[hesselink]

The thing is, you're focusing on when you've detected that there is an issue (a crash). A lot of the issues with NULL are the fact that you can't easily detect if beforehand. It's not indicated in the types, or the syntax. That means that it's incredibly easy for a NULL issue to sneak into an uncommon branch or scenario, only to be hit in production.

[nsajko]

But why was the scenario not tested before production? Should that not be the case anyway?

[cultus]

You can never guarantee you really have 100% test coverage in all scenarios in complex software.

[pjc50]

Indeed. It gets asymptotically more expensive. Whereas a typechecker is a system of tests that's able to "cover" 100% of the code.

[empath75]

It’s cute that you think tests find all problems.