Improvement to sentence about grouping and errors

[hartig]

This PR implements the change that @afs suggested in PR #132. See #132 (comment) for the original comment with this change suggestion.

I have to admit that, even with this change, I still don't understand what exactly the sentence is supposed to say. So, maybe we can improve the sentence even more. In particular, the following points still confuse me in this sentence.

1. In the context of talking about "the result of ListEval," the sentence mentions "solutions containing error values." Assuming "solutions" here means solution mappings (as everywhere else in the spec), this part of the sentence doesn't make sense. Solutions cannot contain "error values." By definition!
2. Related to the previous point, the result of ListEval is not a solution (solution mapping) but it is a list consisting of RDF terms and error values. So, perhaps that's what the word "solutions" was meant to refer to?
3. Is this sentence the only place in the spec that is related to the mentioned removal "at the end of evaluating the group and any aggregation functions"? Or is this removal somehow captured somewhere else in the spec as well (explicitly or implicitly)? I couldn't find anything and, thus, I still don't know what exactly is supposed to be removed. @afs do you have a concrete example that demonstrates the removal mentioned in the sentence?

---

Preview | Diff

[afs]

@afs do you have a concrete example that demonstrates the removal mentioned in the sentence?

IIRC

ListEval happens as grouping forms, collecting rows into the group-by-key.
GROUP (1/?x as ?y) may be an error.

count(*) is different in a way. It does not evaluate anything. It does count the "error group" and so a notion of error must be present at that point. "errors" aren't hidden after ListEval happens.

May be "projection time" is referring to the end of the group-aggregate operation. It is says "error tokens don't appear outside group-aggregate". While that occurs because of the SELECT line, it's not the project operation.

?y is unbound, ?B can be false, in

SELECT ?y (bound(?y) AS ?B)
....
GROUP (1/?x as ?y)