yeshql-3.0.0.0: YesQL-style SQL database abstraction

Copyright(c) 2015 Tobias Dammers
MaintainerTobias Dammers <tdammers@gmail.com>
Stabilityexperimental
Safe HaskellNone
LanguageHaskell2010

Database.YeshQL

Contents

Description

License: MIT

Unlike existing libraries such as Esqueleto or Persistent, YeshQL does not try to provide full SQL abstraction with added type safety; instead, it gives you some simple tools to write the SQL queries yourself and bind them to (typed) functions.

Usage

The main workhorses are yesh1 (to define one query) and yesh (to define multiple queries).

Both yesh and yesh1 can be used as TemplateHaskell functions directly, or as quasi-quoters, and they can generate declarations or expressions depending on the context in which they are used.

Creating Declarations

When used at the top level, or inside a where block, the yesh and yesh1 quasi-quoters will declare one or more functions, according to the query names given in the query definition. Example:

[yesh1|
    -- name:insertUser :: (Integer)
    -- :name :: String
    INSERT INTO users (name) VALUES (:name) RETURNING id |]

...will create a top-level function of type:

    insertUser :: IConnection conn => conn -> String -> IO (Maybe Integer)

Using plain TH, it can also be written as:

yesh1 $ unlines
    [ "-- name:insertUser :: (Integer)"
    , "-- :name :: String"
    , "INSERT INTO users (name) VALUES (:name) RETURNING id"
    ]

Syntax

Because SQL itself does not *quite* provide enough information to generate a fully typed Haskell function, we extend SQL syntax a bit.

Here's what a typical YeshQL definition looks like:

[yesh|
    -- name:insertUser :: (Integer)
    -- :name :: String
    INSERT INTO users (name) VALUES (:name) RETURNING id
    ;;;
    -- name:deleteUser :: rowcount Integer
    -- :id :: Integer
    DELETE FROM users WHERE id = :id
    ;;;
    -- name:getUser :: (Integer, String)
    -- :id :: Integer
    SELECT id, name FROM users WHERE id = :id
    ;;;
    -- name:getUserEx :: [(Integer, String)]
    -- :id :: Integer
    -- :filename :: String
    SELECT id, name FROM users WHERE name = :filename OR id = :id
    |]

Note that queries are separated by _triple_ semicolons; this is done in order to allow semicolons to appear inside queries.

On top of standard SQL syntax, YeshQL query definitions are preceded by some extra information in order to generate well-typed HDBC queries. All that information is written in SQL line comments (-- ...), such that a valid YeshQL definition is also valid SQL by itself (with the exception of parameters, which follow the pattern :paramName).

Let's break it down:

    -- name:insertUser :: (Integer)

This line tells YeshQL to generate an object called insertUser, which should be a function of type IConnection conn => conn -> {...} -> IO (Maybe Integer) (where the {...} part depends on query parameters, see below).

The declared return type can be one of the following:

  • (); the generated function will ignore any and all results from the query and always return ().
  • The keyword rowcount, followed by an integer scalar, e.g. Integer or Int; the generated function will return a row count from INSERT UPDATE ... statements, or 0 from SELECT statements.
  • A tuple, where all elements implement FromSql; the function will return the result set from a SELECT query as a Maybe of such tuples, or always Nothing for other query types. For example, :: (String, Int) produces a function whose type ends in conn -> IO (Maybe (String, Int)). Null-tuples are marshalled to '()', ignoring result sets; one-tuples (written as (a)) are marshalled to scalars.
  • A naked type, i.e., just a type name, without parentheses. The type must implement FromSqlRow; the return type will be a Maybe of that type. E.g., (:: User) will produce a function signature ending in conn -> IO (Maybe User).
  • A list of tuples or naked types, written using square brackets ([ ... ]), returning a list of mapped rows instead of a Maybe.

Note that, unlike Haskell, YeshQL distinguishes (Foo) from Foo: the former takes the first column from a result row and maps it using FromSql, while the latter takes the entire result row and maps it using FromSqlRow.

    -- :paramName :: Type

Declares a Haskell type for a parameter. The parameter :paramName can then be referenced zero or more times in the query itself, and will appear in the generated function signature in the order of declaration. So in the above example, the last query definition:

    -- name:getUserEx :: (Integer, String)
    -- :id :: Integer
    -- :filename :: String
    SELECT id, name FROM users WHERE name = :filename OR id = :id;

...will produce the function:

getUserEx :: IConnection conn => conn -> Integer -> String -> IO [(Integer, String)]
getUserEx conn id filename =
    -- ... generated implementation left out

On top of referencing parameters directly, you can also "drill down" with a projection function, using . syntax similar to property access in, say, JavaScript. The intended use case is passing record types as arguments to the query function, and then dereferencing them inside the query, like so:

    -- name:updateUser :: rowcount Int
    -- :user :: User
    UPDATE users
    SET username  = :user.name
    WHERE id = :user.userID

Note that the part after the . is a plain Haskell function that must be in scope wherever the query is spliced.

Also note that projection functions can be chained, and are not limited to record field accessors.

Loading Queries From External Files

The yeshFile and yesh1File flavors take a file name instead of SQL definition strings. Using these, you can put your SQL in external files rather than clutter your code with long quasi-quotation blocks.

DDL Queries

Normally, you will have one query per function, and that query takes some parameters, and returns a result set or a row count. For DDL queries, however, we aren't interested in the results, and we often want to execute multiple queries with just one function call, e.g. to set up an entire database schema using multiple CREATE TABLE statements.

By adding the @ddl annotation to a query definition, YeshQL will change the following things:

  • The return type of that query, regardless of what you declare, will be '()'. It is recommended to never declare an explicit return type other than '()' for DDL queries, as future versions may report this as an error.
  • The query cannot accept any parameters.
  • The query may consist of multiple individual SQL queries, semicolon-separated. The combined query is sent to the HDBC backend as-is.
  • Instead of run, YeshQL will use runRaw in the code it generates.

In practice, this means that the type of a DDL function thus generated will always be IConnection conn => conn -> IO ().

Example:

[yesh1|
    -- name:makeDatabaseSchema
    -- @ddl
    CREATE TABLE users (id INTEGER, username TEXT);
    CREATE TABLE pages (id INTEGER, title TEXT, slug TEXT, body TEXT);
    |]

Other Functions That YeshQL Generates

On top of the obvious query functions, a top-level YeshQL quasiquotation introduces two more definitions per query: a String variable prefixed describe-, which contains the SQL query as passed to HDBC (similar to the DESCRIBE feature in some RDBMS systems), and another String variable prefixed doc-, which contains all the free-form comments that precede the SQL query in the query definition.

So for example, this quasiquotation:

[yesh1|
    -- name:getUser :: User
    -- :userID :: Integer
    -- Gets one user by the "id" column.
    SELECT id, username FROM users WHERE id = :userID LIMIT 1 |]

...would produce the following three top-level definitions:

getUser :: IConnection conn => Integer -> conn -> IO (Maybe User)
getUser userID conn = ...

describeGetUser :: String
describeGetUser = "SELECT id, username FROM users WHERE id = ? LIMIT 1"

docGetUser :: String
docGetUser = "Gets one user by the \"id\" column."

Synopsis

Quasi-quoters that take strings

yesh :: Yesh a => a Source #

Generate top-level declarations or expressions for several SQL queries. If used at the top level (i.e., generating declarations), all queries in the definitions must be named, and yesh will generate a separate set of functions for each. If used in an expression context, the current behavior is somewhat undesirable, namely sequencing the queries using >>.

Future versions will most likely change this to create a tuple of query expressions instead, such that you can write something like:

let (createUser, getUser, updateUser, deleteUser) = [yesh|
     -- name:createUser :: (Integer)
     -- :username :: String
     INSERT INTO users (username) VALUES (:username) RETURNING id;
     -- name:getUser :: (Integer, String)
     -- :userID :: Integer
     SELECT id, username FROM users WHERE id = :userID;
     -- name:updateUser :: Integer
     -- :userID :: Integer
     -- :username :: String
     UPDATE users SET username = :username WHERE id = :userID;
     -- name:deleteUser :: Integer
     -- :userID :: Integer
     DELETE FROM users WHERE id = :userID LIMIT 1;
 |]

yesh1 :: Yesh a => a Source #

Generate a top-level declaration or an expression for a single SQL query. If used at the top level (i.e., generating a declaration), the query definition must specify a query name.

Quasi-quoters that take filenames

yeshFile :: YeshFile a => a Source #

Generate multiple query definitions or expressions from an external file. Query name derivation works exactly like for yesh1File, except that an underscore and a 0-based query index are appended to disambiguate queries from the same file.

In an expression context, the same caveats apply as for yesh, i.e., to generate expressions, you will almost certainly want yesh1File, not yeshFile.

yesh1File :: YeshFile a => a Source #

Generate one query definition or expression from an external file. In a declaration context, the query name will be derived from the filename unless the query contains an explicit name. Query name derivation works as follows:

  • Take only the basename (stripping off the directories and extension)
  • Remove all non-alphabetic characters from the beginning of the name
  • Remove all non-alphanumeric characters from the name
  • Lower-case the first character.

Note that since there is always a filename to derive the query name from, explicitly defining a query name is only necessary when you want it to differ from the filename; however, making it explicit anyway is probably a good idea.

Low-level generators in the Q monad

Query parsers

AST

data ParsedQuery Source #

Constructors

ParsedQuery 

Fields