S1ChordAngleTest Failure

[ericangle]

I get the following failure when I run mvn test:

[ERROR] Tests run: 20, Failures: 1, Errors: 0, Skipped: 0, Time elapsed: 0.027 s <<< FAILURE! - in com.google.common.geometry.S1ChordAngleTest
[ERROR] com.google.common.geometry.S1ChordAngleTest.testTrigonometry  Time elapsed: 0.003 s  <<< FAILURE!
junit.framework.AssertionFailedError: expected:<0.15643446504023098> but was:<0.1564344650402278>
	at junit.framework.Assert.fail(Assert.java:57)
	...

Note that, as I mentioned in issue 43, I had to change the guava.version from 32.0-jre to 32.0.0-jre in pom.xml to get mvn compile to succeed. That is the only local diff I have compared to the master branch.

In case it's relevant, when I run java -version, that returns

java version "22.0.1" 2024-04-16
Java(TM) SE Runtime Environment (build 22.0.1+8-16)
Java HotSpot(TM) 64-Bit Server VM (build 22.0.1+8-16, mixed mode, sharing)

[jmr]

What JDK are you using? For OpenJDK from Oracle, 22 is not supported: https://endoflife.date/openjdk-builds-from-oracle

Can you try again at HEAD and if that doesn't work, with 24?

I had to change the guava.version

This should no longer be necessary.

[ericangle]

@jmr I just created a fresh git clone of this repo. git log on the master branch shows one commit (4a6abc1) after 7d86027. I also updated java to 24.0.1. Then I ran mvn compile and mvn test, and got the same error:

[ERROR] Failures: 
[ERROR]   S1ChordAngleTest.testTrigonometry:207 expected:<0.15643446504023098> but was:<0.1564344650402278>
[ERROR]   S2BooleanOperationTest.testMeridianSplitting:1983->checkMeridianSplitting:1935 Actual string: # # 10.0014925269841:180, 10:-179, 9:-179, 9.00134850712993:180, 8.00481316618608:180, 8:-178, 11:-178, 11.00654129428:180; 9.00134850712993:180, 9:179, 10:179, 10.0014925269841:180, 11.00654129428:180, 11:178, 8:178, 8.00481316618608:180
Java expected: '# # 10.0014925269841:180, 10:-179, 9:-179, 9.00134850712993:180, 8.00481316618607:180, 8:-178, 11:-178, 11.00654129428:180; 9.00134850712993:180, 9:179, 10:179, 10.0014925269841:180, 11.00654129428:180, 11:178, 8:178, 8.00481316618607:180'
J2CL expected: '# # 10.001492526984:180, 10:-179, 9:-179, 9.0013485071299:180, 8.004813166186:180, 8:-178, 11:-178, 11.00654129428:180; 9.0013485071299:180, 9:179, 10:179, 10.001492526984:180, 11.00654129428:180, 11:178, 8:178, 8.004813166186:180'
[INFO] 
[ERROR] Tests run: 1477, Failures: 2, Errors: 0, Skipped: 0

This is on macOS 15.4.1. java -version shows

java version "24.0.1" 2025-04-15
Java(TM) SE Runtime Environment (build 24.0.1+9-30)
Java HotSpot(TM) 64-Bit Server VM (build 24.0.1+9-30, mixed mode, sharing)

[jmr]

Can you print out all subexpressions of the failing assertion?

s2-geometry-library-java/tests/tests/com/google/common/geometry/S1ChordAngleTest.java

Line 211 in 4a6abc1

assertEquals(atan(tan(radians)), atan(S1ChordAngle.tan(angle)), 1e-15);

radians, tan(radians), atan(tan(radians)), angle, ... Maybe that will help to figure out which part is the problem. The error bound may just be wrong. If I increase the iterations I also get failures. @rsned

[ericangle]

Both line 207 (sine test) and line 211 (tan/arctan test) fail for iter = 19. Here are the subexpressions:

radians = 2.9845130209103035
angle = 171.00000000000014d

Line 207

sin(radians) = 0.15643446504023098
S1ChordAngle.sin(angle) = 0.1564344650402278
S1ChordAngle.sin(angle) - sin(radians) = -3.191891195797325E-15

Line 211

tan(radians) = 2.9845130209103035
S1ChordAngle.tan(angle) = -0.1583844403245331
atan(tan(radians)) = -0.15707963267948977
atan(S1ChordAngle.tan(angle)) = -0.15707963267948655
atan(tan(radians)) - atan(S1ChordAngle.tan(angle)) = -3.219646771412954E-15

Incidentally, S2BooleanOperationTest.testMeridianSplitting() also fails at line 1983, but I could write a separate issue for that error if one doesn't already exist. The failure is:

[ERROR]   S2BooleanOperationTest.testMeridianSplitting:1983->checkMeridianSplitting:1935 Actual string: # # 10.0014925269841:180, 10:-179, 9:-179, 9.00134850712993:180, 8.00481316618608:180, 8:-178, 11:-178, 11.00654129428:180; 9.00134850712993:180, 9:179, 10:179, 10.0014925269841:180, 11.00654129428:180, 11:178, 8:178, 8.00481316618608:180
Java expected: '# # 10.0014925269841:180, 10:-179, 9:-179, 9.00134850712993:180, 8.00481316618607:180, 8:-178, 11:-178, 11.00654129428:180; 9.00134850712993:180, 9:179, 10:179, 10.0014925269841:180, 11.00654129428:180, 11:178, 8:178, 8.00481316618607:180'
J2CL expected: '# # 10.001492526984:180, 10:-179, 9:-179, 9.0013485071299:180, 8.004813166186:180, 8:-178, 11:-178, 11.00654129428:180; 9.0013485071299:180, 9:179, 10:179, 10.001492526984:180, 11.00654129428:180, 11:178, 8:178, 8.004813166186:180'

[ericangle]

Just following up @jmr

[jmr]

Does this fail for you?

  @Test
  public void testTrigonometryGitHubIssue44() {
    double radians = radians = 2.9845130209103035;
    S1ChordAngle angle = S1ChordAngle.fromS1Angle(S1Angle.radians(radians));                             
    assertExactly(angle.radians(), 2.9845130209103035);                                                  
    assertExactly(sin(radians), 0.15643446504023098);
    assertExactly(S1ChordAngle.sin(angle), 0.1564344650402306);                                          
  }  

It looks like your S1ChordAngle.sin(angle) = 0.1564344650402278 value is pretty far off.

[ericangle]

assertExactly(angle.radians(), 2.9845130209103035) fails with error java.lang.AssertionError: expected:<2.984513020910306> but was:<2.9845130209103035>

assertExactly(sin(radians), 0.15643446504023098) passes

assertExactly(S1ChordAngle.sin(angle), 0.1564344650402306) fails with error java.lang.AssertionError: expected:<0.1564344650402278> but was:<0.1564344650402306>

[jmr]

assertExactly(angle.radians(), 2.9845130209103035) fails with error java.lang.AssertionError: expected:<2.984513020910306> but was:<2.9845130209103035>

This is pretty basic and means precision is being lost in the double -> S1ChordAngle construction. Is there some sort of "fast math" option that's being used?

It seems like we should have gotten more complaints, but maybe you're the only one running the tests on macOS.