Unclear possibility to return non-Completion Record values from abstract operations

[arhadthedev]

ReturnIfAbrupt describes ReturnIfAbrupt(argument) as:

1. If argument is an abrupt completion, return argument.
2. Else if argument is a Completion Record, set argument to argument.[[Value]].

Step 2 assumes that argument may be something different from a Completion Record (either normal or abrupt). The same is with other declarations of ReturnIfAbrupt.

However, Implicit Completion Values guarantees that abstract operations always (with undefined edge cases) return a Completion Record:

Unless it is otherwise obvious from the context, an algorithm statement that returns a value that is not a Completion Record, such as:

1. Return "Infinity".

means the same thing as:

1. Return NormalCompletion("Infinity").

that makes the assumption probably excessive.

So I propose:

1. Either (if non-record values really may be returned) add references to cases covered by "Unless it is otherwise obvious" and replace obviousness with formal criteria;

2. Or (otherwise):

   • remove else if foo is a Completion Record, set foo to foo.[[Value]] steps in ReturnIfAbrupt;

   • rephrase Implicit Completion Values:

     The algorithms of this specification often implicitly return Completion Records whose [[Type]] is normal. Unless it is otherwise obvious from the context, an algorithm statement that returns a value that is not a Completion Record, such as:

     to:

     An algorithm statement that returns a value that is not a Completion Record, such as:

[bakkot]

The handling of completion records is inconsistent in a number of ways; see this label in the issue tracker. For example, the implicit coercion you quote doesn't actually make sense: abstract operations can return values which are neither ECMAScript language values nor ~empty~, but a normal completion can only hold values of those types.

I intend to deal with this more comprehensively when I have time, including eliminating all implicit wrapping and unwrapping.

[arhadthedev]

Thank you, I really missed specification and host types (with examples found in #496). Also, I am happy to hear about planned specification-wide revamping of such incoherencies; I also know that it will require a long time to not break anything and properly update external specifications that refer to ECMAScript abstract operations.