i disagree. As i said earlier, a square root function result is not defined for arguments less than zero in real (R) set.
Right, only for complex (C). R < C [C contains R].
This is strictly mathematical behavior, same as division does not defines a value for zero.
My _Elementary Functions_ book has a nice section which defines the square root of -1 to be 1i. I pick up _Handbook of Physical Calculations_ and find an entire section on the mathematics of alternating current which uses complex math exclusively. I see why, given Smalltalk's history, one does not wish to change the way things are done. This is a bit different from implying complex mathematics is not mathematics.
Just think about, how many code i could break, if i return an Infinity (or something like this) as a result of division by zero instead of signaling error.
My recollection (again, not to be trusted) is that Java returns an IEEE infinity rather than throwing an exception. I do not recommend this either, which is why I am puzzled by #ln returning a NaN. Again, my recommendation is that the current Complex implementation be removed from Pharo-Core. Reluctant as I am to suggest it, an allowComplexResults global preference could be used, with the usual "Here Be Dragons" warnings. Such a preference would clearly state that it is to allow functions like #ln and #sqrt to return complex results and would only be settable if a Complex number package was loaded. Cheers, -KenD PS: My apologies, but my laptop died and I am borrowing time on someone else's computer. I intent to browse "sqrt senders" when I get another machine to see what code is breaking.