← Back to context

Comment by CHY872

10 years ago

You make a distinction between restrictions placed by the semantics of the language and the compiler not performing an optimisation.

This isn't a practical distinction for the end user. In both cases, no amount of money thrown at the optimising compiler will help, and the user will have to rewrite their algorithm.

I'd be very leery of replacing arbitrary algorithms. In particular, the user can make the optimisation unsafe so completely so quickly, just by adding debug statements or whatever.

Also, the examples you gave are all of specific classes of problem, which are probably quite easy to dedicate resources to because they affect many, and because optimised assembly exists.