After seeing the post of Bert about ldexp http://croquetweak.blogspot.fr/2014/08/deconstructing-floats-frexp-and-ldexp...
I wanted to check what's in image. But I realized that current Squeak fallback code for Float>>timesTwoPower: is not OK. It's the version from Etoys-Squeakland corresponding to old squeak implementation
The correct one should be that of Kernel-nice.900.mcz http://lists.squeakfoundation.org/pipermail/packages/2015-February/007538.ht... Fortunately it has been moved (duplicated) in-between in both subclasses, thus maybe the Squeakland one is not an override? I can't really trace what happened because it's hidden in the spur transition. But IMO it would be preferable to implement the Smalltalk one in Float only once, and rely on ^super timesTwoPower: in case of primitive failure.
I saw some other instances of code that was fixed but reintroduced (like all the character scanner duplication). What's the strategy about it, can we touch Squeakland, or shall we refrain?