CALL executes operations against the query result of each subquery, NOT all the result of a query command. It wraps query commands and the operations in curly braces {}, passes in the alias that triggers each subquery using a WITH clause and passes out the result of each subquery using a RETURN clause.

Syntax:
CALL {
  with <alias_In>, <alias_In>, ...
  ...
  return <expression> as <alias_Out>, <expression> as <alias_Out>, ...
} Input:

• <alias_In>: The alias to pass into the CALL clause
• <expression>: The return value of the subquery
• <alias_Out>: The alias of return value of the subquery to pass out of the CALL clause, optional when <expression> is alias

find().nodes([1, 5]) as nodes
call {
  with nodes
  n(nodes).e()[:2].n() as p
  skip 1
  return p as path
}
return path

In the UQL above, there are 3 p resulted from the 1st subquery and 2 p resulted from the 2nd subquery, one row is skipped for each of these two results:

Single Query in Subquery

Example: do not use GROUP BY, find how many cards Customer CU001, CU002, and CU003 hold respectively

uncollect ["CU001", "CU002", "CU003"] as user
call { 
  with user
  n({_id == user}).e({@has}).n({@card} as n)
  return count(n) as number
}
return user, number

Analysis: CALL aggregates each user's n to get number, i.e. group all users n based on user and aggregate all users n in group.

Multiple Queries in Subquery

Example: find 10 transfer edges with an amount higher than 8,000 and rate each edge: if there is transfer between its end-node and bank card CA029 within 2 steps, then add 0.5 for that end-node; the sum of ratings of both end-nodes is the final rating of the edge

n({@card} as n1).re({@transfer.amount > 8000} as e).n({@card} as n2).limit(10)
call {
  with n1, n2
  optional n(n1 as a1).e({@transfer})[:2].n({_id == "CA029"}).limit(1) 
  with CASE a1 when 0 then 0 else 0.5 END as c1
  optional n(n2 as a2).e({@transfer})[:2].n({_id == "CA029"}).limit(1) 
  with CASE a2 when 0 then 0 else 0.5 END as c2
  return c1+c2 as c
}
return e{*}, c

Analysis: n1 and n2 are the startting-node and ending-node of edge e, CALL calculates ratings c1 and c2 for n1 and n2 in each row, and sum them as the total rating c for each e; the length of c is the same as the length of n1, n2 and c.