Why don't we switch to C like floating pointed arithmetic instead of automatic expansion to reals?

[Fool via Digitalmars-d]

Here is a relevant example:

https://hal.inria.fr/inria-00171497v1/document

It is used in at least one real world geometric modeling system.