Categorical Databases

Transparent Joins

Many data integration tasks require large numbers of joins. In addition to being a potential perforamnce bottleneck, large joins are difficult to write and debug; a single missing condition in a where clause can cause a query to return an incorrect result. In CQL, foreign keys can be followed (dereferenced) directly, so that many constuctions that require joins in SQL do not require joins in CQL.

This example (built in to the IDE with name Joinless) defines a source schema about schools, faculty, and departments, and a query to find everyone who works in a school whose largest department is mathematics. The query does not require any joins; in SQL, this query would require two joins.

We start by defining a source schema for schools, faculty, and departments, with foreign keys specifying the department and institute of each person, and the biggest department in a school:

typeside Ty = literal { 
	java_types
		String = "java.lang.String"
	java_constants
		String = "return input[0]"
}

schema Schools = literal : Ty {
	entities 
		Person
		School
		Dept
	foreign_keys
		instituteOf : Person -> School 
		deptOf      : Person -> Dept
		biggestDept : School -> Dept
	attributes
		lastName    : Person -> String
		schoolName  : School -> String
		deptName    : Dept   -> String
}

Here is some sample data, taken from the Boston area:

instance BostonSchools = literal : Schools {
	generators 
		ryan david adam greg gregory jason : Person
		harvard mit : School
		math cs : Dept
	multi_equations 
		lastName -> {ryan Wisnesky, david Spivak, adam Chlipala, greg Morrisett, 
					 gregory Malecha, jason Gross}
		schoolName -> {harvard Harvard, mit MIT}
		deptName -> {math Mathematics, cs CompSci}
		instituteOf -> {ryan harvard, david mit, adam mit, greg harvard, 
			            gregory harvard, jason mit}
		deptOf -> {ryan math, david math, adam cs, greg cs, gregory cs, jason cs}
		biggestDept -> {harvard math, mit cs}
}

Our goal is to find all people who work in a school whose biggest department is mathematics. The target schema contains an entity Person and two attributes:

schema Person = literal : Ty {
	entities 
		Person
	attributes
		lastName   : Person -> String
		schoolName : Person -> String
}

To populate this schema we write a query that iterates over all people, dereferencing foreign keys (using the dot operator) instead of performing joins:

query BiggestDeptIsMathQuery = literal : Schools -> Person {
	entities
		Person -> {	from p:Person
				 	where p.instituteOf.biggestDept.deptName = Mathematics
				 	return lastName -> p.lastName 
			  			   schoolName -> p.instituteOf.schoolName}
}

instance BiggestDeptIsMathInBoston = eval BiggestDeptIsMathQuery BostonSchools

The result is displayed in the IDE:

A screen shot of the entire development is shown below: