On: 1BRC merykitty's magic SWAR: 8 lines of code expla...

[Karellen]

Anything is allowed to happen if your program invokes UB. All bets are off (according to the C standard/abstract machine) for the rest of the runtime of your program, from the first time it happens.

In fact, because the compiler is allowed to re-order your program provided that the end result of any non-UB computation is the same (the "as-if" rule[0]), if part of your program invokes UB, it's possible for that to "reach backwards" and invalidate computations that appear to happen before the UB is/was invoked (by a straightforward reading of the source code).

[0] https://en.wikipedia.org/wiki/As-if_rule

[o11c → Karellen]

My point is: we normally speak of UB as a C-source-specific thing. But some UB exists even in compiled code from any language.

[Karellen → o11c]

That's because UB has a well-defined meaning in the C programming language specification, for over 35 years now. And C has been one of the most widespread programming languages in use over that time.

One of the important things about UB and C is that C is a language defined by a standard, not any one compiler - and there are a lot of compilers. If a C compiler does something weird with your source code, we can say whether the compiler is correct or not, or whether your code is at fault. For many languages, that is not the case, and "whatever the compiler does to your code" is de facto correct, and "whatever the compiled version of your code happens to do on any given platform" is well-defined, even if it wasn't what you meant.

I'd argue that if a compiled language does not define what it means by UB, it doesn't have "UB".

But also, if a person talks about "UB", it is reasonable for a listener to assume they are by default talking about the well-defined term in the C programming language standard - unless they specify otherwise. Because that's the most enduring definition of the term we have.