Even better: just let it crash an get a core dump with full context information rather than some log missing information.

But often some "expected" errors can be handled in some way better (retry, ask user, use alternate approach, ...)