~~I'm sorry but I cannot find file MSDFeatureDetectorTest.java to be fixed. ~~

Very sorry, I found it in OpenCV Contrib xfeatures2d.

https://github.com/opencv/opencv_contrib/blob/4.x/modules/xfeatures2d/misc/java/test/MSDFeatureDetectorTest.java

package org.opencv.test.features2d;
import org.opencv.test.OpenCVTestCase;
import org.opencv.test.OpenCVTestRunner;
import org.opencv.xfeatures2d.MSDDetector;

Implementation package is xfeatures2d, but test package is in feature2d.

In opencv-contrib, test is passed with merging this patch. So I retry to run CI.

Usually when dealing with floating point numbers, my concern is to not lose "data" when saving to text for debugging purposes.
Looks like it is good at least for double: https://coliru.stacked-crooked.com/a/be26c5d838792452

If the text output is too heavy, maybe an option to set the desired precision for floating point numbers?

Thank you for your comment. I thought same suggestion(it is old pull request).
But but these two requirement are slightly contradictory. So it is hard to realize.

1. The persistence/file storage must be able to read the same data which is set at written.
2. The persistence/file storage should have an option to set the desired precision.

If we can set the desired precision for floating point numbers, some information of input data will be lost. However we cannot explain that what truncate rule was applied. The truncated number only left.

I believe the persistent function should not be truncated information of inputs. The reading value should be same as writing value. I think this is not unnatural.

And the number of bytes required for the "%.17g" expression after modification is smaller than the "%.16e" expression before modification.

Following are conversion example in test cases.

Thank you very much!

(I'm sorry, it is repeatly)
This patch needs to merge opencv/opencv_contrib#3714
If not, OpenCV regression test will be failed on MSDFeatureDetectorTest.

Thank you very much ! opencv/opencv_contrib#3714 is merged.