Parse indentation level with PEG.js

Parser:

// do not use result cache, nor line and column tracking

{ var indentStack = [], indent = ""; }

start
  = INDENT? l:line
    { return l; }

line
  = SAMEDENT line:(!EOL c:. { return c; })+ EOL?
    children:( INDENT c:line* DEDENT { return c; })?
    { var o = {}; o[line] = children; return children ? o : line.join(""); }

EOL
  = "\r\n" / "\n" / "\r"

SAMEDENT
  = i:[ \t]* &{ return i.join("") === indent; }

INDENT
  = &(i:[ \t]+ &{ return i.length > indent.length; }
      { indentStack.push(indent); indent = i.join(""); pos = offset; })

DEDENT
  = { indent = indentStack.pop(); }

Input:

a
  b
  c
  d
    z
    y
    x

Output:

{
   "a": [
      "b",
      "c",
      {
         "d": [
            "z",
            "y",
            "x"
         ]
      }
   ]
}

It cannot parse an empty object (last x), however, it should be easy to solve. Trick here is the SAMEDENT rule, it succeeds when indentation level hasn't changed. INDENT and DEDENT change current indentation level without changing position in text pos = offset.

Update 2021

Here is a working example which runs in the online playground of Peggy.js. Peggy.js is a fork of PEG.js under active development. PEG.js was discontinued by David Maida.

The example shows, how the INDENT, SAMEDENT and DEDENT rules are parsed, and how to use parsing locations. Check the console log.

It uses these syntaxes, which may not be known from other parser generators:

(top of file)

• {{...}} (Global initializer) – Run ... on parser generation.
• {...} (Per-parse initializer) – Run ... on parser instantiation.

(in-file)

• X {...} (action) – Do ... when X succeeds. Variables from the initializers are available. If ... returns something, it will replace what X returns.
• $X – Return the raw text parsed with X, instead of the result of X.
• ... @X ... (pluck operator) – Replace the result of ... X ... with the result of X.
• X &{...} (predicate) – "and ... also needs to be true for X to succeed".
• X = &(...) – If ... succeeds, X succeeds. ... consumes no input.

See the docs for more information.

{{
    console.clear()
    console.log('Parser generated')
}}

{
    let indentstack = []
    let indent = ''
    function found (what) {
        let loc = location()
        console.log(`[${loc.start.line}:${loc.start.column} - ${loc.end.line}:${loc.end.column}] found ${what}`)
    }
    console.log('Parser instantiated')
}

DOCUMENT = NEWLINES? @THINGS NEWLINES? _

THINGS = ( SAMEDENT @( OBJECT / LINE ) )*

OBJECT = key:KEY childs:(BLOCK / INLINE) {
    found(`object "${key}"`)
    let o = {}
    o[key] = childs
    return o
}

KEY = @$( [^ \t\r\n:]+ ) _ ':' _

BLOCK = NEWLINES INDENT @THINGS DEDENT

INLINE = line:LINE { return [line] }

LINE = text:$( (!EOL .)+ ) NEWLINES? {
    found(`line "${text}"`)
    return text
}

INDENT = &(
    spaces:$( [ \t]+ ) &{
        return spaces.length > indent.length
    } {
        indentstack.push(indent)
        indent = spaces
    }
) {
    found('indent')
}

SAMEDENT = spaces:$( [ \t]* ) &{
    return spaces === indent
} {
    found('samedent')
}

/* Because of this rule, results cache must be disabled */
DEDENT = &{
    indent = indentstack.pop()
    return true
} {
    found('dedent')
}

_ = [ \t]*
EOL = '\r\n' / '\n' / '\r'
NEWLINES = (_ EOL)+

/* Test with this input

H:
  a
  b
  c
  G:
    d
    e
    f

*/

---

Old Answer

Here is a fix for @Jakub Kulhan´s grammar which works in PEG.js v 0.10.0. The last line needs to be changed to = &{ indent = indentStack.pop(); return true;} because PEG.js now does not allow standalone actions ({...}) in a grammar anymore. This line is now a predicate (&{...}) which always succeeds (return true;).

i also removed the pos = offset; because it gives an error offset is not defined. Probably Jakub was referring to some global variable available in older versions of PEG.js. PEG.js now provides the location() function which returns an object which contains offset and other information.

// do not use result cache, nor line and column tracking

{ var indentStack = [], indent = ""; }

start
  = INDENT? l:line
    { return l; }

line
  = SAMEDENT line:(!EOL c:. { return c; })+ EOL?
    children:( INDENT c:line* DEDENT { return c; })?
    { var o = {}; o[line] = children; return children ? o : line.join(""); }

EOL
  = "\r\n" / "\n" / "\r"

SAMEDENT
  = i:[ \t]* &{ return i.join("") === indent; }

INDENT
  = &(i:[ \t]+ &{ return i.length > indent.length; }
      { indentStack.push(indent); indent = i.join(""); })

DEDENT
  = &{ indent = indentStack.pop(); return true;}

Starting with v 0.11.0 PEG.js also supports the Value Plucking operator, @ which would allow writing this grammar even simpler, but as it is currently not in the online parser I will refrain from adding it to this example.