[OpenSmalltalk/opensmalltalk-vm] Bug in primitive 551 (SmallFloat64>> truncated) (#376)
There is a bug in an edge case in #truncated primitive in 64 bits Cog Spur (tested on Linux, both in Cuis and Squeak). It is when #truncated a Float that is slightly larger than SmallInteger maxVal.
SmallInteger maxVal --> 1152921504606846975
If we take (1152921504606846975.0 predecessor), to ensure a Float that still fits in a SmallInteger, everything works fine:
1152921504606846975.0 predecessor asTrueFraction --> 1152921504606846848
1152921504606846975.0 predecessor asTrueFraction --> class SmallInteger
1152921504606846975.0 predecessor truncated --> 1152921504606846848
But taking 1152921504606846976.0:
1152921504606846976.0 asTrueFraction --> 1152921504606846976 (correct result)
1152921504606846976.0 asTrueFraction class --> LargePositiveInteger (ok. Doesn't fit in SmallInteger)
1152921504606846976.0 truncated --> -1152921504606846976
The primitive is answering SmallInteger minVal, (a negative number!) but it should fail, as the correct result doesn't fit in a SmallInteger.