `(0,1)`, and the default
 379 /// `Rand` implementation of `f32` and `f64` for the half-open
 380 /// `[0,1)`.
 381 pub struct Closed01<F>(pub F);
 382
 383 #[cfg(test)]
 384 mod test {
 385     use std::__rand as rand;
 386
 387     pub struct MyRng<R> { inner: R }
 388
 389     impl<R: rand::Rng> ::Rng for MyRng<R> {
 390         fn next_u32(&mut self) -> u32 {
 391             rand::Rng::next_u32(&mut self.inner)
 392         }
 393     }
 394
 395     pub fn rng() -> MyRng<rand::ThreadRng> {
 396         MyRng { inner: rand::thread_rng() }
 397     }
 398
 399     pub fn weak_rng() -> MyRng<rand::ThreadRng> {
 400         MyRng { inner: rand::thread_rng() }
 401     }
 402 }