Honey SQL

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

Build

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.

Note: every commit to the develop branch runs CI (GitHub Actions) and successful runs push a MAJOR.MINOR.9999-SNAPSHOT build to Clojars so the very latest version of HoneySQL is always available either via that snapshot on Clojars or via a git dependency on the latest SHA.

HoneySQL 2.x requires Clojure 1.9 or later. It also supports recent versions of ClojureScript and Babashka.

Compared to the legacy 1.x version, HoneySQL 2.x 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 1.x).

Note: you can use 1.x and 2.x side-by-side as they use different group IDs and different namespaces. This allows for a piecemeal migration. See this summary of differences between 1.x and 2.x if you are migrating from 1.x!

Try HoneySQL Online!

John Shaffer has created this awesome HoneySQL web app, written in ClojureScript, so you can experiment with HoneySQL in a browser, including setting different options so you can generate pretty SQL with inline values (via :inline true) for copying and pasting directly into your SQL tool of choice!

Note on code samples

Sample code in this documentation is verified via lread/test-doc-blocks.

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

This section includes a number of usage examples but does not dive deep into the way the data structure acts as a DSL that can specify SQL statements (as hash maps) and SQL expressions and function calls (as vectors). It is recommended that you read the Getting Started section of the documentation before trying to use HoneySQL to build your own queries!

From Clojure:

(refer-clojure :exclude '[distinct filter for group-by into partition-by set update])
(require '[honey.sql :as sql]
         ;; CAUTION: this overwrites several clojure.core fns:
         ;;
         ;; distinct, filter, for, group-by, into, partition-by, set, and update
         ;;
         ;; you should generally only refer in the specific
         ;; helpers that you want to use!
         '[honey.sql.helpers :refer :all :as h]
         ;; so we can still get at clojure.core functions:
         '[clojure.core :as c])

From ClojureScript, we don't have :refer :all. If we want to use :refer, we have no choice but to be specific:

(refer-clojure :exclude '[filter for group-by into partition-by set update])
(require '[honey.sql :as sql]
         '[honey.sql.helpers :refer [select select-distinct from
                                     join left-join right-join
                                     where for group-by having union
                                     order-by limit offset values columns
                                     update insert-into set composite
                                     delete delete-from truncate] :as h]
         '[clojure.core :as c])

Everything is built on top of maps representing SQL queries:

(def sqlmap {:select [:a :b :c]
             :from   [:foo]
             :where  [:= :foo.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 foo.a = ?" "baz"]
;; sqlmap as symbols instead of keywords:
(-> '{select (a, b, c) from (foo) where (= foo.a "baz")}
    (sql/format))
=> ["SELECT a, b, c FROM foo WHERE foo.a = ?" "baz"]

HoneySQL is a relatively "pure" library, it does not manage your JDBC 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))

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.

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 foo.a = 'baz'"]

As seen above, the default parameterization uses positional parameters (?) with the order of values in the generated vector matching the order of those placeholders in the SQL. As of 2.4.962, you can specified :numbered true as an option to produce numbered parameters ($1, $2, etc):

(sql/format sqlmap {:numbered true})
=> ["SELECT a, b, c FROM foo WHERE foo.a = $1" "baz"]

Namespace-qualified keywords (and symbols) 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.

(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"]
;; this also works with symbols instead of keywords:
(-> '{select (foo/a, foo/b, foo/c)
      from   (foo)
      where  (= foo/a "baz")}
    (sql/format))
=> ["SELECT foo.a, foo.b, foo.c FROM foo WHERE foo.a = ?" "baz"]

As of 2.6.1126, there is a helper macro you can use with quoted symbolic queries (that are purely literal, not programmatically constructed) to provide "escape hatches" for certain symbols that you want to be treated as locally bound symbols (and, hence, their values):

;; quoted symbolic query with local substitution:
(let [search-value "baz"]
  (sql/formatv [search-value]
   '{select (foo/a, foo/b, foo/c)
     from   (foo)
     where  (= foo/a search-value)}))
=> ["SELECT foo.a, foo.b, foo.c FROM foo WHERE foo.a = ?" "baz"]

Note: this is a Clojure-only feature and is not available in ClojureScript, and it is intended for literal, inline symbolic queries only, not for programmatically constructed queries (where you would be able to substitute the values directly, as you build the query).

Documentation for the entire data DSL can be found in the Clause Reference, the Operator Reference, and the Special Syntax reference.

Vanilla SQL clause helpers

For every single SQL clause supported by HoneySQL (as keywords or symbols in the data structure that is the DSL), there is also a corresponding function in the honey.sql.helpers namespace:

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

In general, (helper :foo expr) will produce {:helper [:foo expr]} (with a few exceptions -- see the docstring of the helper function for details).

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 [:= :foo.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 (foo.a = ?) AND (b > ?)" "baz" 10]

Note: the helpers always produce keywords so you can rely on dissoc with the desired keyword to remove. If you are building the data DSL "manually" and using symbols instead of keywords, you'll need to dissoc the symbol form instead.

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]

The power of this approach comes from the abiliity to programmatically and conditionally build up queries:

(defn fetch-user [& {:keys [id name]}]
  (-> (select :*)
      (from :users)
      (cond->
        id    (where [:= :id id])
        name  (where [:= :name name]))
      sql/format))

You can call fetch-user with either :id or :name or both and get back a query with the appropriate WHERE clause, since the helpers will merge the conditions into the query DSL.

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]

or conditionally:

(-> (select :a [:b :bar])
    (cond->
      need-c (select :c)
      x-val  (select [:d :x]))
    (from [:foo :quux])
    (where [:= :quux.a 1] [:< :bar 100])
    (cond->
      x-val  (where [:> :x x-val]))
    sql/format)

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.

The examples in this README use a mixture of data structures and the helper functions interchangably. For any example using the helpers, you could evaluate it (without the call to sql/format) to see what the equivalent data structure would be.

Documentation for all the helpers can be found in the honey.sql.helpers API reference.

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]
;; or as pure data DSL:
(-> {: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]
;; or as pure data DSL:
(-> {: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 unless you specify those columns in the :values-default-columns option, which takes a set of column names that should get the value DEFAULT instead of NULL:

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

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"]
;; or as pure data DSL:
(let [user-id 12345
      role-name "user"]
  (-> {:insert-into [:user_profile_to_role]
       :values [{:user_profile_id 12345,
                 :role_id {:select [:id],
                           :from [:role],
                           :where [:= :name "user"]}}]}
      (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)"]
;; or as pure data DSL:
(-> {:select [:*],
     :from [:foo],
     :where [:in :foo.a {:select [:a], :from [:bar]}]}
    (sql/format))
=> ["SELECT * FROM foo WHERE foo.a IN (SELECT a FROM bar)"]

Because values can be nested queries -- and also because values can be function calls -- whenever you are working with values that are, themselves, structured data, you will need to tell HoneySQL not to interpret that structured data as part of the DSL. This especially affects using JSON values with HoneySQL (e.g., targeting PostgreSQL). There are two possible approaches:

1. Use named parameters instead of having the values directly in the DSL structure (see :param under Miscellaneous below), or
2. Use [:lift ..] wrapped around any structured values which tells HoneySQL not to interpret the vector or hash map value as a DSL.

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"]
;; with numbered parameters:
(-> (insert-into :comp_table)
    (columns :name :comp_column)
    (values
     [["small" (composite 1 "inch")]
      ["large" (composite 10 "feet")]])
    (sql/format {:pretty true :numbered true}))
=> ["
INSERT INTO comp_table (name, comp_column)
VALUES ($1, ($2, $3)), ($4, ($5, $6))
"
"small" 1 "inch" "large" 10 "feet"]
;; or as pure data DSL:
(-> {: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:

(-> (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"]
;; or as pure data DSL:
(-> {: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"]
;; or as pure data DSL:
(-> {: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"]
;; or pure data DSL:
(-> {: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 TABLE films"]
;; or as pure data DSL:
(-> {:truncate :films}
    (sql/format))
=> ["TRUNCATE TABLE 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"]

Note: different databases have different precedence rules for these set operations when used in combination -- you may need to use :nest to add ( .. ) in order to combine these operations in a single SQL statement, if the natural order produced by HoneySQL does not work "as expected" for your database.

Functions

Function calls (and expressions with operators) can be specified as vectors where the first element is either a keyword or a symbol:

(-> (select :*) (from :foo)
    (where [:> :date_created [:date_add [:now] [:interval 24 :hours]]])
    (sql/format))
=> ["SELECT * FROM foo WHERE date_created > DATE_ADD(NOW(), INTERVAL ? HOURS)" 24]

Note: The above example may be specific to MySQL but the general principle of vectors for function calls applies to all dialects.

A shorthand syntax also exists for simple function calls: keywords that begin with % are interpreted as SQL function calls:

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

;; with an alias:
(-> (select [:%count.* :total]) (from :foo) sql/format)
=> ["SELECT COUNT(*) AS total 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 [[:count :*] :total]) (from :foo) sql/format)
=> ["SELECT COUNT(*) AS total FROM foo"]

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

(-> (select [[:max :id]]) (from :foo) sql/format)
=> ["SELECT MAX(id) FROM foo"]
(-> (select [[:max :id] :highest]) (from :foo) sql/format)
=> ["SELECT MAX(id) AS highest FROM foo"]
;; the pure data DSL requires an extra level of brackets:
(-> {:select [[[:max :id]]], :from [:foo]} sql/format)
=> ["SELECT MAX(id) FROM foo"]
(-> {:select [[[:max :id] :highest]], :from [:foo]} sql/format)
=> ["SELECT MAX(id) AS highest FROM foo"]
;; the shorthand makes this simpler:
(-> {:select [[:%max.id]], :from [:foo]} sql/format)
=> ["SELECT MAX(id) FROM foo"]
(-> {:select [[:%max.id :highest]], :from [:foo]} sql/format)
=> ["SELECT MAX(id) AS highest FROM foo"]
;; or even (no alias):
(-> {:select [:%max.id], :from [:foo]} sql/format)
=> ["SELECT MAX(id) FROM foo"]
;; or even (no alias, no other columns):
(-> {:select :%max.id, :from :foo} sql/format)
=> ["SELECT MAX(id) FROM foo"]

Custom columns using functions are built with the same vector format. Be sure to properly nest the vectors so that the first element in the selection is the custom function and the second is the column alias.

(sql/format
  {:select   [:job_name                                      ;; A bare field selection
              [[:avg [:/ [:- :end_time :start_time] 1000.0]] ;; A custom function
               :avg_exec_time_seconds                        ;; The column alias
               ]]
   :from     [:job_data]
   :group-by :job_name})
=> ["SELECT job_name, AVG((end_time - start_time) / ?) AS avg_exec_time_seconds FROM job_data GROUP BY job_name" 1000.0]

If a keyword begins with ', the function name is formatted as a SQL entity rather than being converted to uppercase and having hyphens - converted to spaces). That means that hyphens - will become underscores _ unless you have quoting enabled:

(-> (select :*) (from :foo)
    (where [:'my-schema.SomeFunction :bar 0])
    (sql/format))
=> ["SELECT * FROM foo WHERE my_schema.SomeFunction(bar, ?)" 0]
(-> (select :*) (from :foo)
    (where [:'my-schema.SomeFunction :bar 0])
    (sql/format :quoted true))
=> ["SELECT * FROM \"foo\" WHERE \"my-schema\".\"SomeFunction\"(\"bar\", ?)" 0]
(-> (select :*) (from :foo)
    (where [:'my-schema.SomeFunction :bar 0])
    (sql/format :dialect :mysql))
=> ["SELECT * FROM `foo` WHERE `my-schema`.`SomeFunction`(`bar`, ?)" 0]

Note: in non-function contexts, if a keyword begins with ', it is transcribed into the SQL exactly as-is, with no case or character conversion at all.

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"]
;; or with numbered parameters:
(-> (select :id)
    (from :foo)
    (where [:= :a :?baz])
    (sql/format {:params {:baz "BAZ"} :numbered true}))
=> ["SELECT id FROM foo WHERE a = $1" "BAZ"]
;; or as pure data DSL:
(-> {: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 -- as is common when working with PostgreSQL's JSON/JSONB types).

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]

Entity Names

To quote SQL entity names, 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, SQL entity names 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. As of 2.5.1091, :nrql is also supported:

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

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

See New Relic NRQL Support for more details of the NRQL dialect.

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]
            :eldr [:= :f.e :eldr.t])
      (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]
           :eldr [:= :f.e :eldr.t]]
    :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 INNER JOIN eldr ON f.e = eldr.t
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]
;; with numbered parameters:
(sql/format big-complicated-map
            {:params {:param1 "gabba" :param2 2}
             :pretty true :numbered 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 INNER JOIN eldr ON f.e = eldr.t
LEFT JOIN clod AS c ON f.a = c.d
RIGHT JOIN bock ON bock.z = c.e
WHERE ((f.a = $1) AND (b.baz <> $2)) OR (($3 < $4) AND ($5 < $6)) OR (f.e IN ($7, $8, $9)) OR f.e BETWEEN $10 AND $11
GROUP BY f.a, c.e
HAVING $12 < f.e
ORDER BY b.baz DESC, c.quux ASC, f.a NULLS FIRST
LIMIT $13
OFFSET $14
"
"bort" "gabba" 1 2 2 3 1 2 3 10 20 0 50 10]

;; Printable and readable
(require '[clojure.edn :as edn])

(= big-complicated-map (edn/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! :<=>)
;; all operators are assumed to be variadic:
(-> (select :a) (where [:<=> :a "foo"]) sql/format)
=> ["SELECT a WHERE a <=> ?" "foo"]
(-> (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)]
                          (c/into params-a)
                          (c/into params-b)
                          (c/into params-c)))))
;; example usage:
(-> (select :a) (where [:betwixt :a 1 10]) sql/format)
=> ["SELECT a WHERE a BETWIXT ? AND ?" 1 10]
;; with numbered parameters:
(-> (select :a) (where [:betwixt :a 1 10]) (sql/format {:numbered true}))
=> ["SELECT a WHERE a BETWIXT $1 AND $2" 1 10]

Note: the generation of positional placeholders (?) or numbered placeholders ($1, $2, etc) is handled automatically by format-expr so you get this behavior "for free" in your extensions, as long as you use the public API for honey.sql. You should avoid writing extensions that generate placeholders directly if you want them to work with numbered parameters.

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 (ident? x)
                                (sql/format-expr x)
                                (sql/format-dsl x))]
                          (c/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.

HoneySQL 1.x will continue to get critical security fixes but otherwise should be considered "legacy" at this point.

License

Copyright (c) 2020-2024 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.