Fix Faulty Scaling in GC

[HeikoKlare]

Current Behavior

The GC class provides functionality to apply a transformation (translation, scaling, rotation) to the graphics canvas.
This functionality is broken when applying rotations to the GC, leading to

• completely lost draw operations (e.g. with 45° rotation)
• off-by-one errors in positions
• off-by-one errors in lines widths

The reason is in a faulty way in which the offset for odd line widths is calculated: https://github.com/eclipse-platform/eclipse.platform.swt/blob/4daf005a457fb50b929719cb5b3616e55cc2e5c4/bundles/org.eclipse.swt/Eclipse%20SWT/win32/org/eclipse/swt/graphics/GC.java#L326-L344

To reproduce an extreme case, apply a 45° rotation to the GC and try to draw something. Nothing will be drawn because a very large offset is erroneously calculated.

Expected Behavior

Offset calculation in the GC should be performed properly. No draw operations should be lost and no off-by-one errors should occur.

Additional Information

We have used the following patch which only hides the actual problem under specific usage conditions but does not solve it:

--- a/bundles/org.eclipse.swt/Eclipse SWT/win32/org/eclipse/swt/graphics/GC.java
+++ b/bundles/org.eclipse.swt/Eclipse SWT/win32/org/eclipse/swt/graphics/GC.java
@@ -320,16 +320,19 @@
 			long /*int*/ matrix = Gdip.Matrix_new(1, 0, 0, 1, 0, 0);
 			PointF point = new PointF();
 			point.X = point.Y = 1;
+			double lengthOfInitalVector = Math.sqrt(Math.pow(point.X, 2)+Math.pow(point.Y, 2));
 			Gdip.Graphics_GetTransform(gdipGraphics, matrix);
 			Gdip.Matrix_TransformVectors(matrix, point, 1);
 			Gdip.Matrix_delete(matrix);
-			float scaling = point.X;
+			double lengthOfTransformedVector = Math.sqrt(Math.pow(point.X, 2)+Math.pow(point.Y, 2));
+			float length = (float)(lengthOfTransformedVector/lengthOfInitalVector);
+			float scaling = length;
 			if (scaling < 0) scaling = -scaling;
 			float penWidth = data.lineWidth * scaling;
 			if (penWidth == 0 || ((int)penWidth % 2) == 1) {
 				data.gdipXOffset = 0.5f / scaling;
 			}
-			scaling = point.Y;
+			scaling = length;
 			if (scaling < 0) scaling = -scaling;
 			penWidth = data.lineWidth * scaling;
 			if (penWidth == 0 || ((int)penWidth % 2) == 1) {

[HeikoKlare]

I have already investigated the issue and developed a snippet for reproduction here: https://github.com/HeikoKlare/eclipse.platform.swt/tree/snippet-381-gc-transformation

It is extended with a potential fix and a comparison in the snippet here: https://github.com/HeikoKlare/eclipse.platform.swt/tree/gc-transformation-comparison

I will provide these artifacts, an issue and PR with the fix proposal to the Eclipse repositories as soon as possible.

[HeikoKlare]

Issue is documented here: eclipse-platform/eclipse.platform.swt#788
Snippet for reproduction is provided via this PR: eclipse-platform/eclipse.platform.swt#787
Fix proposal is provided via this PR: eclipse-platform/eclipse.platform.swt#789

[HeikoKlare]

Done by merging eclipse-platform/eclipse.platform.swt#789

[HeikoKlare]

Original PR reverted because of regression. Improved version has been merged: eclipse-platform/eclipse.platform.swt#1073