Skip to content
Go to file

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

We turn our backs on confusion and seek the beginning. -- Sevrin

Note: Clojure 1.9 will introduce a new core library, known as clojure.spec, which makes Herbert obsolete.


A schema language for edn (Clojure data).

Way to Eden

The extensible data notation (edn) defines a useful subset of Clojure data types. As described on

edn is a system for the conveyance of values. It is not a type system, and has no schemas.

The explicit lack of schemas in edn stands in marked contrast to many serialization libraries which use an interface definition language. The edn values essentially speak for themselves, without the need for a separate description or layer of interpretation. That is not to say that schemas aren't potentially useful, they're just not part of the definition of the edn format.

The goal of the Herbert project is to provide a convenient schema language for defining edn data structures that can be used for documentation and validation. The schema patterns are represented as edn values.


Herbert is available from Clojars. Add the following dependency to your project.clj:

Herbert on


The main namespace is miner.herbert. The conforms? predicate takes a schema pattern and a value to test. It returns true if the value conforms to the schema pattern, false otherwise.

The conform function is used to build a test function. Given a schema, it returns a function of one argument that will execute a match against the schema pattern and return a map of bindings if successful or nil for a failed match. If you need to know how the schema bindings matched a value or you want to test against a schema multiple times, you should use conform to define a test function.

Quick example:

(require '[miner.herbert :as h])
(h/conforms? '{:a int :b [sym+] :c str} '{:a 42 :b [foo bar baz] :c "foo"})
;=> true

;; For better performance, create a test function with `h/conform`.
(def my-test (h/conform '{:a (:= A int) :b [sym+] :c str}))
(my-test '{:a 42 :b [foo bar baz] :c "foo"})
;=> {A 42}

Test.Check integration

The property function takes a predicate and a schema as arguments and returns a test.check property suitable for generative testing. (test.check also has a defspec macro for use with clojure.test.) If you just want the generator for a schema, call generator. The sample function is similar to test.check version but takes a schema.

(require '[miner.herbert.generators :as hg])
(require '[clojure.test.check :as tc])

;; trivial example
(tc/quick-check 100 (hg/property integer? 'int))

;; confirm the types of the values
(tc/quick-check 100 (hg/property (fn [m] (and (integer? (:int m)) (string? (:str m)))) 
	                             '{:int int :str str :kw kw}))

;; only care about the 42 in the right place
(tc/quick-check 100 (hg/property (fn [m] (== (get-in m [:v 2 :int]) 42))
                                '{:v (vec kw kw {:int 42} kw) :str str}))

;; samples from a schema generator
(clojure.test.check.generators/sample (hg/generator '[int*]))
;=> (() (9223372036854775807) [9223372036854775807] () [] (1 1) () [-7] (4) [-5])

;; generate samples directly from a schema (notice the "hg" namespace)
(hg/sample '[int*])
;=> (() [-1 0] () (9223372036854775807) [9223372036854775807] [] () () [0 -5] [9223372036854775807] (7 9223372036854775807) [] (12 -11) (-9223372036854775808) [-12 9223372036854775807] [-10 9223372036854775807] (-11) [2] (-11) [-7])

Notation for Schema Patterns

  • Literal constants match themselves:
    nil, true, false, numbers, "strings", :keywords

  • Empty literal collections match themselves:
    [], (), {}

  • A simple schema pattern is named by a symbol:

    • int - integer
    • float - floating-point
    • str - string
    • kw - keyword
    • sym - symbol
    • vec - vector
    • list - list or cons (actually anything that satisfies clojure.core/seq?)
    • seq - any sequential (including vectors)
    • map - map
    • char - character
    • bool - boolean
    • any - anything
  • A few additional schema patterns for numeric sub-types:

    • num - any number
    • pos - positive number
    • neg - negative number
    • zero - zero number
    • even - even integer
    • odd - odd integer
  • A quantified schema pattern: adding a *, + or ? at the end of a symbol for zero-or-more, one-or-more, or zero-or-one (optional):
    int*, str+, sym?

  • A quoted expression matches itself without any other interpretation:
    'foo? matches the symbol foo? literally.

  • A compound schema pattern: using and, or and not
    (or sym+ nil) -- one or more symbols or nil
    (or (vec int*) (list kw+)) -- either a vector of ints or a list of one or more keywords

  • A quantified schema pattern: a list beginning with *, + or ? as the first element.
    (* kw sym) -- zero or more cycles of keywords and symbols

  • A named schema expression is written as a list with the first element being the := operator, followed by a (non-reserved) symbol as the binding name, and the rest of the list being a schema pattern. The names of predicates and special operators (like and, or, etc.) are not allowed as binding names. The name may be used as a parameter to other schema patterns. Also, the name may be used in the pattern expression to create a recursive pattern.
    (:= N int 1 10) -- matches 1 to 10 (inclusive)
    (:= A (or :a [:b A])) -- matches [:b [:b [:b :a]]]

  • A bound symbol matches an element equal to the value that the name was bound to previously.
    [(:= N int) N N] -- matches [3 3 3]

  • A literal vector [in square brackets] matches any sequential (not just a vector) with the contained pattern.
    [(* kw sym)] -- matches (:a foo :b bar :c baz) and [:a foo]

  • A literal map in {curly braces} matches any map with the given literal keys and values matching the corresponding schemas. Optional keywords are written with a ? suffix such as :kw?. (Use a quote mark to match a literal keyword ending with ?. ':k? matches :k? literally without any special interpretation of the ? suffix.) For convenience, an optional keyword schema implicitly allows nil for the corresponding value. An empty literal map {} matches exactly the empty map. Use map to match any map.
    {:a int :b sym :c? [int*]} -- matches {:a 10 :b foo :c [1 2 3]} and {:a 1 :b bar}
    {:x? sym ':k? int} -- matches {:k? 10} but not {:k 10} because the keyword was quoted.

  • The literal map in {curly braces} may also contains a single pair of patterns with a non-literal key pattern. All keys and and values are required to match in the map value. This kind of pattern is useful for matching "functional" maps.
    {kw int} -- matches {:a 10 :b 20}, but not {:a 1 :b "bar"}

  • A literal #{set} with multiple schema patterns denotes the required elements, but does not exclude others. A single element might match multiple patterns. A set with a quantified schema pattern defines the requirement on all elements.
    #{int :a :b} -- matches #{:a :b :c 10}, but not #{:a 10}
    #{int+} -- matches #{1 3 5}, but not #{1 :a 3}

  • Numeric schema patterns, such as int, even, odd, float, or num, may take optional parameters in a list following the pattern name. Numerics take a low and a high parameter. The value must be between to the low and high (inclusive) for it to match. If only one parameter is given, it defines the high, and the low defaults to 0 in that case. If neither is given, there is no restriction on the high or low values. Quantified numeric patterns apply the high and low to all the matched elements.
    (int 1 10) -- matches 4, but not 12

  • String, symbol and keyword schema patterns (such as str, sym and kw) may take an optional regex argument, specified as a string (for edn compatibility) or a Clojure regular expression (like #"[Rr]ege?x"). In that case, the pr-str of the element must match the regex.
    (kw ":user/.*") -- matches :user/foo

  • An inlined schema pattern: a list starting with & as the first element refers to multiple elements in order (as opposed to being within a collection). It can be useful for adding when tests where an extra element would not normally be allowed.
    {:a (:= N int) :b (& (:= F float) (> N F))} -- matches {:a 4 :b 3.14}

  • The map schema predicate matches a map. It takes the same arguments as the {curly brace} literal map schema. With no arguments, (map) matches any map, same as map. Use {} to match the empty map.
    (map :a int :b sym :c? [int*]) -- matches {:a 10 :b foo :c [1 2 3]} and {:a 1 :b bar}

  • The list schema predicate matches a list or cons. It can take multiple optional arguments to specify the schemas for the ordered elements of the list.
    (list sym (* kw int)) -- matches (foo :a 42 :b 52 :c 22)

  • The vec schema predicate matches a vector. It can take multiple optional arguments to specify the schemas for the ordered elements of the vector.
    (vec int (* sym int)) -- matches [4 foo 42 bar 52]

  • The seq schema predicate matches any sequential (vector or list). It's basically the same as using the [square bracket] notation.
    (seq kw int sym) -- matches (:a 10 foo) and [:b 11 bar]

  • The set schema predicate matches a set. It's basically the same as the #{set} literal notation.
    (set :a :b) -- matches #{:a :b :c 10}, but not #{:a 10}
    (set int+) -- matches #{1 3 5}, but not #{1 :a 3}

  • A list starting with class followed by a dotted symbol matches an instance of that Java class. In general, you should avoid bringing classes directly into your schema pattern. It's more flexible and extensible to use tag.
    (class java.util.Date) -- matches a java.util.Date, but not a java.util.Calendar

  • The tag list pattern takes a symbol as its first argument. The pattern matches against the edn-tag of the item. The first argument may also be a string which is interpreted as a regex matching the pr-str of the edn-tag. The optional second argument may be a literal (such as a "string") in which case the value is whatever constant edn/read-string would construct by reading a string with that tag and literal. On the other hand, if the option second argument is a non-literal schema pattern that pattern will be matched against the edn-value of the item. For example, the Herbert map notation would be appropriate for matching a record value. The edn-tag and edn-value are defined in the tagged project. See the miner.tagged.EdnTag protocol for more information. Basically, the edn-tag is the symbol that normally would be used to print as a tagged record (record class my.ns.Rec would use tag my.ns/Rec). Several Java classes corresponding to built-in tagged literals (see clojure.core/default-data-readers) have predefined tags as well. The edn-value is typically a map for a record or the item itself for most other classes. Custom records and Java classes can extend the miner.tagged.EdnTag protocol to participate in tag pattern matching.
    (tag my.ns/Rec {:a int}) -- matches an instance of the record class my.ns.Rec with an integer value for the key :a.
    (tag inst) -- matches any instance of java.util.Date, java.util.Calendar or java.sql.Timestamp

  • The pred list pattern takes as the first argument the name of a Clojure function that should be called to test the input. The function name should be a fully qualified symbol naming the var holding the predicate function. If the predicate is parameterized, the implementing function should take those parameters first. In all cases, the last argument should be the item in question. Note that the predicate should accept all values for consideration without throwing an exception. For example, the even schema predicate is implemented with a test of clojure.core/integer? as well as clojure.core/even? because the latter will throw on non-integer values. The default predicates are defined in the var miner.herbert/default-predicates.
    (pred clojure.string/blank?) -- matches nil or "" or " "

  • The grammar pattern defines a grammar for more complex schema patterns. The first argument is the start-pattern which is the actual pattern to match. It is followed by zero or more rules, declared as an inline pair of a symbol, the term, and its pattern definition. A rule can refer to previously defined terms or use its own term symbol in a recursive pattern. The start-pattern can refer to any of the terms in the grammar form. If you want to go crazy, you can nest another grammar pattern as the definition of a term, but the nested grammar expression is in an isolated scope so its rules are not available to the outer scope.
    (grammar [person+] phone (str "\\d{3}+-\\d{3}+-\\d{4}+") person {:name str :phone phone}) -- matches [{:name "Herbert" :phone "408-555-1212"} {:name "Jenny" :phone "415-867-5309"}]

Regular Expression Support

For conformance testing (as with conforms?), Herbert allows several terms to be parameterized by regular expression (see str, sym, etc). Both the Clojure syntax for regular expressions and the Java String format are allowed (see clojure.core/re-pattern and java.util.regex.Pattern.) Note that Clojure regular expressions (like #"foo+bar*") are not edn types, so you should use Strings if you want your Herbert schemas to be completely edn-compatible. The main difference is that Java String notation requires you to use double backslashes to get the effect of a single backslash in your regex. For example, Clojure #"foo\d" would be written as the String "foo\\d".

Although the full Java regular expression syntax is supported for conformance testing, the Herbert generator implementation supports only a limited form of regular expressions. (Someday, test.check may support a regex generator, but for now, Herbert has to implement string generation as best it can.) The built-in string generator supports basic regular expressions with ASCII characters, such as "[a-z] [^abc] a.b* c+d(ef|gh)? \d\D\w\W\s\S". It does not support advanced regex features such as unicode notation, minimum and maximum match counts, intersection character classes, POSIX character classes, case-insensitity flags, look-ahead, look-back, back-references, greedy, reluctant or possessive quantification, etc.

The dynamic Var *string-from-regex-generator* allows the user to customize the test.check string generator used internally by Herbert. When *string-from-regex-generator* is bound to a test.check generator, Herbert will use this generator for any term that is parameterized by a regular expression. The generator should take one argument, which can be either a java.util.regex.Pattern or a String, as the regex. It should generate strings that match the given regex. When it's bound to nil (the default), Herbert will use its internal string generator as described above.

If you need better support for Java regular expressions when generating Strings, you should consider using the test.chuck library which provides the string-from-regex generator. You can use it with Herbert like this:

(require '[com.gfredericks.test.chuck.generators :as chuck])
(require '[miner.herbert :as h])
(require '[miner.herbert.generators :as hg])

(binding [h/*string-from-regex-generator* chuck/string-from-regex]
	(hg/sample '(str #"\x66oo\dbar{1,3}") 5))

;=> ("foo5bar" "foo2barr" "foo5barrr" "foo2barrr" "foo9bar")

Experimental Features

These features are implemented as an experiment, but I'm not sure I'll keep them as they're a bit of a hack:

  • The when form does not consume any input. The expression is evaluated within the enviroment of the previous bindings -- if it returns a logical true, the match continues. On a logical false, the whole match fails.
    [(:= N int) (:= M int) (when (== (* 3 N) M))] -- matches [2 6]

  • A list starting with =, ==, not=, <, >, <= or >= is an implied when and treated as if the form was within an when test.
    [(:= N int) (:= M int) (== (* 3 N) M)] -- matches [2 6]


(require '[miner.herbert :as h])

(h/conforms? 'int 10)
;=> true

(h/conforms? '(grammar int) 10)
; a very simple "grammar" with no rules, equivalent to the start pattern
;=> true

(h/conforms? '(grammar {show numbers}, show str, numbers [int+]) '{"Lost" [4 8 15 16 23 42]})
; target pattern can use named subpatterns defined by tail of name/pattern pairs
;=> true

(h/conforms? '(:= A (or :a [:b A])) [:b [:b [:b :a]]])
; matches a recursive binding of `A`
;=> true

(h/conforms? '{:a int :b sym :c? [str*]} '{:a 1 :b foo :c ["foo" "bar" "baz"]})
;=> true

(h/conforms? '{:a int :b sym :c? [str*]} '{:a 1 :b foo})
; :c is optional so it's OK if it's not there at all.
;=> true

(h/conforms? '{:a int :b sym :c? [str*]} '{:a foo :b bar})
;=> false

(h/conforms? '{:a (:= A int) :b sym :c? [A+]} '{:a 1 :b foo :c [1 1 1]})
; `A` is bound to the int associated with :a, and then used again to define the values
; in the seq associated with :c.  
;=> true

(h/conforms? '(& {:a (:= A int) :b (:= B sym) :c (:= C [B+])} (when (= (count C) A))) 
           '{:a 2 :b foo :c [foo foo]})

; The & operator just means the following elements are found inline,
; not in a collection.  In this case, we use it to associate the
; when-test with the single map constraint.  The assertion says that
; number of elements in the :c value must be equal to the value
; associated with :a.  Notice that all the elements in the :c seq
; must be equal to the symbol associated with :b.
=> true

((h/conform '[(:= A int) (:= B int) (:= C int+ A B)]) [3 7 4 5 6])
; Inside a seq, the first two ints establish the low and high range of the rest 
; of the int values.
;=> {C [4 5 6], B 7, A 3}

(def my-checker (h/conform '[(:= MAX int) (:= XS int+ MAX)]))
(my-checker [7 3 5 6 4])
;=> {XS [3 5 6 4], MAX 7}

(defn palindrome? [s]
	(and (string? s)
		(= s (clojure.string/reverse s))))
(h/conforms? '(grammar [pal+]
	              palindrome user/palindrome?
                  pal {:len (:= LEN int) :palindrome (and palindrome (cnt LEN))})
             [{:palindrome "civic" :len 5}
              {:palindrome "kayak" :len 5} 
              {:palindrome "level" :len 5}
              {:palindrome "ere" :len 3}
              {:palindrome "racecar" :len 7}])
;=> true


If you want to mix external data into a Herbert pattern, I suggest that you use the backtick library's template function.


Related Projects

If Herbert isn't exactly what you're looking for, here are some other projects that take different approaches to similar problems:

Herbert is obsolete as of Clojure 1.9

Star Trek: The Way to Eden

stardate 5832.3

Space Hippies: "Herbert, Herbert, Herbert ..."
Spock: "Herbert was a minor official notorious for his rigid and limited patterns of thought."
Kirk: "Well, I shall try to be less rigid in my thinking."

video clip:

Way to Eden

Copyright and License

Copyright (c) 2013 Stephen E. Miner.

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


Clojure library defining a schema for edn values


You can’t perform that action at this time.