phetsims / alpenglow Goto Github PK

@jonathanolson found a bug that a shear matrix ( eg. [ [ 1, 1, 0 ], [0, 1, 0], [0, 0, 1]] ), was not being handled int the split function of RenderLinearGradient.

(Fairly difficult, it's unclear if this is the linear/radial gradients having an issue, RenderFromNode having an issue, or an internal other Alpenglow issue. Will require knowledge of color spaces and blending, since there might be a bad blend going on - possibly only on gradients with transparency)

There are some minor visual differences that I'm interested in. I am NOT interested in the rendering differences that are size-small (like the slight differences in rendering small lines or text), however I am interested in the gradient shifts (both linear and radial) that seem to happen when using RenderFromNode.showSim().

In Pendulum Lab, this is most noticeable on the ResetAllButton, but also visible on the gradient on the masses themselves.

To reproduce:

1. Check out Scenery's 'vello' branch (at least at commit phetsims/scenery@e7d846a, pull once it's checked out - I've merged main into it)
2. Move font files into scenery/js/display/vello/ (talk to @jonathanolson)
3. Open up Pendulum Lab, go to the first screen
4. Run phet.alpenglow.RenderFromNode.showSim() in the console. It will display a Canvas OVER everything else with the alpenglow software-rendered image.
5. To switch back and forth to see visual differences, select the Canvas in devtools, and toggle visibility on it (might need to add the CSS inline), like the following:

Alpenglow-rendered:

Scenery (SVG) rendered:

Alpenglow:

Scenery SVG:

This issue will track review comments and questions for ClippableFace.

This issue will be used to track review questions/comments as well as unit test commits for RenderPrograms

(Hard, this is the bug I was describing with the circular clipping where we're getting a mismatch of edges).

It seems like EdgedFace.getBinaryCircularClip is outputting invalid data. As discussed, later stages need the start/end points of edges to match up precisely. I probably screwed up something like this in CircularClipping.binaryCircularClipEdges.

To reproduce, apply the patch below (or just... enable the one test in RasterizeTests.ts), and run the unit tests.

Subject: [PATCH] Enable test
---
Index: js/raster/RasterizeTests.ts
IDEA additional info:
Subsystem: com.intellij.openapi.diff.impl.patch.CharsetEP
<+>UTF-8
===================================================================
diff --git a/js/raster/RasterizeTests.ts b/js/raster/RasterizeTests.ts
--- a/js/raster/RasterizeTests.ts	(revision 5e56f9c48be2d1fe4b67f2f93ce2980a7d1f4e8d)
+++ b/js/raster/RasterizeTests.ts	(date 1697254991298)
@@ -14,7 +14,7 @@
 
 QUnit.module( 'Rasterize' );
 
-QUnit.skip( 'split clip start end matches', assert => {
+QUnit.test( 'split clip start end matches', assert => {
 
   const colors = _.range( 0, 100 ).map( i => new RenderColor( new Vector4( i / 100, 0, 0, 1 ) ) );
   const outputSize = 768 * 2;

If you browse https://github.com/phetsims/alpenglow/blob/main/doc/raster-clip.md, it just shows a not-very-helpful error.

@marlitas, any thoughts on what I'm doing wrong here? It is working in the WebStorm/Intellij preview, AND the Mermaid live editor

(Moderate difficulty, somewhat related to #5, but hopefully slightly easier to track down)

Run things similar to #5, but with geometric-optics. First screen.

Scenery SVG:

Alpenglow:

(as a side note... check out the antialiasing differences on the black line between the two! some is going on with Chrome SVG)