Hacker News Reader

Comment by Avshalom | Hacker News Reader

Cosmic rays are no excuse not to throw an exception on integer overflow.

If, for some inputs, the program is supposed to calculate the value 42, but it throws an integer overflow, then it is not correct.

Detecting overflows and throwing nice exceptions instead of continuing execution with garbage values isn't the same thing as ensuring correctness.

Is it really the case that some programming languages are more resilient to hardware errors than others?

Java does array bounds checking in principle, but it is usually optimized out of the machine code. Isn't it then virtually just as susceptible to hardware error as C? And the JVM is written in C anyway.

This has been the subject of some papers, like

https://www.cs.princeton.edu/~appel/papers/memerr.pdf [2003]

The JVM is written in C, but that C has been subject to some formal analysis. And even if it hadn't, I'd bet the defect rate of a Java program running on a C JVM would be much much lower than that of a C program.

But, in keeping with this topic, that C has not been subject to formal analysis from the angle of defending against rays doing random things to the CPU and RAM or other hardware errors.

So let's stop unit testing all of our code, since it's provably incorrect anyways, due to cosmic rays flipping random bits of RAM once in every gazillion calculations.

> And the JVM is written in C anyway.

Which one?

https://en.wikipedia.org/wiki/List_of_Java_virtual_machines

Depending which one we are speaking about, they are implemented in Assembly, C, C++ or even Java.

Reply on news.ycombinator.com