Completion record linting doesn't work quite right with parentheses

[ptomato]

In Temporal I have an operation RoundDuration which returns "either a normal completion containing a Record with fields [[DurationRecord]] (a Duration Record) and [[Total]] (a mathematical value), or a throw completion". I'm trying to access one of the fields on the returned Record without making an intermediate binding:

1. Let result be (? RoundDuration(arguments)).[[DurationRecord]].

This results in the error "RoundDuration returns a Completion Record, but is not consumed as if it does" from ecmarkup.

Changing it to this makes the error go away:

1. Let roundRecord be ? RoundDuration(arguments).
2. Set result to roundRecord.[[DurationRecord]].

[ljharb]

Why are the parens needed at all in the first case? The question mark should attach to the AO call, not the field value.

[bakkot]

The question mark should attach to the AO call, not the field value.

... Why? That's not the obvious-to-me parse of that expression. (Compare e.g. ! a().b in JS.)

[jmdyck]

There are two examples in the spec where the operand of ? is a field-extraction, both in InnerModuleEvaluation:

• return ? _module_.[[EvaluationError]]
• return ? _requiredModule_.[[EvaluationError]]

In each case, the intended parse is
return ? (_foo_.[[EvaluationError]])
not
return (? _foo_).[[EvaluationError]]

[bakkot]

I think I can fix this easily enough, but to be honest the extra binding seems like a good thing from a readability point of view, so I'm probably not going to prioritize it right away.

[ljharb]

@bakkot because i don’t expect completion records to be stored in record fields, but i do expect them to be the result of an AO call. I don’t think JS has a comparison because completion records are so weird a concept.

[bakkot]

Completion records can be stored in record fields. They're just a value like any other.

[ljharb]

Sure, that's just not something we ever do, afaik, hence my expectation.

[jmdyck]

I already gave an example where we do that: the [[EvaluationError]] field of a Cyclic Module Record is a throw completion or ~empty~.

[ljharb]

Ah, fair enough. My expectation still holds, and using parens should ofc work, and i agree the separate alias is cleaner.