Undefined Behaviour as usual

Posted Feb 23, 2024 22:18 UTC (Fri) by khim (subscriber, #9252)
In reply to: Undefined Behaviour as usual by fw
Parent article: Stenberg: DISPUTED, not REJECTED

> Not really, Java and C# do not have undefined behavior, and yet there are optimizing compilers.

Java and C# absolutely do have undefined behavior. It's just handled like Rust handles it: “safe” language guarantees absence of UB by compiler whole “unsafe” part allows on to write programs with UB.

Java forces you to write these parts in entirely different language using JNI while C# have an unsafe subset, similarly to Rust, but in both cases UB is still, very much, form the basis for all optimizations.

> Even for C, it's a rather extreme position to say that register allocation (probably among the top three optimizations to implement in a compiler for current architectures) depends on undefined behavior.

Of course it does. Everything in C depends on absence of undefined behavior. Simply because it's permitted to convert pointer to function into pointer to chat *. This may break your program if it does register allocation in a different way.

And it's not even theoretical issue! Back in MS-DOS era register variables could only use si and di (and they weren't used for anything else) and thus it was possible to write code that would [ab]use that in it's signal handler.

> For others like constant propagation it's a bit of a stretch, too.

That one may be exploiting by finding and changing constants in the compiler code. And that, too, was used back when compilers weren't smart enough to break such tricks.