[OpenJDK Rasterizer] Marlin artefact issue (Pisces too ?)

[Laurent Bourgès]

Jim,

I should create a new bug concerning both Marlin & Pisces but I prefer
discussing the problem first.

A Marlin user reported me an issue with (text outline) rendering artefacts
(due to partial cleanup in arrays).
Using Marlin with -Dsun.java2d.renderer.doChecks=true (logs re-enabled), we
detected that edgeBucketCounts arrays were not properly zero-filled due to
NaN coordinates ie array[0] > 0 !

As it only happens with very large coordinates and round joins, we finally
got a reproducer test class.

However, we tracked the problem down into the Stroker.drawBezApproxForArc()
method (from Pisces):
cv = NaN.

I am not very good at curve maths but I figured out that cosext2 means
cos(ext)^2 as there is below sqrt(0.5 +/- cosext2) !

Of course, the problem is sqrt(negative) gives NaN !

Moreover, it only happens with very large out-of-clip coordinates (2M), see
the test class !
It probably means that float values have not enough precision in previous
math operations and it finally overflows 0.5 !

Is it correct to clamp cosext2 in [-0.5, 0.5] range as I propose ?

    // the input arc defined by omx,omy and mx,my must span <= 90 degrees.
    private void drawBezApproxForArc(final float cx, final float cy,
                                     final float omx, final float omy,
                                     final float mx, final float my,
                                     boolean rev)
    {
        float cosext2 = (omx * mx + omy * my) / (2f * lineWidth2 *
lineWidth2);

// PROPOSED FIX TO BE CONFIRMED:
        // clamp value within [-0.5, 0.5] range:
        if (cosext2 < -0.5f) {
            cosext2 = -0.5f;
        } else if (cosext2 > 0.5f) {
            cosext2 = 0.5f;
        }

        // cv is the length of P1-P0 and P2-P3 divided by the radius of the
arc
        // (so, cv assumes the arc has radius 1). P0, P1, P2, P3 are the
points that
        // define the bezier curve we're computing.
        // It is computed using the constraints that P1-P0 and P3-P2 are
parallel
        // to the arc tangents at the endpoints, and that |P1-P0|=|P3-P2|.
        float cv = (float) ((4.0 / 3.0) * sqrt(0.5 - cosext2) /
                            (1.0 + sqrt(cosext2 + 0.5)));

If you have any explanations,  please tell me !

PS: Anyway NaN handling must be properly tested and filtered in Marlin's
pipeline (like DuctusRenderingEngine does) ...
but here it concerns intermediate points (not inputs).

Test code:
/*
 * Copyright (c) 2015, Oracle and/or its affiliates. All rights reserved.
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 * or visit www.oracle.com if you need additional information or have any
 * questions.
 */

import java.awt.BasicStroke;
import java.awt.Color;
import java.awt.Font;
import java.awt.Graphics2D;
import java.awt.RenderingHints;
import java.awt.font.FontRenderContext;
import java.awt.font.GlyphVector;
import java.awt.geom.AffineTransform;
import java.awt.image.BufferedImage;
import java.io.File;
import java.io.IOException;
import javax.imageio.ImageIO;

/**
 * @test
 * @summary Check the Stroker.drawBezApproxForArc() bug:
 * abs(cosext2) > 0.5 generates curves with NaN coordinates
 * @run main TextClipErrorTest
 */
public class TextClipErrorTest {

    public static void main(String[] args) {
        BufferedImage image = new BufferedImage(256, 256,
                BufferedImage.TYPE_INT_ARGB);

        Graphics2D g2d = image.createGraphics();
        g2d.setColor(Color.red);
        try {
            g2d.setRenderingHint(RenderingHints.KEY_ANTIALIASING,
                    RenderingHints.VALUE_ANTIALIAS_ON);

            Font font = g2d.getFont();
            FontRenderContext frc = new FontRenderContext(
                    new AffineTransform(), true, true);

            GlyphVector gv1 = font.createGlyphVector(frc, "\u00d6");

            g2d.setStroke(new BasicStroke(4.0f,
                    BasicStroke.CAP_ROUND, BasicStroke.JOIN_ROUND));

            AffineTransform at1 = AffineTransform.getTranslateInstance(
                    -2091202.554154681, 5548.601436981691);
            g2d.draw(at1.createTransformedShape(gv1.getOutline()));

            GlyphVector gv2 = font.createGlyphVector(frc, "Test 2");

            AffineTransform at2 = AffineTransform.getTranslateInstance(
//                    -218.1810476789251, 85.12774919422463);
                    10, 50);
            g2d.draw(at2.createTransformedShape(gv2.getOutline()));


            final File file = new File("TextClipErrorTest.png");
            System.out.println("Writing file: " + file.getAbsolutePath());
            ImageIO.write(image, "PNG", file);

        } catch (IOException ex) {
            ex.printStackTrace();
        }
        finally {
            g2d.dispose();
        }
    }
}


Cheers,
Laurent
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/graphics-rasterizer-dev/attachments/20151203/e1e5871a/attachment-0001.html>