How to specify a template type will be unused in a Result type class?

[Kingdutch]

I'm trying to model a Result type class from functional programming because in a list of operations an Exception is very disruptive. Being able to indicate something was a success or a failure and using static analysis to force people to deal with both possibilities is very useful.

I'm able to get 90% of the way to a solution. However, the problem is that when constructing either the Ok or the Error variant of a Result I don't have a way to tell PHPStan "This template value is never used".

This is what I've been able to create https://phpstan.org/r/e9c7213e-b7ad-47f8-93b4-ec19e479b688

How would I indicate for ok and error that the ErrorT or OkT respectively won't be used? I'm not able to reduce the 2 template types to 1 because it's important that there's distinction between the type that happens on success and the type that happens on failure (and in some cases both might be string or int so I also can't use the type itself in calling code to distinguish success from failure).

If it helps from a conceptual standpoint: in other languages a Result is often an enum with values of Ok and Error both of which can have their own type contained, but in PHP we unfortunately don't have enum types that can have a value.

[ondrejmirtes]

There is more than one way how to model Result/Ok/Error. For example: https://phpstan.org/r/55b876e9-566c-4815-9d19-132cfbf84254

[Kingdutch]

Thanks for showing me an alternative implementation!

One thing I do notice is that the type you use on the extending Ok class is Result<T, never> which I would expect to be the same as the annotation for the ok function of my original implementation (self<T, never>).

Could you help me understand why my implementation gives the error

Method Drupal\Core\Result::ok() should return Drupal\Core\Result<T, never> but returns Drupal\Core\Result<T, T>.

Could you help me understand why PHPStan seemingly behaves differently for the function return type vs the class extends? I'd love to improve my mental model.

[ondrejmirtes]

The thing is how it interprets new self(TRUE, $value) - what type is inferred based on that. It simply does not understand the constructor signature as you intend it to:

   * @param bool $isOk
   *   TRUE if the result is OkT or FALSE otherwise.
   * @param ($isOk is true ? OkT : ErrorT) $value
   *   The value for the result.

See also #6732

[Kingdutch]

Oh that issue provides a lot of clarity! :D

Then another workaround if I really must have a single class rather than your (very elegant) 3-class solution is to insert a type annotation: https://phpstan.org/r/1f74947c-b350-4ae9-9162-67f23fd69265

That does move the solution a bit more into the "Trust me" realm since of course if the class ever changes without the @var annotation changing too, you'll have hidden type-errors.

Thanks!