Returns are also gotos then.
But there is a good case to be made that exception handlers are gotos in the Dijkstra sense, at least when used for anything other than exceptions, like passing errors around.
> These retain structure and this are not considered harmful
This might be your opinion, and it is a very reasonable opionin. But it is just not what Dijkstra is arguing. He is very clearly arguing for “single entry single exit”.
Agreed. Return forces you into a single exit. Upon hitting return, the code can only return back to where the function was originally called. It cannot 'arbitrarily' jump to some other place in code as you could do in an unstructured programming language like, say, BASIC. Which is what Dijkstra was pushing for, being a strong proponent of structured programming.
I don't know of any modern programming language that does allow anything outside of a single exit, exception handlers and setjmp/longjmp excepted. There is a good case to be made that the latter two reintroduce the very problem Dijkstra warned of and are generally considered harmful for the same reason.
Exiting in the middle of a block would be just as bad as entering in the middle, according to the argument he is making.
However, Dijkstra accepts abortion clauses, which is what return really is (it is not an exit clause). My take is that his argument is that an unbridled go to is too primitive and that he believed go to statements should be bridled by additional structure that help describe the process, not that go to should be avoided entirely.
While I think we can agree that return is go to, it is a bridled go to. It strictly limits what a programmer can do, avoiding the mess Dijkstra claims an unbridled go to promotes. It is predictable and understandable, clearly describing the intent.
Exception handlers have no such strictness. It is not clear, without studying the program in its entirety, where your code will end up. That can be a good tradeoff when you are dealing with exceptions. The only reasonable response to encountering an exception is to ultimately crash, so at that point who cares? But, indeed, using exception handlers for control flow (e.g. passing errors around) is considered harmful.
It is not my view. I happen to disagree with Dijkstra on that point.