I just fixed a bug related to the unsupported floating point. ~master phobos compiles again and runs the unit-tests.
I will probably have to spent the rest of the month with fixing bugs that function-call support has uncovered.
Since now newCTFE sees much more code before bailing out. I have also determined the cause of most of the miscompiled code. The cause is the incorrect treatment of utf8. (currently the evaluator really just does (dchar c= (_c & 7f)) which leads to corrupted strings in mixins; and that has pretty crazy effects. It was very hard to find this.