Strange anomalies in code

[MWstudios]

Greetings. I've been porting Box2D to C#. During the porting, I've discovered a few strange anomalies in the C++ original and I'd like to know if these were intended or typos:

1. In contact.cs, line 710: The function doesn't check if the function is primary and swap the shapes, so if shapeA is a circle and shapeB is a polygon, the function does not swap them around and fails.

b2Manifold b2ComputeManifold( b2Shape* shapeA, b2Transform transformA, b2Shape* shapeB, b2Transform transformB )
{
	b2ManifoldFcn* fcn = s_registers[shapeA->type][shapeB->type].fcn;
	b2SimplexCache cache = { 0 };
	return fcn( shapeA, transformA, shapeB, transformB, &cache );
}

2. In contact_solver.cs, line 1852 and 1904: The first and second penetration point compute impulses differently. I don't know whether that's intentional.

b2FloatW negImpulse = b2AddW( b2MulW( c->normalMass1, b2AddW( b2MulW( pointMassScale, vn ), bias ) ),
							  b2MulW( pointImpulseScale, c->normalImpulse1 ) );

b2FloatW negImpulse = b2AddW( b2MulW( c->normalMass2, b2MulW( pointMassScale, b2AddW( vn, bias ) ) ),
							  b2MulW( pointImpulseScale, c->normalImpulse2 ) );

3. In solver.cs, line 383: Both offsets are calculated from p1. Again, I don't know if that's intentional.

b2Vec2 p1 = b2TransformPoint( transform, shape->segment.point1 );
b2Vec2 p2 = b2TransformPoint( transform, shape->segment.point2 );
b2Vec2 e = b2Sub( p2, p1 );
b2Vec2 c1 = continuousContext->centroid1;
b2Vec2 c2 = continuousContext->centroid2;
float offset1 = b2Cross( b2Sub( c1, p1 ), e );
float offset2 = b2Cross( b2Sub( c2, p1 ), e );

Also, I'd like to know if InterlockedFetchAddInt() returns the value before or after addition. I'm under the impression that it's the value before addition, which means it behaves differently than C#. Thanks in advance!

[HughPH]

I thought about porting to C# to update Box2D.NetStandard but Box2D 3.x builds to a shared lib, so I just created bindings instead.

https://learn.microsoft.com/en-us/cpp/intrinsics/interlockedexchangeadd-intrinsic-functions?view=msvc-170

"The return value is the initial value of the variable pointed to by the Addend parameter."

So yes, different behaviour to Interlocked.Add

You can P/Invoke the C++ libs if you want identical behaviour.

[MWstudios]

The C# port is actually done. I tried posting a link to the repository in the discussions but I'm unable to submit a discussion for some reason.

As for the interlocked function, I worked around it by, well, simply subtracting again from the returned value. Still unsure about the code above though.

[erincatto]

1. b2ComputeManifold is not used -> removed.
2. This is indeed a bug, but has only a minor effect so I never noticed
3. This is math is correct
   Thanks for reporting!