Replace JavaFX text with AWT

[breiler]

This PR aims to replace JavaFX font with AWT for extruding texts

[breiler]

I'll look into why those tests are failing.

[madhephaestus]

One of the external checks that i do is to run the StlExportTest and put TextStl2.stl through PrusaSlicer or Bamboo Slicer. they are very poor quality slicing algorithms and acts as a worst case scenario mesh integrity test.

I use MeshLab to look at TextStl2.stl for any non manifold points or edges, and for any boundary edges. There should be 0 of all of those.

[breiler]

Thanks

Here is a screenshot from Prusa Slicer, the top is from this branch and the bottom is from the development branch:

Here is the new version in meshlab:

And here is the development version:

My version is obviously worse, but what is the metric I that I should be looking for? Should it look as nice as possible or attempt to count the number of errors?

[madhephaestus]

In the StlExporrtTest there is a line commented out at the top, set it to 1e-11 and see if the errors go away entirely. It indicates issues with the initial point data having duplicate points or near-duplicate points. It could also indicate a polygon with backtracking or even crossing polygons. I had to do a ton of data correction on the javafx text and it has always been a source of bad polygons into the csg opperations. If you run the slice in Prusa see if there are missing layers. This is the "good enough" test, if prusa can slice it. I use text as the stress test since the polygons are generated and often full of errors.

…

On Sat, Dec 14, 2024, 2:05 AM Joacim Breiler ***@***.***> wrote: Thanks Here is a screenshot from Prusa Slicer, the top is from this branch and the bottom is from the development branch: image.png (view on web) <https://github.com/user-attachments/assets/8e309fd2-f5da-4285-966f-3615f4bc356e> Here is the new version in meshlab: image.png (view on web) <https://github.com/user-attachments/assets/6df94778-274c-43ec-a1cd-9c5dd54ff6b7> And here is the development version: image.png (view on web) <https://github.com/user-attachments/assets/a0da1b79-c661-4e8d-80f2-4ab093c22584> My version is obviously worse, but what is the metric I that I should be looking for? Should it look as nice as possible or attempt to count the number of errors? — Reply to this email directly, view it on GitHub <#63 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJSKRVNIVXKHMTBNHLVGVL2FPKB5AVCNFSM6AAAAABTQNX2VOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKNBSHEYTMMRTGM> . You are receiving this because you commented.Message ID: ***@***.***>

[breiler]

Using the development branch, uncommenting that line will slightly improve the result, but for me there are still missing triangles:

Using this code:

package eu.mihosoft.vrl.v3d;

import java.io.File;
import java.io.IOException;
import java.nio.file.Paths;

import org.junit.Test;

public class StlExportTest {

	@Test
	public void makeBadSTL() throws IOException {
		Plane.setEPSILON(1.0e-11);
		//Vector3d.setEXPORTEPSILON(10);
		CSG.setUseGPU(false);
		CSG.setPreventNonManifoldTriangles(true);
		CSG badExport2 = CSG.text(" A QUICK BROWN FOX JUMPS OVER THE LAZY DOG", 10,30,"Serif Regular").movey(30);
		System.out.println("First text loaded");
		CSG badExport = CSG.text("THis is some text a quick brown fox jumps over the lazy dog.", 10);
		System.out.println("Second text loaded");
//		badExport2=new Cube(20).toCSG().movey(30);
//		badExport=new Cube(20).toCSG();
		
		badExport=badExport.union(badExport2);
		//String filename ="TextStl.stl";
		FileUtil.write(Paths.get("TextStl.stl"),
				badExport.toStlString());
		System.out.println("Load saved stl");
		File file = new File("TextStl.stl");
		CSG loaded = STL.file(file.toPath());
		System.out.println("Perform difference");
		badExport=loaded.scaleToMeasurmentX(160).scaleToMeasurmentY(30);
		badExport=new Cube(180,40,10).toCSG().toZMin().toXMin().toYMin().movey(-5).difference(badExport).rotx(35).roty(45);
		FileUtil.write(Paths.get("TextStl2.stl"),
				badExport.toStlString());
	}

}

[madhephaestus]

So I adjusted the 2 epsilon parameters and was able to get down to 4 boundary faces in the development branch. The issue is related to how the CSG stack decides 2 points are coincident vs how other mesh manipulating stacks assume points are coincident. Text seems to be a source of difficult polygons, which is why i use it as the stress test for repairing damaged meshes. I have been addressing this recently in order to support Prusa and Bamboo slicers.

[breiler]

How are you measuring this? From just looking at it in Prusa with the default settings and in Meshlab I can clearly see polygons missing when uncommenting both those lines in the development branch:

Plane.setEPSILON(1.0e-11);
Vector3d.setEXPORTEPSILON(10);

TextStl2.stl.txt

What OS are you using? For instance I saw a big difference in how fonts are vectorized on Linux/Windows vs MacOSX.

[madhephaestus]

Plane.setEPSILON(1.0e-10); Vector3d.setEXPORTEPSILON(1.0e-9);
I used these settings and was able to get a lower number of failed polygons on Linux. I have not run these tests on WIndows or Macos, but you are right, OS's process fonts differently and produce very different results, which requires the loader to do some point validation before continuing with the polygon as a source for a 3d shape.

[madhephaestus]

One other thing to look at is to see if the triangles are missing, or are just backwards, IE wound clockwise instead of Counter-clockwise. The CW wound trinagles come into MeshLab as a dark grey instead of light grey.

When a polygon gets removed from the CSG its because the triangulation algorithm failed to find a convex corner after tracing over the polygon. This could be because of a backtrack, a portion of the polygon that juts out, then backtracks over the same line forming a 0 area section, or it can be 2 points that are coincident, or an edge of the polygon that crosses another edge of the polygon. I have seen all 3 of these types of polygon irregularities when generating shapes using fonts.

[madhephaestus]

You can also look at TextStl.stl only for the moment, the second one test loading back in the first STL and performing operations on it. If you can make just the text stl with no non-manifold features and no boundary edges, then we can call that a Win and merge this PR.

[breiler]

Sorry for being a newbie, now that I know what to look for it makes a bit more sense. This is the result when running the tests. Don't merge it just yet, I would like to do some more tests.

My branch
TextStl.stl:

TextStl2.stl:

Development branch
TextStl.stl:

TextStl2.stl