This is fantastic, thank you! I'll need some time to go through it in detail.

Regarding the test TestHardBioUnits#test4A1I, I've had a look and it looks like a bug in SuperPositionQCP. If you switch to use SuperPositionSVD, then the tests works fine. I think that is a good solution for the problem here.

We can then create a separate issue for the SuperPositionQCP bug.

Just played around with the QCP algorithm. The test will find the correct solution if I 'trim' the precision of the double values coming from BinaryCIF. Otherwise the result is completely off. Kinda strange.

Something along the lines of:

private Point3d[] prepare(Point3d[] point3ds) {
    return Stream.of(point3ds)
            .map(v -> new Point3d((int) (v.x * 1000) * 0.001,
                    (int) (v.y * 1000) * 0.001,
                    (int) (v.z * 1000) * 0.001))
            .toArray(Point3d[]::new);
}

Also one missing thing is documenting the breaking changes. Could you add them to the CHANGELOG.md file so that when we release 6.0.0 it's already done?

Done in 53982b0.

Sounds good. I've opened an issue (#914).

I'll merge this next week if there are no other comments.

I'll merge this next week if there are no other comments.

Would next week be a good time to discuss what might go in the version 6.0.0 release? There is a roadmap issue #879

Would next week be a good time to discuss what might go in the version 6.0.0 release? There is a roadmap issue #879

Most items covered in #879 are done now. But if there's anything else that needs to be included, please add any ideas to the list. It is a good opportunity to clean up APIs.

I've reverted the changes to QCP (that would change the threshold).
Instead TestHardBioUnits#test4A1I now uses SVD and there's a dedicated test for the issue in TestSuperPositionQCP as discussed in #914.