I waited a year with the hope that dxli could solve these issues;
he did solve 788 somehow by using the eigen lib;
however, I don't think we need to add a dependency to solve these problems;
and I'm not willing to pull in all the changes he made without understanding them.
788 will probably be solved simply by replacing the current generic "quadratic" approach
with dedicated line-circle and circle-circle intersection algorithms.
Vasilis has offered to do the work next month,
which gives anyone interested some time to debug and salvage the current approach.
is it possible to test that code and see if the intersections work in your case?
I am not very clear what this exactly get solved with the patches.
My math is very limited and I don't understand it to get better precision and to understand the implications.
For what I see it sounds like precise enough for CAD work in the area where LibreCAD is active (not medical equipment and anything object you can walk around).