Cross-references

The ability to refer to other parser rules is what gives PEG parsers their power. A rule can refer to any other rule in the grammar by name, and this has the effect of using the named rule to match the next piece of input. Here’s a simplified example of using rules to match an email address:

email.peg
grammar Email
  email     <-  username "@" host
  username  <-  [a-z]+ ("." [a-z]+)*
  host      <-  [a-z]+ "." ("com" / "co.uk" / "org" / "net")
require('./email').parse('bob@example.com')
   == { text: 'bob@example.com',
        offset: 0,
        elements: 
         [ { text: 'bob', offset: 0, elements: [...] },
           { text: '@', offset: 3, elements: [] },
           { text: 'example.com', offset: 4, elements: [...] } ],
        username: { text: 'bob', offset: 0, elements: [...] },
        host: 
         { text: 'example.com',
           offset: 4,
           elements: [...] } }

As you can see in the above parse tree, the rules referenced by the email rule add named nodes called username and host to the parse tree. This gives you an easier way to traverse the tree than using the elements array.

var tree = require('./email').parse('bob@example.com');
tree.username.text    == 'bob'
tree.host.text        == 'example.com'

References allow you to create recursive matchers, which is why PEGs can match languages that regular expressions typically can’t. Here’s a grammar for matching nested lists of numbers:

lists.peg
grammar Lists
  value   <-  list / number
  list    <-  "[" value ("," value)* "]" 
  number  <-  [0-9]

We can parse a string using this grammar and browse the tree it generates:

var tree = require('./lists').parse('[[1,2],3]')

tree.text
   == '[[1,2],3]'

tree.elements[1].text
   == '[1,2]'

tree.elements[1].elements[2].elements[0].elements[1].text
   == '2'