Selda: A Haskell SQL Library

Chapter 3: Advanced Queries

While fetching rows from a single table is all well and good, most applications require a bit more interesting queries. This chapter will teach you how to correlate data between multiple tables, and to conjure up table rows from thin air.

Product Queries

Perhaps the most basic capability of a relational database is to correlate data between different tables, hence the relational part. In Selda, the monadic bind operator performs this function. In less theoretical terms, to combine data from multiple tables, simply use the select operation on each desired table. This mechanism should be familiar to you if you've ever used list comprehensions.

To illustrate the concept, this query returns a list of pairs of all people in our database and any homes they happen to own.

peopleWithHomes :: Query s (Row s Person :*: Row s Home)
peopleWithHomes = do
  person <- select people
  home <- select homes
  restrict (person ! #name .== home ! #ownerName)
  return (person :*: home)

The type signature of this query merits some explanation, as this is the first time we've seen one explicitly spelled out. As mentioned in chapter 1, all Selda queries are executed in the Query monad, which is parameterised over a scope type s.

All rows and columns originating from any given scope s are tagged with that s. This prevents queries in another scope from accessing those values, as all Selda operations ensure that the s of the current query and the s of the data being operated on are identical.

So far, we haven't encountered any queries that are convoluted enough for this to matter. Once we get to joins and aggregates however, the necessity of the scope parameter will become apparent.

Set Membership

The peopleWithHomes query does not include any persons who don't have a home. Not wanting to marginalize homeless people further, let's write a function that finds all such persons.

The set of homeless people can be defined as the set of people who do not appear in the set of home owners. This means that we can check whether any particular person is homeless using the isIn function, which determines whether some particular value is contained in some other result set.

homelessPeople :: Query s (Row s Person)
homelessPeople = do
  person <- select people
  restrict (not_ $ (person ! #name) `isIn` (#ownerName `from` select homes))
  return person

The set of home owners is produced by the query #ownerName `from` select homes. The from function is a convenient shorthand for extracting a single column from a query, and is defined as from s q = fmap (!s) q.

Ad Hoc Rows

Sometimes, it can be handy to conjure up database rows "from thin air". Let's say, for instance, that we want to create a Person row with only the information present in the homes table. We can accomplish this by using the new function.

personsFromHomes :: Query s (Row s Person)
personsFromHomes = do
  home <- select homes
  return $ new [#name := home ! #ownerName]

new takes a list of updates — as seen in chapter 2 — which are applied to a row where each value initially has its type's default value. Thus, the personsFromHomes function will return a list of Person values with their names set to something sensible, and the other fields set to 0 or Nothing.

Querying into Another Table

This ability to create entirely new rows can be convenient when you want to return some particular type from a query, but you don't want to get it straight from a table. One prime example of this is the queryInto function — corresponding to the SELECT INTO SQL statement — which inserts the result set of a query straight into another table and returns the number of inserted rows.

Using this function, we can remedy the homelessness situation by simply granting a cheap home in Tokyo to any homeless persons in the people table.

homesForEveryone :: SeldaM s Int
homesForEveryone = queryInto homes $ do
  person <- select people
  restrict (not_ $ (person ! #name) `isIn` (#ownerName `from` select homes))

  return $ new
    [ #ownerName := person ! #names
    , #city := "Tokyo"
    , #rent := 50000

Using the TypeApplications language extension, we can make the new expression even clearer, explicitly telling the compiler just what type of row we want to conjure up:

return $ new @Home [ ... ]

This is often useful — essential, even — when, for instance, creating an row entirely made up of default values, with no updates given to fix the type of the row.