author  Jim Blandy <jimb@mozilla.com> 
Fri, 16 Oct 2015 12:21:39 0700  
changeset 268117  8e864a608d2f912b1cc208a2f07ed113f94037e5 
parent 268116  c992bef6f751f12b861bce09ea352983f202eb9a 
child 268118  01dd15e9d65895469dd87a4b7c4699a34384e20e 
push id  29541 
push user  philringnalda@gmail.com 
push date  Sat, 17 Oct 2015 17:18:38 +0000 
treeherder  mozillacentral@cfd6d490994a [default view] [failures only] 
perfherder  [talos] [build metrics] [platform microbench] (compared to previous push) 
reviewers  me 
milestone  44.0a1 
first release with  nightly linux32
nightly linux64
nightly mac
nightly win32
nightly win64

last release without  nightly linux32
nightly linux64
nightly mac
nightly win32
nightly win64

 a/mfbt/FastBernoulliTrial.h +++ b/mfbt/FastBernoulliTrial.h @@ 235,17 +235,17 @@ class FastBernoulliTrial { * at would still be fine if our numbers were mathematically perfect. So, * while we've considered IEEE's edge cases, we haven't done anything that * should be actively bad when using other representations. * * (In the below, read comparisons as exact mathematical comparisons: when * we say something "equals 1", that means it's exactly equal to 1. We * treat approximation using intervals with open boundaries: saying a * value is in (0,1) doesn't specify how close to 0 or 1 the value gets.  * When we use closed boundaries like [1, 2**53], we're careful to ensure + * When we use closed boundaries like [2**53, 1], we're careful to ensure * the boundary values are actually representable.) * *  After the comparison above, we know mProbability is in (0,1). * *  The gaps below 1 are 2**53, so that interval is (0, 12**53]. * *  Because the floatingpoint gaps near 1 are wider than those near * zero, there are many small positive doubles ε such that 1ε rounds to