Crystal parser combinators.
16bf2c90 — Kim Burgess 3 years ago
chore: bump version
6c6606af — Kim Burgess 3 years ago
docs: fix outdated behaviour in README examples
4f5969cd — Kim Burgess 3 years ago
docs: add example to README


browse  log 



You can also use your local clone with git send-email.


Pars is a library for building monadic parser combinators in crystal-lang. It works with minimal object allocation to extract domain-specific representation from String or Bytes input.

A combinator parser is a system that allows for the creation of small parsers which can then combine to represent more complex semantics. This process then repeats, allowing for increasing complexity. Small parsers combine with logic (OR, AND, XOR) and sequencing to create larger, more meaningful parsers. Ultimately providing a single parser that models a full domain grammar.

This style of parser allows for creating interpreted programming languages, decoding markup, reading files of different formats, decoding communication protocols and other uses where there is a need to extract information from String or Bytes data based on defined syntax.

For a more in-depth introduction, see Monadic Parser Combinators.


Lets start with a domain model.

enum Greeting

alias Target = String

record Expression, greeting : Greeting, target : Target

Now we can build some parsers.

# Creates a parser that returns a single letter character.
letter = Parse.char_if &.letter?

# Parser for 1 or more letters.
letters = letter * (1..)

# We can join these into a new Parser that provides joins these letters as a String.
word = letters.map &.join  # this is already available as `Parse.word` too

# This can now build a parser for one of our domain objects.
greeting = word.map &->Greeting.parse(String)

# But we still have a little more to consume...
comma = Parse.char ','
space = Parse.char ' '
sep = (comma + space) | space
# ...and some more...
target = word
exclamation = Parse.char('!') * (0..1)

# But now we can combine these into the full expression parser
expression = ((greeting << sep) &+ target << exclamation).map do |(g, t)|
  Expression.new g, t

typeof(expression) # => Parser(Expression)

expression.parse "Hello, world!" # => Expression(@greeting=Hello, @target="world")

expression.parse "Hello human" # => Expression(@greeting=Hello, @target="human")

expression.parse "Well then..." # => ParseError: unknown enum Greeting value: Well

Importantly though this is only the start. We can continue to build complexity.

other_expression = Parse.string "Well then..."
new_parser = expression | other_expression

And create results that use crystal's beautiful type system.

result = new_parser.parse("Well then...")
case result
in Expression
  # ...
in String
  # ...
in ParseError
  # ...


require "pars"
include Pars

While not required, including Pars is highly recommended for ease of access.

#Primitive parsers

char_a = Parse.char 'a'

puts char_a.parse "abc" #=> a

This example creates a Parser(Char) from Parse.char, and parses the string "abc" on it. The character parser looks at the beginning of the string, and looks for the first character. If the first character matches the character supplied, then the parse will succeed and the parse result will return the character that matched.

puts char_a.parse "bca" #=> expected 'a', got 'b'

This example uses the same char_a parser, but parses string "bca" on it. Because it doesn't start with 'a', the parse fails and returns a ParseError. A ParseError contains a message about the parse failure, available via ParseError#message. As such, Parser(T)#parse returns a union of (T | ParseError), as it can return either.

str_cat = Parse.string "cat"

puts str_cat.parse "cat" #=> cat
puts str_cat.parse "cats are cool" #=> cat
puts str_cat.parse "dog" #=> expected 'cat', got 'd'

This example creates a new primitive parser, the Parser(String) created by Parse.string(String). It expects an exact copy of the string provided; in this example the text "cat".

When constructing parsers for non-string based input, Parser.byte is also provided.

null_byte = Parse.byte 0x0

puts null_byte.parse Bytes[0xDE, 0xAD, 0xBE, 0xEF] #=> expected `0`, got '222'

Similarly Parser.bytes is available for matching a specific byte sequence.

bovine = Parse.bytes Bytes[0xBE, 0xEF]

puts bovine.parse Bytes[0xBE, 0xEF] #=> Bytes[0xBE, 0xEF]

#Conditional parsers

In some cases, you may want to retrieve a value from the input that matches certain criteria. Two base conditional parsers provide this:

space = Parse.char_if &.whitespace?

or for binary inputs

low_val = Parse.byte_if { |b| b <= 10 }

#Optional parsers

char_a = Parse.char 'a'
char_b = Parse.char 'b'
parse_ab = char_a | char_b

puts parse_ab.parse "abc" #=> a
puts parse_ab.parse "bca" #=> b
puts parse_ab.parse "cab" #=> expected 'b', got 'c'

This example creates three parsers:

  • a Parser(Char) that expects a character of 'a',
  • a Parser(Char) that expects a character of 'b', and
  • a Parser(Char) created using the | operator that will try the left parser first, then the right, and use the successful parser.

The | operator allows you to create branching parsers by using OR logic. It first tries the parser on the left, then the right. If both fail, it will throw the ParseError given by the rightmost parser.

This process is tedious for large masses of characters, such as if you wanted to accept all letters of the alphabet. For this sake, there exists Parse.one_char_of, which looks for any character in the provided string of list.

parse_alphabet = Parse.one_char_of "abcdefghijklmnopqrstuvwxyz"

puts parse_alphabet.parse "abc" #=> a
puts parse_alphabet.parse "bca" #=> b
puts parse_alphabet.parse "xyz" #=> x
puts parse_alphabet.parse "yzx" #=> y
puts parse_alphabet.parse "123" #=> expected 'z', got '1'

This example creates a parser that accepts a char from the provided list. As seen, alphabetical characters parse, but numerical characters do not, as they were not in the original string of the alphabet.

Prebuilt parsers exist for common character types: Parse.lowercase, Parse.uppercase, Parse.letter, Parse.digit, Parse.alphanumeric, Parse.whitespace.

#Repetitive parsers

To create a parser that repeats, use the * operator. This is available on any Parser(T), and outputs a Parser(Array(T)).

When used with an integer, this creates a parser that matches an exact number of times.

triple_a = Parse.char('a') * 3

triple_a.parse("aaa") #=> ['a', 'a', 'a']
triple_a.parse("aa") #=> expected 'a', input ended

To match a variable number of times, use a Range.

some_a = Parse.char('a') * (1..3)

some_a.parse("aaa") #=> ['a', 'a', 'a']
some_a.parse("aa") #=> ['a', 'a']

Endless ranges are also supported, which will continue to match until a ParseError occurs.

existential_dread = Parse.char('a') * (5..)

existential_dread.parse("aaa") #=> expected 'a', input ended
existential_dread.parse("aaaaaaaaaa") #=> ['a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a', 'a']
word = Parse.letter * (1..)

puts word.parse "hello world" #=> ['h', 'e', 'l', 'l', 'o']
puts word.parse "abc" #=> ['a', 'b', 'c']

A clear issue exists with the above example: it returns a list of the characters. If we want to convert this into a usable String, we have to transform the parser.

#Transforming parsers

Existing parsers can be "transformed" to create new parsers with new logic. This provide the ability to move from primitive types to domain-specific types. To transform a parser, use the Parser(T)#map(T -> B) method. This accepts a block that receives the resulting value of a parse as a parameter, and outputs a transformed/mapped value.

For example, if you created a parser that accepted numbers:

digit = Parse.one_char_of "0123456789"

Upon parsing it, it would yield characters on success:

puts (digit.parse "1").class #=> Char

we find that the result is a Char, not any form of a Number! To solve this, we can transform the parser:

digit = (Parse.one_char_of "0123456789").map &.to_i

puts digit.parse "1" #=> 1
puts (digit.parse "1").class #=> Int32

Success! Now the parsed value from our parser is the correct type, Int32.

Back to the issue we found in the word parser from the previous section, we can transform the Array(Char) to a String.

word = (Parse.letter * (1..)).map &.join

puts word.parse "hello world" #=> hello
puts word.parse "abc" #=> abc

This identical word parser is available as Parse.word (Parser(String)).

#Logical combinations

The | (OR) operator already discussed as accompanied by other logical operators.

  • A & B (AND) creates a new parser that ensure both A and B successfull parse for the same input and returns the results as the Tuple {A, B}.
  • A ^ B (XOR) creates a new parser that succeeds with the result of A or B, but fails if both succeed.

#Sequencing parsers

  • A >> B creates a new parser that ensures both A and B parse sequentially, but results with the value of B.
  • A << B creates a new parser that ensures both A and B parse sequentially, but results with the value of A.
  • A + B creates a new parser that ensure both A and B parse sequentially, returning the results as an Array.
  • A &+ B creates a new parser that ensure both A and B parse sequentially, returning the results as a Tuple.
letter = Parse.letter
digit = Parse.digit
parser_take_digit = letter >> digit
parser_take_letter = letter << digit

puts parser_take_digit.parse "a1" #=> 1
puts parser_take_digit.parse "b2" #=> 2

puts parser_take_letter.parse "a1" #=> a
puts parser_take_letter.parse "b2" #=> b

In this example, two parsers are created, letter and digit. Then, two new parsers are created using the >> and << operators. The first parses both sequentially but results with the result of digit, and the second does the same but results with the value of letter. Upon parsing these, the two parsers must work sequentially, but returns with the parser's result the operator is pointing toward.

#Parsing lists

Parse has a special parser that can parse a list of parsable items by parser A, delimited by parser B. Using this, we can create a parser that parses through a list of words (using Parser.word), delimited by a second parser that looks for commas.

word = Parser.word
optional_whitespace = Parser.whitespace * (0..)
comma = (Parser.char ',') << optional_whitespace

list_parser = Parse.list word, comma

puts list_parser.parse "hello, world" #=> ["hello", "world"]
puts list_parser.parse "how,are,    you" #=> ["how", "are", "you"]
puts list_parser.parse "123, 456" #=> []
puts list_parser.parse "hello world, how are you" #=> ["hello"]

#Complex sequential parsers

In the event you need to create complex sequential parsers, you can use Parser(T)#bind. The bind method takes a block that receives the output of Parser(T) as a value, and must return a new Parser of any type, or Parser(B). We can recreate the parser_take_digit and parser_take_letter parsers using this functionality:

letter = Parse.letter
digit = Parse.digit

parser_take_digit = letter.bind do |char_result|
  digit.bind do |digit_result|
    Parse.const digit_result

The original two parsers chain their execution, and ultimately a Parse.const parser returns. Parse.const is a parser that takes in any value of type T. When parsed, it always returns the value of type T. In this case, we create it with the Char result from digit.

parser_letter_digit = letter.bind do |char_result|
  digit.bind do |digit_result|
    Parse.const({char_result, digit_result}) # a constant parser with a `Tuple(Char, Char)`

This parser will parse strings like a1, b2, c3, etc., but return both of the retrieved values as a Tuple.

result = parser_letter_digit.parse "a1"

puts result[0] #=> a
puts result[1] #=> 1

This form of parser sequencing can become tedious. As a result, the library has a special macro inspired by Haskell's do statement. It allows you to chain parsers like above, but in a much more linear and organized manner. Here is the most recent sequential parser parser_letter_digit using Parse.do:

parser_letter_digit = Parse.do({
  char_result <= letter,
  digit_result <= digit,
  Parse.const({char_result, digit_result})

The body of the Parse.do macro is a list of actions separated by commas. The last element of this list must be an expression that is ultimately returned through the new parser.

For each of the other elements in the list, they must be either parser results or local variables.

  • Parser results look like result_variable_name <= parser,. In this case, the result from parser is stored as result_variable_name.
  • Local variables are variable_name = value,. In this case, variable_name is set to value.

Utilizing these tools, more complex parsers are expressible.

word = Parse.word

optional_whitespace = Parse.whitespace * (0..)
equals = optional_whitespace >> (Parse.char '=') << optional_whitespace

key_value_pair = Parse.do({
  key <= word,
  _ <= equals,
  value <= word,
  Parse.const({key, value})

comma = (Parse.char ',') << optional_whitespace

key_value_list = Parse.list key_value_pair, comma

puts key_value_list.parse "hello = world" #=> [{"hello", "world"}]
puts key_value_list.parse "how = are, you= sir" #=> [{"how", "are"}, {"you", "sir"}]
puts key_value_list.parse "all=     sorts,of   =supported, white = spaces" #=> [{"all", "sorts"}, {"of", "supported"}, {"white", "spaces"}]

#Custom parsers

Custom parsers can wrap arbitrary logic. This is sometimes necessary if existing primitive parsers cannot combine effectively or efficiently.

def char_parser(char)
  Parser(Char).new do |context|
    if context.exhausted?
      ParseResult(Char).error "expected '#{char}', got end of input", context
    elsif context.head === char
      ParseResult(Char).new char, context.next
      ParseResult(Char).error "expected '#{char}', got '#{context.head}", context

This defines char_parser(Char), which creates a parser that expects a character as specified. This implementation is the same as the internal implementation Parse.char(Char). See the source code for more applications of Parsers derived from blocks.


Generate docs with crystal docs.


Pars is a fork of Pars3k. It shares much of the same internals and structure but is not API compatible. The public API uses features, idioms and operators specific to crystal-lang. While it may look and feel different, a significant hat-tip needs to go to the original work by Voximity and the authors of libraries which inspired it.