Peter Alexander:
What about code that relies on overflow? It's well-defined
behaviour, so it should be expected that people rely on it (I
certainly do sometimes)
Do you rely on signed or unsigned overflow?
My opinions on this topic have changed few times.
A modern system language should offer the programmer both
integral types for the rare situations where the overflow or wrap
around are expected or acceptable, and other "default" integral
types to be used in all the other situations, where overflow or
wrap-around are unexpected and not desired. The implementation
then should offer ways to optionally perform run-time tests on
the second group of integrals.
A very good system language should also offer various means to
statically verify the bounds of a certain percentage of values
and expression results, to reduce the amount of run-time tests
needed (here things like "Liquid Types" help).
D currently doesn't have such safe built-in types, and it doesn't
offer means to create such efficient types in library code. I
think such means should be provided:
http://d.puremagic.com/issues/show_bug.cgi?id=9850
Bye,
bearophile
