Irrelevant transient dependencies reported

[lstrojny]

class VO {}

abstract class A {
    final public function getVO(): VO { return new VO($this->generate()); }
    abstract protected function generate(): string;
}

class B extends A {
    protected function generate(): string { return "from B"; }
}

Now assume each class lives in its own layer called like the class wit this ruleset:

VO: []
A: [VO]
B: [A]

deptrac will complain that B depends on VO even though I would say it does not. Bug or feature?

[patrickkusebauch]

This is absolutely a feature. Looks what happens when you "squash" B into the final shape the class will have at runtime:

class B {
    final public function getVO(): VO { return new VO($this->generate()); }
    protected function generate(): string { return "from B"; }
}

You absolutely depend on VO. You can look at it another way:

If I deleted VO, would B lint?

Obviously, it would not, it would complain about non-existent VO.

Deptrac should provide you with a relatively nice message about this being a dependency coming from A via inheritance as well. At least with the default console formatter.

[patrickkusebauch]

Thinking about it a bit more.

Is the argument there that since final public function getVO(): VO { return new VO($this->generate()); } is final it is always executed in the context of A and therefore not B should not be dependent on VO? And if the method was not final it would be?

That would be an interesting case.

If that would be the case what would happen if I had:

class C {
  public function foo(): void {
    $var = new B(); //This is clear dependency on B
    $vo = $var->getVO(); //Is this dependency on B or on A?
    var_dump($vo); //This could be a dependency on VO, but it is not yet implemented in deptrac
  }
}

It creates another conundrum. I am not opposed to this, just wondering what were your expectations. It is important that deptrac behaves the way people expect it do behave.

[lstrojny]

Hah, interesting indeed. I love the "if you delete it, will it compile" heuristic, since it’s super easy to grasp. That being said, the final case is not atypical in a ports-and-adapters kinda situation.
I guess this needs more long showers and long walks before I can say something more insightful here. But I will come back.

[patrickkusebauch]

I will close this for now due to inactivity and lack of a clear ask in this issue. Feel free to re-open if you need.