Regal

Royally reified regular expressions

tl;dr

Regal lets you manipulate regular expressions as data, by providing a Hiccup-like regex syntax, and ways to convert between this Hiccup syntax (Regal syntax), compiled regex patterns, and test.check generators. It also helps with writing cross-platform code by providing consistent semantics across JS/Java runtimes, and it allows converting JavaScript regex to Java regex semantically (useful for e.g. dealing with JSON Schema in Clojure)

The slightly longer version

Regal provides a syntax for writing regular expressions using plain Clojure data: vectors, keywords, strings. This is known as Regal notation.

Once you have a Regal form you can either compile it to a regex object (java.util.regex.Pattern or JavaScript RegExp), or you can use it to create a Generator (see test.check) for generating values that conform to the given pattern.

It is also possible to parse regular expression patterns back to Regal forms.

Regal is Clojure and ClojureScript compatible, and has fixed semantics across platforms. Write your forms once and run them anywhere! It also allows manipulating multiple regex flavors regardless of the current platform, so you can do things like converting a JavaScript regex pattern to one that is suitable for Java's regex engine.

Support Lambda Island Open Source

Regal is part of a growing collection of quality Clojure libraries and tools released on the Lambda Island label. If you find value in our work please consider becoming a backer on Open Collective

Project status

Regal is alpha level software, this does not mean it is of low quality or not fit for use, it does mean that future breakage of the API is still possible.

The following aspects of the library are generally well tested and developed, and we intend to retain compatibility as much as practically possible.

• Regal syntax as described in this README
• Generating regex patterns from regal forms
• Parsing regex patterns to regal forms

The following aspects have known issues or are otherwise untested or incomplete, and you can expect them to change significantly as we further develop them:

• Creating test.check generators from regal forms
• clojure.spec-alpha integration
• Malli integration

Installation

deps.edn

lambdaisland/regal {:mvn/version "0.0.143"}

project.clj

[lambdaisland/regal "0.0.143"]

An example

(require '[lambdaisland.regal :as regal]
         '[lambdaisland.regal.generator :as regal-gen])

;; Regal expression, like Hiccup but for Regex
(def r [:cat
        [:+ [:class [\a \z]]]
        "="
        [:+ [:not \=]]])

;; Convert to host-specific regex
(regal/regex r)
;;=> #"[a-z]+\Q=\E[^=]+"

;; Match strings
(re-matches (regal/regex r) "foo=bar")
;;=> "foo=bar"

;; ... And generate them
(regal-gen/gen r)
;;=> #clojure.test.check.generators.Generator{...}

(regal-gen/sample r)
;;=> ("t=�" "d=5Ë" "zja=·" "uatt=ß¾" "lqyk=É" "xkj=q\f��" "gxupw=æ" "pkadbgmc=¯²" "f=Ã�J" "d=ç")

A swiss army knife

Regal can convert between three different represenations for regular expressions, Regal forms, patterns(i.e. strings), and regex objects. Here is an overview of how to get from one to the other.

↓From / To→	Form	Pattern	Regex
Form	identity	lambdaisland.regal/pattern	lambdaisland.regal/regex
Pattern	lambdaisland.regal.parse/parse-pattern	identity	lambdaisland.regal/compile
Regex	lambdaisland.regal.parse/parse	lambdaisland.regal/regex-pattern	identity

Regal forms

Forms consist of vectors, keywords, strings, character literals, and in some cases integers. For example:

[:cat [:alt [:char 11] [:char 13]] \J [:rep "hello" 2 3]]

Forms have platform-independent semantics. The same regal form will match the same strings both in Clojure and ClojureScript, even though Java and JavaScript (and even different versions of Java or JavaScript) have different regex "flavors". In other words, we generate the regex that is right for the target platform.

;; Clojure
(regal/regex :vertical-whitespace) ;;=> #"\v"

;; ClojureScript
(regal/regex :vertical-whitespace) ;;=> #"[\n\x0B\f\r\x85\u2028\u2029]"

Regal currently knows about three "flavors"

• :java8 Java 1.8 (earlier versions are not supported)
• :java9 Java 9 or later
• :ecma ECMAScript (JavaScript)

By default it takes the flavor that is best suited for the platform, but you can override that with lambdaisland.regal/with-flavor

(regal/with-flavor :ecma
  (regal/pattern ...))

Note that using regal/regex with a flavor that does not correspond with the flavor of the platform may yield unexpected results, when dealing with "foreign" regex flavors always stick to string representations (i.e. patterns).

Pattern

The second regex representation regal knows about is the pattern, i.e. the regex pattern in string form.

(regal/regex-pattern #"\u000B\v") ;; => "\\u000B\\v"

Depending on the situation there are several reasons why you might want to use this pattern representation over the compiled regex object.

• simple strings, so easy to (de-)serialize
• value semantics (can be compared)
• allow manipulating regex pattern of regex flavors other than the one supported by the current runtime

Note that in Clojure the syntax available in regex patterns differs from the syntax available in strings, in particluar when it comes to notations starting with a backslash. e.g. #"\xFF" is a valid regex, while "\xFF" is not a valid string. We encode regex patterns in strings, which practically speaking means that backslashes are escaped (doubled).

(regal/regex-pattern #"\xFF") ;;=> "\\xFF"
(regal/compile "\\xFF")       ;;=> #"\xFF"

Regex

To use the regex engine provided by the runtime (e.g. through re-find or re-seq) you need a platform-specific regex object. This is what lambdaisland.regal/regex gives you.

Grammar

• Strings and characters match literally. They are escaped, so . matches a period, not any character, ^ matches a caret, etc.
• A few keywords have special meaning.
  • :any : match any character, like .. Does not match newlines.
  • :start match the start of the input
  • :end : match the end of the input
  • :digit : match any digit (0-9)
  • :non-digit : match non-digits (not 0-9)
  • :word : match word characters (A-Za-z0-9_)
  • :non-word : match non-word characters (not A-Za-z0-9_)
  • :newline : Match \n
  • :return : Match \r
  • :tab : Match \t
  • :form-feed : Match \f
  • :line-break : Match \n, \r, \r\n, or other unicode newline characters
  • :alert : match \a (U+0007)
  • :escape : match \e (U+001B)
  • :whitespace : match any whitespace character. Uses \s on JavaScript, and a character range of whitespace characters on Java with equivalent semantics as JavaScript \s, since \s in Java only matches ASCII whitespace.
  • :non-whitespace : match non-whitespace
  • :vertical-whitespace : match vertical whitespace, including newlines and vertical tabs #"\n\x0B\f\r\x85\u2028\u2029"
  • :vertical-tab : match a vertical tab \v (U+000B)
  • :null : match a NULL byte/char
• All other forms are vectors, with the first element being a keyword
  • [:cat forms...] : concatenation, match the given Regal expressions in order
  • [:alt forms...] : alternatives, match one of the given options, like (foo|bar|baz)
  • [:* form] : match the given form zero or more times
  • [:+ form] : match the given form one or more times
  • [:? form] : match the given form zero or one time
  • [:*? form] : lazily match the given form zero or more times
  • [:+? form] : lazily match the given form one or more times
  • [:?? form] : lazily match the given form zero or one time
  • [:class entries...] : match any of the given characters or ranges, with ranges given as two element vectors. E.g. [:class [\a \z] [\A \Z] "_" "-"] is equivalent to [a-zA-Z_-]
  • [:not entries...] : like :class, but negates the result, equivalent to [^...]
  • [:repeat form num] : repeat a form fixed number of times, like {5}
  • [:repeat form min max] : repeat a form a number of times, like {2,5}
  • [:lazy-repeat form num] : lazily repeat a form fixed number of times, like {5}?
  • [:lazy-repeat form min max] : lazily repeat a form a number of times, like {2,5}?
  • [:capture forms...] : capturing group with implicit concatenation of the given forms
  • [:char number] : a single character, denoted by its unicode codepoint
  • [:ctrl char] : a control character, e.g. [:ctrl \A] => ^A => #"\cA"
  • [:lookahead ...] : match if followed by pattern, without consuming input
  • [:negative-lookahead ...] : match if not followed by pattern
  • [:lookbehind ...] : match if preceded by pattern
  • [:negative-lookbehind ...] : match if not preceded by pattern
  • [:atomic ...] : match without backtracking (atomic group)
• A clojure.spec.alpha definition of the grammar can be made available as :lambdaisland.regal/form by explicitly requiring lambdaisland.regal.spec-alpha

You can add your own extensions (custom tokens) by providing a :registry option mapping namespaced keywords to Regal expressions.

Use with spec.alpha

(require '[lambdaisland.regal.spec-alpha :as regal-spec]
         '[clojure.spec.alpha :as s]
         '[clojure.spec.gen.alpha :as gen])

(s/def ::x-then-y (regal-spec/spec [:cat [:+ "x"] "-" [:+ "y"]]))

(s/def ::xy-with-stars (regal-spec/spec [:cat "*" ::x-then-y "*"]))

(s/valid? ::xy-with-stars "*xxx-yy*")
;; => true

(gen/sample (s/gen ::xy-with-stars))
;; => ("*x-y*"
;;     "*xx-y*"
;;     "*x-y*"
;;     "*xxxx-y*"
;;     "*xxx-yyyy*"
;;     "*xxxx-yyy*"
;;     "*xxxxxxx-yyyyy*"
;;     "*xx-yyy*"
;;     "*xxxxx-y*"
;;     "*xxx-yyyy*")

Use with Malli

(require '[malli.core :as m]
         '[malli.error :as me]
         '[malli.generator :as mg]
         '[lambdaisland.regal.malli :as regal-malli])

(def malli-opts {:registry {:regal regal-malli/regal-schema}})

(def form [:+ "y"])

(def schema (m/schema [:regal form] malli-opts))

(m/form schema)
;; => [:regal [:+ "y"]]

(m/type schema)
;; => :regal

(m/validate schema "yyy")
;; => true

(me/humanize (m/explain schema "xxx"))
;; => ["unknown error"]

(me/humanize (m/explain schema "xxx") {:errors {:regal {:error/message {:en "Pattern does not match"}}}})
;; => ["Pattern does not match"]

(mg/sample schema)
;; => ("y" "yy" "yy" "yyyy" "yyyy" "y" "yyy" "yyyyy" "yyyyyyyyy" "yyyyyyyy")

BYO test.check / spec-alpha

Regal does not declare any dependencies. This lets people who only care about using Regal Expressions to replace normal regexes to require lambdaisland.regal without imposing extra dependencies upon them.

If you want to use lambdaisland.regal.generator you will require org.clojure/test.check. For lambdisland.regal.spec-alpha you will additionally need org.clojure/spec-alpha.

Contributing

Everyone has a right to submit patches to this projects, and thus become a contributor.

Contributors MUST

• adhere to the LambdaIsland Clojure Style Guide
• write patches that solve a problem. Start by stating the problem, then supply a minimal solution. *
• agree to license their contributions as MPLv2.
• not break the contract with downstream consumers. **
• not break the tests.

Contributors SHOULD

• update the CHANGELOG and README.
• add tests for new functionality.

If you submit a pull request that adheres to these rules, then it will almost certainly be merged immediately. However some things may require more consideration. If you add new dependencies, or significantly increase the API surface, then we need to decide if these changes are in line with the project's goals. In this case you can start by writing a pitch, and collecting feedback on it.

* This goes for features too, a feature needs to solve a problem. State the problem it solves, then supply a minimal solution.

** As long as this project has not seen a public release (i.e. is not on Clojars) we may still consider making breaking changes, if there is consensus that the changes are justified.

Prior Art

• irregex for Chicken Scheme
• CL-PPRE create-scanner (Common Lisp)
• test.chuck string-from-regex
• rx for Emacs Lisp (and source)
• cgrand/regex
• ClojureVerbalExpressions
• FLIP, Bobrow and Teitelman, 1966

License

Copyright © 2020 Arne Brasseur

Licensed under the term of the Mozilla Public License 2.0, see LICENSE.