[Issue 24030] New: A `lazy` parameter shouldn't be allowed to be "called" twice

[d-bugmail at puremagic.com]

https://issues.dlang.org/show_bug.cgi?id=24030

          Issue ID: 24030
           Summary: A `lazy` parameter shouldn't be allowed to be "called"
                    twice
           Product: D
           Version: D2
          Hardware: All
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P1
         Component: dmd
          Assignee: nobody at puremagic.com
          Reporter: qs.il.paperinik at gmail.com

It’s quite straightforward: The intended use for `lazy` is to defer evaluation;
everyone seeing this for the first time expects that. The fact that a function
taking a `lazy` parameter may evaluate the underlying delegate more than once
is surprising. Only once a programmer learns that there is a delegate
underlying, it makes some sense.

This is a breaking change, but there’s a simple transition path: Use a delegate
explicitly, and you’ll surprise nobody.

Checking that a `lazy` parameter is evaluated at most once only needs a very
limited form of control-flow analysis. In `@system` code, this could be ignored
and calling a `lazy` parameter more than once becomes Undefined Behavior.

If desired, on non-release builds, the only-call-once rule can be asserted by
introducing a hidden `bool` variable on the caller’s side:

```d
int f(lazy int);

1 + f(1);
```

is lowered to:

```d
int f(int delegate());

bool called; // unique name
1 + f({
    assert(!called);
    called = true;
    return 1;
});
```

What about `lazy` `void` parameters?

They’re no different. You’re after the side-effect, but again, getting it more
than once will be surprising; use a delegate.

--