Repository navigation
Random.NextBytes is very slightly biased against 0xFF #41457
Description
NextBytes(byte[]) is implemented like so:
public virtual void NextBytes(byte[] buffer)
{
if (buffer == null) throw new ArgumentNullException(nameof(buffer));
for (int i = 0; i < buffer.Length; i++)
{
buffer[i] = (byte)InternalSample();
}
}(side note: why no unchecked on the cast...?)
The Span<byte> version has the same effective implementation if Next() has not been overridden.
The problem is that InternalSample() returns values in the range [0, int.MaxValue) - that is, int.MaxValue is not included. This skews the distribution of possible values for bytes:
0x00-0xFE- 8388608 values0xFF- 8388607 values
The catch is that I'm not sure that this is fixable, given that any fix for this would be a breaking change in the behavior, and we've been reluctant to do so for other parts of Random.
Activity
Dotnet-GitSync-Bot commentedon Aug 27, 2020
I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.
GrabYourPitchforks commentedon Aug 27, 2020
Possible dupe of #23198.
Clockwork-Muse commentedon Aug 27, 2020
@GrabYourPitchforks - I mean, that issue is about the fact that there's a deviation in the algorithm used to generate what
InternalSample()returns.This one is more about how there's a slight bias due to the fact that
InternalSample()doesn't returnint.MaxValue. That is, even if there was no deviation in the algorithm, there would still be this issue.Although yes, it may end up being a dupe due to the resolution. I just wanted this explicitly called out.
GrabYourPitchforks commentedon Aug 31, 2020
Ah, yes, you're right in that it's not an exact dupe. But the resolution would be the same. Either they're all resolved as "by design", or they're fixed all at once.
Clockwork-Muse commentedon Jan 21, 2021
Note to future readers:
The fix implemented at this time only affects randoms created without an explicit seed. Due to compatibility concerns, randoms initialized with a seed retain the old behavior.