Right-shift of certain negative BigIntegers incorrectly results in 0

[slozier]

Description

It appears that right shifting certain negative BigInteger values by 32 incorrectly results in 0.

var a = BigInteger.Parse("-18446744073709543424");
Debug.Assert(a >> 32 == (a >> 31) >> 1);

I've seen this occur for -18446744073709543424 and -79228162514264337593543917568. Appears to be related to the underlying array representation since it's possible to construct the same values in a way that the shift works.

Configuration

Runtime Environment:
OS Name: Windows
OS Version: 10.0.19042
OS Platform: Windows
RID: win10-x64
Base Path: C:\Program Files\dotnet\sdk\5.0.100-rc.2.20479.15\

Host (useful for support):
Version: 5.0.0-rc.2.20475.5
Commit: c5a3f49

Regression?

Appears to occur on .NET Framework as well so probably a bug that's been there for a long time?

[vcsjones]

/cc @tannergooding (no idea where bot for labels is)

[ghost]

Tagging subscribers to this area: @tannergooding, @pgovind, @jeffhandley
See info in area-owners.md if you want to be subscribed.

[tannergooding]

Will look into this more, I'm guessing there is a bug in the sign propagation logic.: https://source.dot.net/#System.Runtime.Numerics/System/Numerics/BigInteger.cs,1926

[wzchua]

This is a dupe of #27358