Change to use local Random object instead of global rand() in the RtcEventLog unit test.

webrtc/test/random.h

 /*
  *  Copyright (c) 2015 The WebRTC project authors. All Rights Reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
 #ifndef WEBRTC_TEST_RANDOM_H_
 #define WEBRTC_TEST_RANDOM_H_
 
+#include <limits>
+
 #include "webrtc/typedefs.h"
 #include "webrtc/base/constructormagic.h"
 
 namespace webrtc {
 
 namespace test {
 
 class Random {
  public:
   explicit Random(uint32_t seed);
 
   // Return pseudo-random number in the interval [0.0, 1.0).
   float Rand();
 
-  // Return pseudo-random number mapped to the interval [low, high].
-  int Rand(int low, int high);
+  template <typename T>
+  T Rand() {

[the sun, 2015/10/22 12:30:38]

This is very pretty!

Why are some functions called Rand() and some Uniform()?

Should the float version also be a template specialization or is that just
adding cumbersome syntax?

[terelius, 2015/10/26 18:18:41]

I considered both ideas, but couldn't make up my mind. :)

My (weak) argument for doing it this way was that the functions which takes a
distribution parameter should be named after that distribution, e.g Uniform. I
kept Rand in analogy with the C rand() function when you want a random element
but haven't explicitly specified a distribution.

I don't have any strong opinions. What would you prefer?

[the sun, 2015/10/27 09:03:12]

Ok, I understand your argument. I think we should stick to Rand() though (leave
Gaussian and Exponential as they are), but rename the Uniform() functions. I
don't have super solid reasons either:

1. The distribution is a property of the implementation and doesn't differ
between the Rand() and Uniform() functions.
2. The different naming will likely cause more confusion than it is helping. I
imagine users commonly will look for a function to return a PRN within a range,
of a certain type and so, rather than a certain distribution.

Also, leave the float impl as it is now.

[terelius, 2015/10/29 12:54:04]

But isn't it a bit weird (and error prone) to overload Rand such that it
computes something completely differently when called with a parameter compared
to no parameter?

For example:
int i = prng.Rand(99) // Works. Sets i to a random number in [0,99].
int i = prng.Rand()   // Compiles and runs without errors, but i=0.

+    static_assert(std::numeric_limits<T>::is_integer &&
+                      std::numeric_limits<T>::radix == 2 &&
+                      std::numeric_limits<T>::digits <= 32,
+                  "Rand is only supported for built-in integer types that are "
+                  "32 bits or smaller.");
+    return static_cast<T>(Uniform(std::numeric_limits<uint32_t>::max()));
+  }
+
+  template <bool>
+  bool Rand() {
+    return Uniform(0, 1) == 1;
+  }
+
+  // Uniformly distributed pseudo-random number in the interval [0, t].
+  uint32_t Uniform(uint32_t t);
+
+  // Uniformly distributed pseudo-random number in the interval [low, high].
+  uint32_t Uniform(uint32_t low, uint32_t high);
 
   // Normal Distribution.
   int Gaussian(int mean, int standard_deviation);
 
   // Exponential Distribution.
   int Exponential(float lambda);
 
   // TODO(solenberg): Random from histogram.
   // template<typename T> int Distribution(const std::vector<T> histogram) {
 
  private:
   uint32_t a_;
   uint32_t b_;
 
   RTC_DISALLOW_IMPLICIT_CONSTRUCTORS(Random);
 };
 }  // namespace test
 }  // namespace webrtc
 
 #endif  // WEBRTC_TEST_RANDOM_H_