Possible Bug In Java Implementation

[Anonymous]

To whom it may concern,
When I browse here:

http://geographiclib.sourceforge.net/cgi-bin/GeodSolve

...and enter the following values:

90.0 10.0 180.0 -1000000.0

...to solve the direct problem using all default values,
I get what appears to be a reasonable latitude and longitude
for the second point:

lat2 lon2 fazi2 (°) = 81.04623282 -170.00000000 0.00000000

But, when I download GeographicLib-1.45.zip from here:

https://sourceforge.net/projects/geographiclib/files/distrib/

...and then compile and run the Java
version, I get the following instead:

lat2 lon2 fazi2 (°) = 81.04623281595062 10.0 0.0

The longitude in the Java version appears to be incorrect.
I would expect these two approaches to yield the same result.
My test code:

public static void main(String[] args)
{
GeodesicData geo = Geodesic.WGS84.Direct(90.0, 10.0, 180.0, -1000000.0);
System.out.println("lat2 lon2 fazi2 (°) = " + geo.lat2 + " "
+ geo.lon2 + " " + geo.azi2);
}

This same behavior occurs at the south pole as well.
I compiled the Java code with Java V1.8.0_74 64-bit on
Windows 7. Maybe I'm not using the classes correctly?
Any help would be appreciated.

[Charles Karney]

Yes, this is a bug which was introduced in version 1.44 and it is
present also in version 1.45. It's fixed in version 1.46-SNAPSHOT.
Here's the patch to 1.45

diff --git a/java/src/main/java/net/sf/geographiclib/GeodesicLine.java b/java/src/main/java/net/sf/geographiclib/GeodesicLine.java
index 307b316..9e7cc62 100644
--- a/java/src/main/java/net/sf/geographiclib/GeodesicLine.java
+++ b/java/src/main/java/net/sf/geographiclib/GeodesicLine.java
@@ -179,7 +179,8 @@ public class GeodesicLine {
     { Pair p = GeoMath.sincosd(GeoMath.AngRound(_lat1));
       sbet1 = _f1 * p.first; cbet1 = p.second; }
     // Ensure cbet1 = +epsilon at poles
-    { Pair p = GeoMath.norm(sbet1, cbet1); sbet1 = p.first; cbet1 = p.second; }
+    { Pair p = GeoMath.norm(sbet1, cbet1);
+      sbet1 = p.first; cbet1 = Math.max(Geodesic.tiny_, p.second); }
     _dn1 = Math.sqrt(1 + g._ep2 * GeoMath.sq(sbet1));

     // Evaluate alp0 from sin(alp1) * cos(bet1) = sin(alp0),

I'll add a unit test to cover this issue.

Thanks for bringing this to my attention.