Sound neat. But I wonder about those "quality assured computations that have been used and tested through many years" - when they convert the source code to a different language, using a brand new converter + a reimplemented standard library + 5% manual editing, how can they be sure the computations are still correct? Especially since they mention that the results are different due to 64-bit precision.
Those are good questions, I should have written more about that in the article! I'll give some brief answers:
Regarding the age of the converter: Even though it is new it is thought through. Check out the detailed documentation and technical articles on the website for an explanation. The converter basically deals with floating point computations carefully and in a way programmers commonly deal with them when doing scientific computations.
The reimplemented standard library are methods such as draw a line from point a to b, or draw this text at this point. Draw pixels at a certain coordinate, check if a key was pressed. So, the basic answer is that the functions were quite simple to get right. Had the library been more complicated, it would have been a different story.
Yes, you are right about the 5% edits. I should have added a section on Q&A done after the translation was done. It is of course very important to test code carefully after translation.