Matrix translation direction changed or origin moved? Or matrix modification in place does something that I don't understand? v2.88.9 to v3.116.0 change

[KRA2008]

(I'm doing this with MAUI on Android but it happens on iOS too)

I was on v2.88.9 for forever but recently upgraded to the latest, but I found a change in behavior I didn't expect and I've found a workaround but I don't understand what happened and I'm trying to figure out either where this is documented or what happened or what I'm not understanding. The change appears when you upgrade from v2.88.9 to v3.116.0. All I'm trying to do is draw a square in the middle of the canvas, then rotate it by 45 degrees about its center. In order to do this you have to use three matrix operations: 1: move to the origin 2: do the rotation 3: move back to original location. Here's the very simple code to make it work:

private void SKCanvasView_OnPaintSurface(object? sender, SKPaintSurfaceEventArgs e)
{
    var canvas = e.Surface.Canvas;
    
    var transform4D = SKMatrix44.CreateIdentity();
    var bounds = canvas.DeviceClipBounds;
    
    var center = bounds.Width / 2f;
    var oneThird = bounds.Width / 3f;
    
    //v3.116.0
    transform4D = transform4D.PostConcat(SKMatrix44.CreateTranslation(center, center, 0));
    transform4D = transform4D.PostConcat(SKMatrix44.CreateRotationDegrees(0, 0, 1, 45));
    transform4D = transform4D.PostConcat(SKMatrix44.CreateTranslation(-center, -center, 0));
    
    //v 2.88.9
    //transform4D.PostConcat(SKMatrix44.CreateTranslation(-center, -center, 0));
    //transform4D.PostConcat(SKMatrix44.CreateRotationDegrees(0, 0, 1, 45));
    //transform4D.PostConcat(SKMatrix44.CreateTranslation(center, center, 0));
    
    canvas.SetMatrix(transform4D.Matrix);
    var whitePaint = new SKPaint();
    whitePaint.Color = SKColor.Parse("ffffff");
    
    canvas.DrawRect(oneThird, oneThird, oneThird, oneThird, whitePaint);
}

Why did my needed translations invert? (from -center to center)
At the same time, PostConcat went from being done in-place to needing to be assigned to the variable. Is that relevant?

[junkerderprovinz]

In 3.x SKMatrix44 is no longer a wrapper around the native Skia matrix. It's a struct laid out like System.Numerics.Matrix4x4, which uses row vectors, so a * b applies a first and PostConcat(m) is just m * this (source). Each matrix you PostConcat now runs before what's already in the chain, while the native post_concat in 2.88 ran it after, and that's why your translations swapped. The new return value comes from the struct change and has nothing to do with the order. To keep your original reading order, use PreConcat with the old sequence (-center, rotate, +center), or write it as SKMatrix44.CreateTranslation(-center, -center, 0) * SKMatrix44.CreateRotationDegrees(0, 0, 1, 45) * SKMatrix44.CreateTranslation(center, center, 0).

[KRA2008]

@junkerderprovinz thanks a lot for the explanation. i took linear algebra in college so i believe i understand the importance of the order of the operands in matrix multiplication as contrasted with scalar multiplication, but i'm still missing why this change occurred the way it did. it seems backwards. to me it seems like pre and post should be swapped in their functionality (as they were in v2.x). do you know if there is some documentation or discussion somewhere that explains why it was done this way? do you agree with the way this was done? i'm not picking up an opinion in your comment.

i'm probably just going to stay away from post or pre entirely and just use *, since the syntax got just as verbose as concat and now counterintuitive (at least according to my intuition).

[junkerderprovinz]

I couldn't find anything written about it; the closest is #2778, where the 3.x struct was moved onto Matrix4x4. My guess is that the formula was carried over as it was: PostConcat(m) is m * this in both versions, but in 2.88 that product was read the column vector way (this first, then m), while Matrix4x4 reads it the row vector way (m first), so the meaning flipped even though the code looks the same. I agree it's counterintuitive, especially since it no longer matches what Skia's own postConcat does, so it might be worth opening an issue and asking whether it's intended. Using * directly is what I'd do too.

[KRA2008]

#5357