Hi Stephan.
But here we could go the same path and move the normalization to #bitAnd:, #bitOr: and #bitXor.
Note that #bitOr: with normalized numbers does not need #normalize. Incidentally, some time ago I had produced faster versions of bitOr:, bitAnd: and bitXor: for LargeIntegers that were between 33 and 40 times faster or so. Unfortunately, they do not work with the new bitwise behavior standard for negative integers.
Another place that I think would be very nice to revamp is the digitLogic methods in Integer (suggestion by Dan Ingalls)... I thought about doing it but then run out of time and then fell out of my stack... so now you have reminded me of that. Perhaps after I finish some small tasks remaining...
Then we'd have the unique convention that the private methods don't normalize and the callers are responsible for that.
I like this very much. Andres.