Parse JSON into an Abstract Syntax Tree using Flex/Bison

In a previous post we Parsed JSON using Flex/Bison. This example was very basic. We used a FLEX scanner to tokenize our JSON and created grammar rules in BISON to parse this token stream. This post will take it further and create an Abstract Syntax Tree (AST). The AST is a power data structure typically used in compilers that will store the structure of our JSON. Effectively we will read in the JSON structure a tree-like data structure in C.

Bison Grammar

We will use same grammar from json.org for our Bison grammar. They represent this in McKeeman Form. Below is the converted McKeeman Form grammar from json.org to Bison. Note, this grammar isn’t exact to what is on json.org, but close. For example we don’t have characters, digit, digits, integer, fraction … our FLEX scanner is handling a lot of this and it’s not found in the Bison parser.

// from our parser.y Bison file
json:
    | value
    ;

value: object
     | array
     | STRING
     | NUMBER_FLOAT
     | NUMBER_INTEGER
     | VTRUE
     | VFALSE
     | VNULL
     ;

object: LCURLY RCURLY
      | LCURLY members RCURLY
      ;

members: member
       | members COMMA member
       ;

member: STRING COLON value
      ;

array: LBRAC RBRAC
     | LBRAC elements RBRAC
     ;

elements: element
        | elements COMMA element
        ;

element: value
       ;

Notes about this Grammar

Here are some notes to clarify the grammar.

The result of JSON is a value. Here are some examples of valid JSON: null, 5, [] and {}. These are all values.
We have simple values: "true", "false", null, number and string.
We have complex values which are the array and object.
The object contains members which is a list of type member. Each member has a name and a value. Members are separated by a comma.
The array contains elements which is a list of type element. Each element is a value. The element doesn’t have a name like we have in the memeber which is a part of an object. Elements are separated by a comma.

An AST for JSON

The AST for JSON only has two structures. One structure is a value_type which contains the type of value it is, and a void * to that data. The other structure is a member_type which is part of a JSON object that contains a name and a struct value_type *value to the value.

AST Structures

We define our AST structures in a file called ast.h. These make up part of the Bison declaration section.

// a blurb from ast.h
enum VALUE_TYPE
{
  OBJECT_TYPE,
  ARRAY_TYPE,
  STRING_TYPE,
  NUMBER_TYPE,
  INTEGER_TYPE,
  BOOLEAN_TYPE,
  NULL_TYPE
};

struct value_type
{
  enum VALUE_TYPE type;
  // the data pointer can be a primitive like a string, int, boolean or null
  // or to an array or object
  void *data;
};

struct member_type
{
  char *name;
  struct value_type *value;
};

Bison Declarations

The Bison declarations will map our C types in our AST to the Rules Syntax of Bison. Note the struct List * for membs and elems.

We don’t define members or elements in our AST, we instead use a Linked List for them. The members list will be a list of the member_type and the elements list will be a list of value_type. The value_type for an array will have the void data pointer pointing to the list and the value_type for object will have the void data pointer pointing to the members list. We’ll see this in the Bison Declarations.

%start json

%union {
  char *string;
  float decimal;
  int integer;
  struct value_type *v;
  struct member_type *memb;
  struct List *membs;
  struct List *elems;
}

%token LCURLY RCURLY LBRAC RBRAC COMMA COLON
%token VTRUE VFALSE VNULL
%token <string> STRING;
%token <decimal> NUMBER_FLOAT;
%token <integer> NUMBER_INTEGER;

%type <v> value object array json element
%type <memb> member
%type <membs> members
%type <elems> elements

Bison Grammar Rules

Now that we have our AST definitions in place we can define the rules to build an AST structure as JSON is parsed.

%%

json:
    {
     $$ = NULL;
    }
    | value
    {
      $$ = $1;
      json = $$;
    }
    ;

value: object {
        $$ = $1;
     }
     | array {
        $$ = $1;
     }
     | STRING {
        $$ = new_value_string($1);
     }
     | NUMBER_FLOAT {
       $$ = new_value_number($1);
     }
     | NUMBER_INTEGER {
       $$ = new_value_integer($1);
     }
     | VTRUE {
       $$ = new_value_boolean(1);
     }
     | VFALSE {
       $$ = new_value_boolean(0);
     }
     | VNULL {
       $$ = new_value_null();
     }
     ;

object: LCURLY RCURLY {
        $$ = new_value_object(NULL);
      }
      | LCURLY members RCURLY {
        $$ = new_value_object($2);
      }
      ;

members: member {
         $$ = new_members();
         list_add_last($$, $1);
       }
       | members COMMA member {
         list_add_last($1, $3);
       }
       ;

member: STRING COLON value {
        $$ = new_member_type($1, $3);
      }
      ;

array: LBRAC RBRAC {
       $$ = new_value_array(NULL);
     }
     | LBRAC elements RBRAC {
       $$ = new_value_array($2);
     }
     ;

elements: element {
          $$ = new_elements();
          new_array_element($$, $1);
        }
        | elements COMMA element {
          new_array_element($1, $3);
        }
        ;

element: value {
         $$ = $1;
       }
       ;

%%
body

Implementation of the AST Functions

The implementation of the AST functions is simple and quite repetitive. Look particularly at the new_array_element and new_object_member and how they relate into the Bison parser.

struct value_type*
new_value_array(struct List *elements)
{
  struct value_type *v = malloc(sizeof(struct value_type));

  v->type = ARRAY_TYPE;
  v->data = elements;

  return v;
}

struct value_type*
new_value_object(struct List *members)
{
  struct value_type *v = malloc(sizeof(struct value_type));

  v->type = OBJECT_TYPE;
  v->data = members;

  return v;
}

struct value_type*
new_value_string(char *string)
{
  struct value_type *v = malloc(sizeof(struct value_type));

  v->type = STRING_TYPE;
  v->data = strip_quotes(string);

  return v;
}

struct value_type*
new_value_integer(int i)
{
  struct value_type *v = malloc(sizeof(struct value_type));
  int *ii = malloc(sizeof(int));

  *ii = i;

  v->type = INTEGER_TYPE;
  v->data = ii;

  return v;
}

struct value_type*
new_value_boolean(int b)
{
  struct value_type *v = malloc(sizeof(struct value_type));
  int *bb = malloc(sizeof(int));

  *bb = b;

  v->type = BOOLEAN_TYPE;
  v->data = bb;

  return v;
}

struct value_type*
new_value_number(float f)
{
  struct value_type *v = malloc(sizeof(struct value_type));
  int *ff = malloc(sizeof(float));

  v->type = NUMBER_TYPE;
  v->data = ff;

  return v;
}

struct value_type*
new_value_null()
{
  struct value_type *v = malloc(sizeof(struct value_type));
  v->type = NULL_TYPE;
  v->data = NULL;

  return v;
}

struct List *
new_members()
{
 struct List *members = malloc(sizeof(struct List));
 list_init(members, member_match, member_free);
 return members;
}

struct member_type*
new_member_type(char *name, struct value_type *value)
{
  struct member_type *m = malloc(sizeof(struct member_type));
  size_t len;

  m->name = strip_quotes(name);
  m->value = value;

  return m;
}

struct List *
new_elements()
{
  struct List *elements = malloc(sizeof(struct List));
  list_init(elements, member_match, member_free);
  return elements;
}

int
new_array_element(struct List *elements, struct value_type *element)
{
  return list_add_last(elements, element);
}

int
new_object_member(struct List *members, struct member_type *member)
{
  return list_add_last(members, member);
}

Traversing the AST

Once we’ve parse a file that contains JSON into our AST we will then traverse through the AST and print out the values. This will be done with a recursive function.

Take the following formatted file as input to our parser.

{
  "people":
    {
      "first": "bob",
      "last" : "stevens",
      "children": [ "sue", "anne" ],
      "wallet": null,
      "legs": true,
      "hair": false,
      "age" : 23,
      "height": 42.1
    }
}

We can parse it and print it out. It’s not pretty printed and the numeric types won’t round out as exactly as passed in. However, it works.

$ ./parse_json test/test6.json
{"people":{"first":"bob","last":"stevens","children":["sue","anne"],"wallet":null,"legs":true,"hair":false,"age":23,"height":42.099998}}

To recursively print the AST we have the following. The recursion happens only when we have an object or array type.

void
json_print_recursive(struct value_type *value)
{
  struct List* members;
  struct List* elements;
  struct member_type *member;

  char *str;
  int *integer;
  float* number;

  if(value == NULL)
  {
    perror("value is null!?\n");
    return;
  }

  switch(value->type)
  {
  case OBJECT_TYPE:
    members = value->data;

    if(members == NULL)
    {
      printf("{}");
      break;
    }

    printf("{");
    for(int i=0; i<list_size(members); i++)
    {
      member = list_get_index(members, i);
      printf("\"%s\":", member->name);
      json_print_recursive(member->value);
      if(i < (list_size(members)-1))
      {
        printf(",");
      }
    }
    printf("}");

    break;
  case ARRAY_TYPE:
    elements = value->data;

    printf("[");
    for(int i=0; i<list_size(elements); i++)
    {
      value = list_get_index(elements, i);
      json_print_recursive(value);
      if(i < (list_size(elements)-1))
      {
        printf(",");
      }
    }
    printf("]");

    break;
  case STRING_TYPE:
    str = value->data;
    printf("\"%s\"", str);
    break;
  case INTEGER_TYPE:
    integer = value->data;
    printf("%d", *integer);
    break;
  case NUMBER_TYPE:
    number = value->data;
    printf("%f", *number);
    break;
  case BOOLEAN_TYPE:
    integer = value->data;
    if(*integer != 0)
      printf("true");
    else
      printf("false");
    break;
  case NULL_TYPE:
      printf("null");
    break;
  default:
    perror("unknown value type");
  }
  return;
}

Downloading and Usage

Download the full source of parse_json.

To use it do the following.

$ tar xf parse_json-1.1.tar.gz
$ cd parse_json
$ ./configure
$ make
$ ./src/parse_json some_file.json

The files are all inside the src folder:

scanner.l the Flex Scanner
parser.y the Bison Parser
ast.h and ast.c the C code for the Abstract Syntax Tree
main.c where the parsing starts

#json