Fix off-by-one error in PRNG.

webrtc/test/random.cc

Index: webrtc/test/random.cc
diff --git a/webrtc/test/random.cc b/webrtc/test/random.cc
index 8877ed4c20c5bdfca2fcef03806f8b84d64b92bb..6bc707976c7e65573832f840411b7f80bb4036a2 100644
--- a/webrtc/test/random.cc
+++ b/webrtc/test/random.cc
@@ -22,16 +22,16 @@ Random::Random(uint32_t seed) : a_(0x531FDB97 ^ seed), b_(0x6420ECA8 + seed) {
 }
 
 float Random::Rand() {
-  const float kScale = 1.0f / 0xffffffff;
-  float result = kScale * b_;
+  const double kScale = 1.0f / 0xfffffffful;
+  double result = kScale * b_;
   a_ ^= b_;
   b_ += a_;
-  return result;
+  return static_cast<float>(result);
 }
 
 int Random::Rand(int low, int high) {
   RTC_DCHECK(low <= high);
-  float uniform = Rand() * (high - low + 1) + low;
+  float uniform = Rand() * (high - low) + low;

[pbos-webrtc, 2015/10/19 15:22:21]

Wait, this actually causes a bad distribution, if we do this (with 1.0
inclusive) then 2, 3 gives that 3 is infinitely (ok, 1 out of 2**32) unlikely.

Revert this line and divide kScale by 2**32 instead so that 1.0 is inclusive,
then the previous line was OK.

[terelius, 2015/10/19 15:27:23]

I was actually about to file a bug about this independently. As Peter points
out, this gives the wrong distribution, but with the +1, the computation
overflows.

I am quite convinced that the only way to solve this completely is to use
unsigned.

   return static_cast<int>(uniform);
 }