Honey SQL

SQL as Clojure data structures. Build queries programmatically -- even at runtime -- without having to bash strings together.

Build

The latest stable version (1.0.444) on Clojars and on cljdoc (note: honeysql/honeysql):

This project follows the version scheme MAJOR.MINOR.COMMITS where MAJOR and MINOR provide some relative indication of the size of the change, but do not follow semantic versioning. In general, all changes endeavor to be non-breaking (by moving to new names rather than by breaking existing names). COMMITS is an ever-increasing counter of commits since the beginning of this repository.

Prerelease builds of the upcoming 2.x version of HoneySQL will soon be available:

This is the README for the upcoming 2.x version of HoneySQL which provides a streamlined codebase and a simpler method for extending the DSL. It also supports SQL dialects out-of-the-box and will be extended to support vendor-specific language features over time (unlike the 1.x version).

See this summary of differences between v1 and v2 if you want to help test v2!

Note on code samples

All sample code in this README is automatically run as a unit test using seancorfield/readme.

Some of these samples show pretty-printed SQL: HoneySQL 2.x supports :pretty true which inserts newlines between clauses in the generated SQL strings.

Usage

(refer-clojure :exclude '[for group-by set update])
(require '[honey.sql :as sql]
         ;; caution: this overwrites for, group-by, set, and update
         '[honey.sql.helpers :refer :all :as h])

Everything is built on top of maps representing SQL queries:

(def sqlmap {:select [:a :b :c]
             :from   [:foo]
             :where  [:= :f.a "baz"]})

Column names can be provided as keywords or symbols (but not strings -- HoneySQL treats strings as values that should be lifted out of the SQL as parameters).

format

format turns maps into next.jdbc-compatible (and clojure.java.jdbc-compatible), parameterized SQL:

(sql/format sqlmap)
=> ["SELECT a, b, c FROM foo WHERE f.a = ?" "baz"]

HoneySQL is a relatively "pure" library, it does not manage your sql connection or run queries for you, it simply generates SQL strings. You can then pass them to a JDBC library, such as next.jdbc:

(jdbc/execute! conn (sql/format sqlmap))

If you want to format the query as a string with no parameters (e.g. to use the SQL statement in a SQL console), pass :inline true as an option to sql/format:

(sql/format sqlmap {:inline true})
=> ["SELECT a, b, c FROM foo WHERE f.a = 'baz'"]

Note: you'll need to add your preferred JDBC library as a dependency in your project -- HoneySQL deliberately does not make that choice for you.

Namespace-qualified keywords are generally treated as table-qualified columns: :foo/bar becomes foo.bar, except in contexts where that would be illegal (such as the list of columns in an INSERT statement). This approach is likely to be more compatible with code that uses libraries like next.jdbc and seql, as well as being more convenient in a world of namespace-qualified keywords, following the example of clojure.spec etc.

[In HoneySQL 1.x, this was the behavior when :namespace-as-table? true was specified]

(def q-sqlmap {:select [:foo/a :foo/b :foo/c]
               :from   [:foo]
               :where  [:= :foo/a "baz"]})
(sql/format q-sqlmap)
=> ["SELECT foo.a, foo.b, foo.c FROM foo WHERE foo.a = ?" "baz"]

Vanilla SQL clause helpers

There are also functions for each clause type in the honey.sql.helpers namespace:

(-> (select :a :b :c)
    (from :foo)
    (where [:= :f.a "baz"]))

Order doesn't matter (for independent clauses):

(= (-> (select :*) (from :foo))
   (-> (from :foo) (select :*)))
=> true

When using the vanilla helper functions, repeated clauses will be merged into existing clauses, in the natural evaluation order (where that makes sense):

(-> sqlmap (select :d))
=> {:from [:foo], :where [:= :f.a "baz"], :select [:a :b :c :d]}

If you want to replace a clause, you can dissoc the existing clause first, since this is all data:

(-> sqlmap
    (dissoc :select)
    (select :*)
    (where [:> :b 10])
    sql/format)
=> ["SELECT * FROM foo WHERE (f.a = ?) AND (b > ?)" "baz" 10]

where will combine multiple clauses together using SQL's AND:

(-> (select :*)
    (from :foo)
    (where [:= :a 1] [:< :b 100])
    sql/format)
=> ["SELECT * FROM foo WHERE (a = ?) AND (b < ?)" 1 100]

Column and table names may be aliased by using a vector pair of the original name and the desired alias:

(-> (select :a [:b :bar] :c [:d :x])
    (from [:foo :quux])
    (where [:= :quux.a 1] [:< :bar 100])
    sql/format)
=> ["SELECT a, b AS bar, c, d AS x FROM foo AS quux WHERE (quux.a = ?) AND (bar < ?)" 1 100]

In particular, note that (select [:a :b]) means SELECT a AS b rather than SELECT a, b -- helpers like select are generally variadic and do not take a collection of column names.

Inserts

Inserts are supported in two patterns. In the first pattern, you must explicitly specify the columns to insert, then provide a collection of rows, each a collection of column values:

(-> (insert-into :properties)
    (columns :name :surname :age)
    (values
     [["Jon" "Smith" 34]
      ["Andrew" "Cooper" 12]
      ["Jane" "Daniels" 56]])
    (sql/format {:pretty true}))
=> ["
INSERT INTO properties
(name, surname, age)
VALUES (?, ?, ?), (?, ?, ?), (?, ?, ?)
"
"Jon" "Smith" 34 "Andrew" "Cooper" 12 "Jane" "Daniels" 56]

If the rows are of unequal lengths, they will be padded with NULL values to make them consistent.

Alternately, you can simply specify the values as maps:

(-> (insert-into :properties)
    (values [{:name "John" :surname "Smith" :age 34}
             {:name "Andrew" :surname "Cooper" :age 12}
             {:name "Jane" :surname "Daniels" :age 56}])
    (sql/format {:pretty true}))
=> ["
INSERT INTO properties
(name, surname, age) VALUES (?, ?, ?), (?, ?, ?), (?, ?, ?)
"
"John" "Smith" 34
"Andrew" "Cooper"  12
"Jane" "Daniels" 56]

The set of columns used in the insert will be the union of all column names from all the hash maps: columns that are missing from any rows will have NULL as their value.

Nested subqueries

The column values do not have to be literals, they can be nested queries:

(let [user-id 12345
      role-name "user"]
  (-> (insert-into :user_profile_to_role)
      (values [{:user_profile_id user-id
                :role_id         (-> (select :id)
                                     (from :role)
                                     (where [:= :name role-name]))}])
      (sql/format {:pretty true})))

=> ["
INSERT INTO user_profile_to_role
(user_profile_id, role_id) VALUES (?, (SELECT id FROM role WHERE name = ?))
"
12345
"user"]

(-> (select :*)
    (from :foo)
    (where [:in :foo.a (-> (select :a) (from :bar))])
    sql/format)
=> ["SELECT * FROM foo WHERE foo.a IN (SELECT a FROM bar)"]

Composite types

Composite types are supported:

(-> (insert-into :comp_table)
    (columns :name :comp_column)
    (values
     [["small" (composite 1 "inch")]
      ["large" (composite 10 "feet")]])
    (sql/format {:pretty true}))
=> ["
INSERT INTO comp_table
(name, comp_column)
VALUES (?, (?, ?)), (?, (?, ?))
"
"small" 1 "inch" "large" 10 "feet"]

Updates

Updates are possible too:

(-> (h/update :films)
    (set {:kind "dramatic"
           :watched [:+ :watched 1]})
    (where [:= :kind "drama"])
    (sql/format {:pretty true}))
=> ["
UPDATE films
SET kind = ?, watched = watched + ?
WHERE kind = ?
"
"dramatic"
1
"drama"]

If you are trying to build a compound update statement (with from or join), be aware that different databases have slightly different syntax in terms of where SET should appear. The default above is to put SET before FROM which is how PostgreSQL (and other ANSI-SQL dialects work). If you are using MySQL, you will need to select the :mysql dialect in order to put the SET after any JOIN clause.

Deletes

Deletes look as you would expect:

(-> (delete-from :films)
    (where [:<> :kind "musical"])
    (sql/format))
=> ["DELETE FROM films WHERE kind <> ?" "musical"]

If your database supports it, you can also delete from multiple tables:

(-> (delete [:films :directors])
    (from :films)
    (join :directors [:= :films.director_id :directors.id])
    (where [:<> :kind "musical"])
    (sql/format {:pretty true}))
=> ["
DELETE films, directors
FROM films
INNER JOIN directors ON films.director_id = directors.id
WHERE kind <> ?
"
"musical"]

If you want to delete everything from a table, you can use truncate:

(-> (truncate :films)
    (sql/format))
=> ["TRUNCATE films"]

Set operations

Queries may be combined with a :union, :union-all, :intersect or :except keyword:

(sql/format {:union [(-> (select :*) (from :foo))
                     (-> (select :*) (from :bar))]})
=> ["(SELECT * FROM foo) UNION (SELECT * FROM bar)"]

There are also helpers for each of those:

(sql/format (union (-> (select :*) (from :foo))
                   (-> (select :*) (from :bar))))
=> ["(SELECT * FROM foo) UNION (SELECT * FROM bar)"]

Functions

Keywords that begin with % are interpreted as SQL function calls:

(-> (select :%count.*) (from :foo) sql/format)
=> ["SELECT COUNT(*) FROM foo"]

(-> (select :%max.id) (from :foo) sql/format)
=> ["SELECT MAX(id) FROM foo"]

Since regular function calls are indicated with vectors and so are aliased pairs, this shorthand can be more convenient due to the extra wrapping needed for the regular function calls in a select:

(-> (select [[:count :*]]) (from :foo) sql/format)
=> ["SELECT COUNT(*) FROM foo"]

(-> (select [[:max :id]]) (from :foo) sql/format)
=> ["SELECT MAX(id) FROM foo"]

Bindable parameters

Keywords that begin with ? are interpreted as bindable parameters:

(-> (select :id)
    (from :foo)
    (where [:= :a :?baz])
    (sql/format {:params {:baz "BAZ"}}))
=> ["SELECT id FROM foo WHERE a = ?" "BAZ"]

Miscellaneous

Sometimes you want to provide SQL fragments directly or have certain values placed into the SQL string rather than turned into a parameter.

The :raw syntax lets you embed SQL fragments directly into a HoneySQL expression. It accepts either a single string to embed or a vector of expressions that will be converted to strings and embedded as a single string.

The :inline syntax attempts to turn a Clojure value into a SQL value and then embeds that string, e.g., [:inline "foo"] produces 'foo' (a SQL string).

The :param syntax identifies a named parameter whose value will be supplied via the :params argument to format.

The :lift syntax will prevent interpretation of Clojure data structures as part of the DSL and instead turn such values into parameters (useful when you want to pass a vector or a hash map directly as a positional parameter value, for example when you have extended next.jdbc's SettableParameter protocol to a data structure).

Finally, the :nest syntax will cause an extra set of parentheses to be wrapped around its argument, after formatting that argument as a SQL expression.

These can be combined to allow more fine-grained control over SQL generation:

(def call-qualify-map
  (-> (select [[:foo :bar]] [[:raw "@var := foo.bar"]])
      (from :foo)
      (where [:= :a [:param :baz]] [:= :b [:inline 42]])))

call-qualify-map
=> '{:where [:and [:= :a [:param :baz]] [:= :b [:inline 42]]]
     :from (:foo)
     :select [[[:foo :bar]] [[:raw "@var := foo.bar"]]]}

(sql/format call-qualify-map {:params {:baz "BAZ"}})
=> ["SELECT FOO(bar), @var := foo.bar FROM foo WHERE (a = ?) AND (b = 42)" "BAZ"]

(-> (select :*)
    (from :foo)
    (where [:< :expired_at [:raw ["now() - '" 5 " seconds'"]]])
    (sql/format))
=> ["SELECT * FROM foo WHERE expired_at < now() - '5 seconds'"]

(-> (select :*)
    (from :foo)
    (where [:< :expired_at [:raw ["now() - '" [:lift 5] " seconds'"]]])
    (sql/format))
=> ["SELECT * FROM foo WHERE expired_at < now() - '? seconds'" 5]

(-> (select :*)
    (from :foo)
    (where [:< :expired_at [:raw ["now() - '" [:param :t] " seconds'"]]])
    (sql/format {:params {:t 5}}))
=> ["SELECT * FROM foo WHERE expired_at < now() - '? seconds'" 5]

(-> (select :*)
    (from :foo)
    (where [:< :expired_at [:raw ["now() - " [:inline (str 5 " seconds")]]]])
    (sql/format))
=> ["SELECT * FROM foo WHERE expired_at < now() - '5 seconds'"]

PostGIS

A common example in the wild is the PostGIS extension to PostgreSQL where you have a lot of function calls needed in code:

(-> (insert-into :sample)
    (values [{:location [:ST_SetSRID
                         [:ST_MakePoint 0.291 32.621]
                         [:cast 4325 :integer]]}])
    (sql/format {:pretty true}))
=> ["
INSERT INTO sample
(location) VALUES (ST_SETSRID(ST_MAKEPOINT(?, ?), CAST(? AS integer)))
"
0.291 32.621 4325]

Identifiers

To quote identifiers, pass the :quoted true option to format and they will be quoted according to the selected dialect. If you override the dialect in a format call, by passing the :dialect option, identifiers will be automatically quoted. You can override the dialect and turn off quoting by passing :quoted false. Valid :dialect options are :ansi (the default, use this for PostgreSQL), :mysql, :oracle, or :sqlserver:

(-> (select :foo.a)
    (from :foo)
    (where [:= :foo.a "baz"])
    (sql/format {:dialect :mysql}))
=> ["SELECT `foo`.`a` FROM `foo` WHERE `foo`.`a` = ?" "baz"]

Locking

The ANSI/PostgreSQL/SQLServer dialects support locking selects via a FOR clause as follows:

• :for [<lock-strength> <table(s)> <qualifier>] where <lock-strength> is required and may be one of:
  • :update
  • :no-key-update
  • :share
  • :key-share
• Both <table(s)> and <qualifier> are optional but if present, <table(s)> must either be:
  • a single table name (as a keyword) or
  • a sequence of table names (as keywords)
• <qualifier> can be :nowait, :wait, :skip-locked etc.

If <table(s)> and <qualifier> are both omitted, you may also omit the [..] and just say :for :update etc.

(-> (select :foo.a)
    (from :foo)
    (where [:= :foo.a "baz"])
    (for :update)
    (sql/format))
=> ["SELECT foo.a FROM foo WHERE foo.a = ? FOR UPDATE" "baz"]

If the :mysql dialect is selected, an additional locking clause is available: :lock :in-share-mode.

(sql/format {:select [:*] :from :foo
             :where [:= :name [:inline "Jones"]]
             :lock [:in-share-mode]}
            {:dialect :mysql :quoted false})
=> ["SELECT * FROM foo WHERE name = 'Jones' LOCK IN SHARE MODE"]

Dashes are allowed in quoted names:

(sql/format
  {:select [:f.foo-id :f.foo-name]
   :from [[:foo-bar :f]]
   :where [:= :f.foo-id 12345]}
  {:quoted true})
=> ["SELECT \"f\".\"foo-id\", \"f\".\"foo-name\" FROM \"foo-bar\" AS \"f\" WHERE \"f\".\"foo-id\" = ?" 12345]

Big, complicated example

Here's a big, complicated query. Note that HoneySQL makes no attempt to verify that your queries make any sense. It merely renders surface syntax.

(def big-complicated-map
  (-> (select-distinct :f.* :b.baz :c.quux [:b.bla "bla-bla"]
                       [[:now]] [[:raw "@x := 10"]])
      (from [:foo :f] [:baz :b])
      (join :draq [:= :f.b :draq.x])
      (left-join [:clod :c] [:= :f.a :c.d])
      (right-join :bock [:= :bock.z :c.e])
      (where [:or
               [:and [:= :f.a "bort"] [:not= :b.baz [:param :param1]]]
               [:and [:< 1 2] [:< 2 3]]
               [:in :f.e [1 [:param :param2] 3]]
               [:between :f.e 10 20]])
      (group-by :f.a :c.e)
      (having [:< 0 :f.e])
      (order-by [:b.baz :desc] :c.quux [:f.a :nulls-first])
      (limit 50)
      (offset 10)))

big-complicated-map
=> {:select-distinct [:f.* :b.baz :c.quux [:b.bla "bla-bla"]
                     [[:now]] [[:raw "@x := 10"]]]
    :from [[:foo :f] [:baz :b]]
    :join [:draq [:= :f.b :draq.x]]
    :left-join [[:clod :c] [:= :f.a :c.d]]
    :right-join [:bock [:= :bock.z :c.e]]
    :where [:or
             [:and [:= :f.a "bort"] [:not= :b.baz [:param :param1]]]
             [:and [:< 1 2] [:< 2 3]]
             [:in :f.e [1 [:param :param2] 3]]
             [:between :f.e 10 20]]
    :group-by [:f.a :c.e]
    :having [:< 0 :f.e]
    :order-by [[:b.baz :desc] :c.quux [:f.a :nulls-first]]
    :limit 50
    :offset 10}

(sql/format big-complicated-map
            {:params {:param1 "gabba" :param2 2}
             :pretty true})
=> ["
SELECT DISTINCT f.*, b.baz, c.quux, b.bla AS \"bla-bla\", NOW(), @x := 10
FROM foo AS f, baz AS b
INNER JOIN draq ON f.b = draq.x
LEFT JOIN clod AS c ON f.a = c.d
RIGHT JOIN bock ON bock.z = c.e
WHERE ((f.a = ?) AND (b.baz <> ?)) OR ((? < ?) AND (? < ?)) OR (f.e IN (?, ?, ?)) OR f.e BETWEEN ? AND ?
GROUP BY f.a, c.e
HAVING ? < f.e
ORDER BY b.baz DESC, c.quux ASC, f.a NULLS FIRST
LIMIT ?
OFFSET ?
"
"bort" "gabba" 1 2 2 3 1 2 3 10 20 0 50 10]

;; Printable and readable
(= big-complicated-map (read-string (pr-str big-complicated-map)))
=> true

Extensibility

Any keyword (or symbol) that appears as the first element of a vector will be treated as a generic function unless it is declared to be an operator or "special syntax". Any keyword (or symbol) that appears as a key in a hash map will be treated as a SQL clause -- and must either be built-in or must be registered as a new clause.

If your database supports <=> as an operator, you can tell HoneySQL about it using the register-op! function (which should be called before the first call to honey.sql/format):

(sql/register-op! :<=>)
;; default is a binary operator:
(-> (select :a) (where [:<=> :a "foo"]) sql/format)
=> ["SELECT a WHERE a <=> ?" "foo"]
;; you can declare that an operator is variadic:
(sql/register-op! :<=> :variadic true)
(-> (select :a) (where [:<=> "food" :a "fool"]) sql/format)
=> ["SELECT a WHERE ? <=> a <=> ?" "food" "fool"]

Sometimes you want an operator to ignore nil clauses (:and and :or are declared that way):

(sql/register-op! :<=> :ignore-nil true)

Or perhaps your database supports syntax like a BETWIXT b AND c, in which case you can use register-fn! to tell HoneySQL about it (again, called before the first call to honey.sql/format):

;; the formatter will be passed your new operator (function) and a
;; sequence of the arguments provided to it (so you can write any arity ops):
(sql/register-fn! :betwixt
                  (fn [op [a b c]]
                    (let [[sql-a & params-a] (sql/format-expr a)
                          [sql-b & params-b] (sql/format-expr b)
                          [sql-c & params-c] (sql/format-expr c)]
                      (-> [(str sql-a " " (sql/sql-kw op) " "
                                sql-b " AND " sql-c)]
                          (into params-a)
                          (into params-b)
                          (into params-c)))))
;; example usage:
(-> (select :a) (where [:betwixt :a 1 10]) sql/format)
=> ["SELECT a WHERE a BETWIXT ? AND ?" 1 10]

You can also register SQL clauses, specifying the keyword, the formatting function, and an existing clause that this new clause should be processed before:

;; the formatter will be passed your new clause and the value associated
;; with that clause in the DSL (which is often a sequence but does not
;; need to be -- it can be whatever syntax you desire in the DSL):
(sql/register-clause! :foobar
                      (fn [clause x]
                        (let [[sql & params]
                              (if (keyword? x)
                                (sql/format-expr x)
                                (sql/format-dsl x))]
                          (into [(str (sql/sql-kw clause) " " sql)] params)))
                      :from) ; SELECT ... FOOBAR ... FROM ...
;; example usage:
(sql/format {:select [:a :b] :foobar :baz})
=> ["SELECT a, b FOOBAR baz"]
(sql/format {:select [:a :b] :foobar {:where [:= :id 1]}})
=> ["SELECT a, b FOOBAR WHERE id = ?" 1]

If you find yourself registering an operator, a function (syntax), or a new clause, consider submitting a pull request to HoneySQL so others can use it, too. If it is dialect-specific, let me know in the pull request.

License

Copyright (c) 2020-2021 Sean Corfield. HoneySQL 1.x was copyright (c) 2012-2020 Justin Kramer and Sean Corfield.

Distributed under the Eclipse Public License, the same as Clojure.