A Parsing Expression Grammar ( hence peg) is a way to create grammars similar in principle to regular expressions but which allow better code integration. Specifically, peg is an implementation of the Packrat parser generator originally implemented as peg/leg by Ian Piumarta in C. A Packrat parser is a "descent recursive parser" capable of backtracking and negative look-ahead assertions which are problematic for regular expression engines .

• https://en.wikipedia.org/wiki/Parsing_expression_grammar
• https://bford.info/packrat/
• https://piumarta.com/software/peg/

go install github.com/pointlander/peg@latest

peg [<option>]... <file>

Usage of peg:
  -inline
      parse rule inlining
  -noast
      disable AST
  -output string
      specify name of output file
  -print
      directly dump the syntax tree
  -strict
      treat compiler warnings as errors
  -switch
      replace if-else if-else like blocks with switch blocks
  -syntax
      print out the syntax tree
  -version
      print the version and exit

This sample Makefile will convert any file ending with .peg into a .go file with the same name. Adjust as needed.

.SUFFIXES: .peg .go

.peg.go:
	peg -noast -switch -inline -strict -output $@ $<

all: grammar.go

Use caution when picking your names to avoid overwriting existing .go files. Since only one PEG grammar is allowed per Go package (currently) the use of the name grammar.peg is suggested as a convention:

grammar.peg
grammar.go

First declare the package name and any import(s) required:

package <package name>

import <import name>

Then declare the parser:

type <parser name> Peg {
	<parser state variables>
}

Next declare the rules. Note that the main rules are described below but are based on the peg/leg rules which provide additional documentation.

The first rule is the entry point into the parser:

<rule name> <- <rule body>

The first rule should probably end with !. to indicate no more input follows.

first <- . !.

This is often set to END to make PEG rules more readable:

END <- !.

. means any character matches. For zero or more character matches, use:

repetition <- .*

For one or more character matches, use:

oneOrMore <- .+

For an optional character match, use:

optional <- .?

If specific characters are to be matched, use single quotes:

specific <- 'a'* 'bc'+ 'de'?

This will match the string "aaabcbcde".

For choosing between different inputs, use alternates:

prioritized <- 'a' 'a'* / 'bc'+ / 'de'?

This will match "aaaa" or "bcbc" or "de" or "". The matches are attempted in order.

If the characters are case-insensitive, use double quotes:

insensitive <- "abc"

This will match "abc" or "Abc" or "ABc" and so on.

For matching a set of characters, use a character class:

class <- [a-z]

This will match "a" or "b" or all the way to "z".

For an inverse character class, start with a caret:

inverse <- [^a-z]

This will match anything but "a" or "b" or all the way to "z".

If the character class is case-insensitive, use double brackets:

insensitive <- [[A-Z]]

(Note that this is not available in regular expression syntax.)

Use parentheses for grouping:

grouping <- (rule1 / rule2) rule3

For looking ahead a match (predicate), use:

lookAhead <- &rule1 rule2

For inverse look ahead, use:

inverse <- !rule1 rule2

Use curly braces for Go code:

gocode <- { fmt.Println("hello world") }

For string captures, use less than and greater than:

capture <- <'capture'> { fmt.Println(text) }

Will print out "capture". The captured string is stored in buffer[begin:end].

Testing a grammar usually requires more than the average unit testing with multiple inputs and outputs. Grammars are also usually not for just one language implementation. Consider maintaining a list of inputs with expected outputs in a structured file format such as JSON or YAML and parsing it for testing or using one of the available options for Go such as Rob Muhlestein's tinout package.

• Golang, see go.mod for version
• golangci-lint latest version
• gofumpt

  go install mvdan.cc/gofumpt@latest
  

go run build.go

or

go generate

go run build.go test

golangci-lint run

gofumpt -l -w .

• bootstrap/main.go - bootstrap syntax tree of peg
• tree/peg.go - syntax tree and code generator
• peg.peg - peg in its own language

Andrew Snodgrass

Here are some projects that use peg to provide further examples of PEG grammars:

• https://github.com/tj/go-naturaldate - natural date/time parsing
• https://github.com/gnames/gnparser - scientific names parsing