This is an automated email from the ASF dual-hosted git repository.
desruisseaux pushed a commit to branch geoapi-4.0
in repository https://gitbox.apache.org/repos/asf/sis.git
The following commit(s) were added to refs/heads/geoapi-4.0 by this push:
new 8ec8097 Better fix for rounding error, using the actual error as
determined by Math.ulp(double).
8ec8097 is described below
commit 8ec8097fee26cab76d696721208fc170a19d0daa
Author: Martin Desruisseaux <[email protected]>
AuthorDate: Thu Oct 18 09:48:13 2018 +0200
Better fix for rounding error, using the actual error as determined by
Math.ulp(double).
---
.../apache/sis/measure/SexagesimalConverter.java | 35 +++++++++++++++++-----
1 file changed, 27 insertions(+), 8 deletions(-)
diff --git
a/core/sis-utility/src/main/java/org/apache/sis/measure/SexagesimalConverter.java
b/core/sis-utility/src/main/java/org/apache/sis/measure/SexagesimalConverter.java
index dc1dbaf..857db03 100644
---
a/core/sis-utility/src/main/java/org/apache/sis/measure/SexagesimalConverter.java
+++
b/core/sis-utility/src/main/java/org/apache/sis/measure/SexagesimalConverter.java
@@ -24,6 +24,8 @@ import javax.measure.UnitConverter;
import org.apache.sis.util.ArgumentChecks;
import org.apache.sis.util.resources.Errors;
import org.apache.sis.util.resources.Vocabulary;
+import org.apache.sis.internal.util.Numerics;
+import org.apache.sis.math.MathFunctions;
import static org.apache.sis.math.MathFunctions.truncate;
@@ -281,10 +283,27 @@ class SexagesimalConverter extends AbstractConverter {
* <li>value * 10000 = 465708.66000000003</li>
* <li>deg = 46, min = 57, deg = 8.660000000032596</li>
* </ol>
+ *
+ * We perform a rounding based on the representation in base 10
because extractions of degrees and
+ * minutes fields from the sexagesimal value themselves use arithmetic
in base 10. This conversion
+ * is used in contexts where the sexagesimal value, as shown in a
number in base 10, is definitive.
+ *
+ * @param remainder the value to fix, after other fields (degrees
and/or minutes) have been subtracted.
+ * @param magnitude value of {@code remainder} before the degrees
and/or minutes were subtracted.
*/
- private static double fixRoundingError(final double remainder) {
- final double c = Math.rint(remainder * 1E+6) / 1E+6;
- return (Math.abs(remainder - c) < 1E-9) ? c : remainder;
+ private static double fixRoundingError(double remainder, final double
magnitude) {
+ /*
+ * We use 1 ULP because the double value parsed from a string
representation was at 0.5 ULP
+ * from the real value, and the multiplication by 'divider' add
another 0.5 ULP rounding error.
+ * Removal of degrees and/or minutes fields as integers do not add
rounding errors.
+ */
+ int p = Math.getExponent(Math.ulp(magnitude)); // Power
of 2 (negative for fractional value).
+ if (p < 0 && p >= -Numerics.SIGNIFICAND_SIZE) { //
Precision is a fraction digit >= Math.ulp(1).
+ p = Numerics.toExp10(-p); //
Positive power of 10, rounded to lower value.
+ final double scale = MathFunctions.pow10(p);
+ remainder = Math.rint(remainder * scale) / scale;
+ }
+ return remainder;
}
/**
@@ -294,18 +313,18 @@ class SexagesimalConverter extends AbstractConverter {
*/
@Override
public double convert(final double angle) throws
IllegalArgumentException {
- double deg,min,sec;
+ double deg,min,sec,mgn;
if (hasSeconds) {
- sec = angle * divider;
+ sec = mgn = angle * divider;
deg = truncate(sec/10000); sec -= 10000*deg;
min = truncate(sec/ 100); sec -= 100*min;
- sec = fixRoundingError(sec);
+ sec = fixRoundingError(sec, mgn);
} else {
sec = 0;
- min = angle * divider;
+ min = mgn = angle * divider;
deg = truncate(min / 100);
min -= deg * 100;
- min = fixRoundingError(min);
+ min = fixRoundingError(min, mgn);
}
if (min <= -60 || min >= 60) { // Do
not enter for NaN
if (Math.abs(Math.abs(min) - 100) <= (EPS * 100)) {
