Keep both sequences and multisets of solution mappings as algebraic expressions?

[hartig]

The algebraic expressions produced by the translation in Section 18.2 Translation to the SPARQL Algebra may contain, as a sub-expression, a multiset of solution mappings or a sequence of solution mappings. In particular,

• a multiset of solution mappings becomes a sub-expression if the WHERE clause of the given query contains a VALUES clause: see the point "If the form is InlineData" in Section 18.2.2.6 Translate Graph Patterns,
• a sequence of solution mappings becomes a sub-expression if the given query contains a trailing VALUES clause: see Section 18.2.4.3 VALUES.

A question is whether we really need both. Probably not, because in the case of trailing VALUES clause (i.e., Section 18.2.4.3 VALUES), the sequence is directly passed into a 'ToMultiSet' operator! Yet, related to this question is also issue #215 which highlights that we may actually want to keep the sequence case.

Also, perhaps a matter of taste, but I don't find it particularly elegant to have sequences and/or multisets of solution mappings directly as a possible sub-expression within the algebraic syntax. It would feel more natural to me to have a dedicated operator as a wrapper for such a sequence or multiset; such an operator may, for instance, be called 'InlineSolutions' or just 'Solutions' or 'Table'.

[afs]

The key is sequences retain order, multisets do not.
sequences only occur in modifiers.

[TallTed]

The key is sequences retain order, multisets do not.

That seems to me a strong argument to retain both, and if not both, then sequences.

[afs]

I can see @hartig's point about clutter - the main concept is multisets.
If there could be only one, then it must be possibility order retaining.
However, saying "sequences without significant order" repeatedly then deserves a name.

[hartig]

Perhaps I should emphasize that this issue here is not about whether the spec should generally distinguish between multiset of solution mappings and sequences of solution mappings. It certainly should! In the parts where it assumes a sequence, the spec should say that it is a sequence, and vice versa for multisets.

No, this issue is only about the kinds of expressions that are possible in the algebraic syntax; i.e., the syntax into which ASTs are converted by the translation in Section 18.2 Translation to the SPARQL Algebra. Currently, this syntax needs to cover both options; that is, a multiset of solution mappings must be allowed as an expression in this syntax and a sequence of solution mappings must be allowed as well. The two bullet points in the issue description above point to the different parts of the translation algorithm where this expressive power of the algebraic syntax is used.

[hartig]

I don't think this issue should be closed. I am still planning to come back to it, but this is better done after the work on #228 is finished.