Scalar Subqueries and List Subqueries

[thobe]

Solves #177.

CIP2017-03-29

[thobe]

There is still some work to be done here, but comments are welcome.

Ideas for examples to include in the document:

• Sorting lists
• Filtering lists
• Unpacking a singleton list

[thobe]

These fish-brackets are a place holder, we really need something better to delineate a scalar subquery. Suggestions are most welcome. It could be a keyword, I considered 'SINGLE', '(', SingleValueSubquery, ')' ;, but single is already used as a predicate function (actually also a type of subquery) in Cypher.

Another suggestion was 'SCALAR', '(', SingleValueSubquery, ')' ;, but that didn't feel quite right.

[boggle]

More random ideas:

• VALUE ( subquery ) (but: too generic?)
• QUERY ( subquery ) (but: too generic?
• EVAL ( subquery ) (but: sounds too much like metaprogramming?)
• $( subquery ) (but: feels like we should reserve this syntax, perhaps for metaprogramming/templating)
• (> ()-[]->(c) RETURN c <) - Kinda like these "injection (or inception?) parens"
• (| ()-[]->(c) RETURN c |) - Bananas ftw :)

[boggle]

Maybe the whole approach is also wrong.. for crpq we're discussing path pattern predicates.. maybe here we're missing inline function or expression-level template definitions? Given those, the quest for scalar subqueries really just becomes a search for anonymous function syntax.

DECLARE  $subquery(args)=expr
...
WITH expr1 + $subquery(...)
...

[boggle]

One more... if we don't allow subquery short forms, i.e. starting with a pattern, round parenthesis might actually suffice from a parsing point of view... (MATCH (a)-[r]->(b) RETURN r). It's inconsistent with the other subquery forms under discussion but it would save us the hassle of coming up with weird fancy bracket syntax.

[petraselmer]

• 'SCALAR' as a keyword is simple and "does what it says on the tin"; i.e. it's very clear in its intention. What are the objections to its use?

• $( subquery ) is rather elegant. Re boggle's comments about the drawback, would it not be reasonable to infer from context?

• The inline function/expression-level template defs option looks very unwieldy from a readability point of view and does not marry up at all well with our other Subquery Family syntax. I don't see any advantage in it.

[thobe]

Yes, I've had that thought as well...

if we don't allow subquery short forms /.../ round parenthesis might actually suffice from a parsing point of view

[thobe]

I don't remember what the objections were to SCALAR - perhaps we never suggested it in prior conversation? I know that we suggested VALUE and decided that it was definitely too generic.

I'm leaning towards SCALAR now, since I agree with Stefan that since we've used $ for parameters, we should reserve variations with that character for variations of parametrization.

[Mats-SX]

This sentence sounds a bit strange to me. Perhaps

The goal of a List Subquery

or

The result of a List Subquery is the list formed by collecting all of the values of all rows produced by the subquery.

[Mats-SX]

Cypher style: space before property predicate.
Cypher style: space after colon in property predicate.

[Mats-SX]

Would the UNWIND form not require a RETURN clause?

[thobe]

The implication here is that the implicitly RETURN value is the value unpacked by UNWIND.

UNWIND with RETURN is handled by the regular SingleValueReadSubquery.

The use cases for this form is:

1. Unpacking a singleton-list: SCALAR( UNWIND myList )
2. Sorting a list: [UNWIND myList AS x ORDER BY x.age]
3. Filtering a list: [UNWIND myList AS x WHERE x.name CONTAINS ' ']

et.c.

[boggle]

Just realized... using round parenthesis for scalar subqueries seems not well aligned with EXISTS {} which also returns a single value... in a way.. Not sure how to solve this given that we overloaded exists(n.prop)