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.