Skip to content

patmorin/ssCC

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

77 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ssCC - A Parser Code Generator

What is it?

ssCC is a tool used to create a parser based on the definition of a language.

Can you be more specific?

ssCC reads in language definition files and outputs Java code that parses input into a parse tree, which may then be use for your own purposes such as interpreting or compiling.

Syntax

ssCC [options] definitions

options

--package packagename
	includes all of the created classes in the specified package

--prefix prefixname
	prefixes all created classes with the specified string

definitions is one of the following

tokendef grammardef

--tokenizer-only tokendef

--parser-only grammardef

examples

This will create a tokenizer and parser with all classes belonging to the MyProject package.

ssCC --package MyProject token.def grammar.def

This one create only a tokenizer and prefixes each class with "BOB"

ssCC --prefix BOB --tokenizer-only token.def

Language Definitions

How is the language defined?

A language is defined in two steps and in their own seperate files: first by its tokens, then by its grammar.

Tell me about tokens and how to define them

Tokens are individual "pieces" of your language which have a type and a value. As an example you may define the english language has having token types such as punctuation, nouns and verbs with possible values such as "!", "George" and "to run" respectively.

Tokens are defined one per line in a regular text file. The syntax of a token definition is as follows:

tokenname: definition

The name must be of only alphanumeric characters ('a' to 'b' and '0' to '9') and must start with a letter.

You may also define internal tokens which are not reported by the created tokenizer but may be used in other token definitions. To mark a token as being internal place a colon in front of the name as such:

:internal: definition

The definition part is written much like regular expressions. Here are the operators you may use:

*   Match zero or more

+   Match one or more

?   Match one or none

|   Match the pattern on either side, much like an OR

\   Escape, matches the next character (used to match characters that are operators, such as \+ or \[ or \:)

\\n  Matches a newline characters

\\r  Matches a carriage return character

\\t  Matches a tab character

\\s  Matches a space character

()  Group patterns

[abc]  Character class, matches any character within the brackets

[^abc] Negative character class, matches any character that is _not_ within the brackets

:tok:  Uses the pattern of the named token (not necessarily internal) to match, 
       note that the token must be defined *before* embeding it

Also, comments may be written with the # sign either after a definition or on its own line.

Note that spaces and tabs are ignored in the definition, therefore the definition "J o e" is the same as "Joe", although most importantly if you want to match the space in "Joe Shmoe" the definition should use the \s special character like so "Joe \s Shmoe".

Show me some token definition examples

# matching "khan", "khaan", "khaaan", "khaaaan"...
tok1: kha+n

# matching "fun" or "sun"
tok2: [fs]un

# matching "wild cats" or "wild dogs"
tok3: wild \s (cats | dogs)

# matches a quoted string
tok5: " [^"]* "

# matching a variable name (alphanumeric, starting with a letter)
:alpha: [abcdefghijklmnopqrstuvwxyz]
:digit: [0123456789]
tok4: :alpha: (:alpha: | :digit:)*

Alright, now what about language grammars

A language's grammar defines the form of all strings that belong to the language. In very simple terms, it defines which tokens can follow which tokens.

The (context-free) grammar is defined by a set of rules written one per line in its own text file. Each rule is of the following form:

RuleName -> definition 

The name must again be of only alphanumeric characters ('a' to 'b' and '0' to '9') and must start with a letter.

The definition is comprised of token and rule names along with a few operators. These operators are much again like regular expressions:

*  Match zero or more

?  Match one or none

|  Match the pattern on either side

() Group patterns

Notice that not all operators used in token definitions are available here.

There are additionally two special symbols that you may use in the definition:

\\0  Epsilon, an 'empty' match, see examples (only use a single backslash)

[>1] Multiple child flag, must be placed at the end of the definition,
     signals that the rule should be included in the parse tree only if the node has more than one child,
     see the discussion about parse trees for more information 

Note that literals are not allowed in this definition, only the names of tokens and grammar rules.

You may again also use the # sign for comments, either after a definition or on it's own line.

Show me some grammar definition examples

In these examples we use the convention of having token names in lowercase and grammar rule names in word-case (ie: first letter of each word in uppercase).

# matching a phone number of the form "(613) 555-1234"
PhoneNumber -> AreaCode space FirstPart dash SecondPart

AreaCode -> leftparen threedigits rightparen

FirstPart -> threedigits

SecondPart -> fourdigits


# matching a person's name with optional title and optional middle names
FullName -> NameTitle FirstName MiddleNames LastName

NameTitle -> title | \\0  # equivalent to "title?"

FirstName -> name

MiddleNames -> name*

LastName -> name


# matching a "if" block with possible multiple "elsif" and "else" 
# (some rule definitions omitted for the sake of brevity)
IfStatement -> if Condition Block (elsif Condition Block)* (else Block)?

Condition -> leftparen LogicalStatement rightparen 

Block -> leftsquiggly Statement* rightsquiggly

Resulting Code

What does ssCC give me after feeding it the definition files?

ssCC will write several Java classes, of which the two main ones are Tokenizer and Parser.

As you might have guessed, Tokenizer is created based on your token definition, while Parser is created based on the grammar definition.

What does Tokenizer do?

The Tokenizer class is designed to read in a stream of characters and to output a stream of tokens, which are encapsulated in the Token class, based on your definition.

Methods

Tokenizer(Reader r)
	The constructor takes in an object of type Reader (such as an InputStreamReader)

Token nextToken()
	Returns the next token found in the character stream based on the token definitions file

void pushToken()
	Pushes the last returned token so that the next call to nextToken() returns it, 
	can be called multiple times to push through the history of returned tokens,
	however a limit is imposed and defaulted to 20. See setTokenHistorySize()

int getLineNumber()
	Returns the current line number in the Reader

void setTokenHistorySize(int size)
	Sets how many tokens the Tokenizer will remember in order to use pushToken(), defaulted to 20

int getTokenHistorySize() 
	Returns the number of tokens remembered for pushToken()

What does Parser do?

The Parser class is constructed to read in a stream of Tokens from a Tokenizer and to output a parse tree based on the given grammar definition. You can then use the parse tree however you like, most often for interpreting or compiling.

Methods

Parser(iTokenizer t)
	The constructor takes an object that implements the iTokenizer interface, including the Tokenizer generated by ssCC

ASTxxxNode parse()
	Parser's only method returns the parse tree generated from the input of the given Tokenizer based on the grammar definition file.
	To be more specific it returns the top node of the tree of type ASTxxxNode, of which the xxx is defined by the first rule in the grammar.

What about all the other classes that are created?

Here is a quick outline of each class created, minus exception classes:

  • ASTNode - Base parse tree node class

  • ASTToken - A superclass of ASTNode, represents a token in the parse tree

  • ASTxxxNode - A superclass of ASTNode, one created for every grammar rule (xxx is replaced by the rule's name)

  • iTokenizer - An interface implemented by Tokenizer

  • Parser - The main parsing class, takes a Tokenizer and outputs a parse tree

  • Token - A token outputed from Tokenizer

  • Tokenizer - The main tokenizing class, reads in a character stream and outputs a stream of Token

  • Visitor - An interface implementing the visitor pattern, used to traverse the parse tree (read on for more information)

Usage

Enough with the technical, how do I use ssCC?

First we create the two definition files, let's call them "tokens.def" and "grammar.def"

tokens.def

number: [0123456789]+

plus: \+
minus: -
multiply: \*
divide: /

grammar.def

Equation -> Sum*

Sum -> Term ( (plus | minus) Term )* [>1]

Term -> number ( (multiply | divide) number)* [>1]

We then feed them to ssCC like so: ssCC tokens.def grammar.def

As mentioned previously we get a handful of java classes in return, among which are Tokenizer and Parser.

To be able to use these classes in our own program we must create a main class, let's call it SuperApp.

Here's a sample of code for SuperApp:

import java.io.*;

class SuperApp {
	
	public static void main(String args[]) {
	
		try {
		
			Tokenizer tokenizer = new Tokenizer(new InputStreamReader(System.in));
			
			Parser parser = new Parser(tokenizer);
			
			ASTEquationNode equation = parser.parse();
			
			equation.accept(new InterpretorVisitor(), null);
			
		}
		catch (Exception e) {
		
			System.out.println(e);
			
		}
	}
	
}

You'll notice that in the last line of the try block that the parse tree (or rather the parse tree's root node) "accepts" an object of the type InterpretorVisitor. First we will explain what the parse tree is, then we'll show how to traverse it using a Visitor.

What is the parse tree?

The parse tree is a collection of nodes that are attached in a parent/child relationship.

Each rule in your grammar is one possible node type, and each rule and token included in its definition is a child of that rule.

example

with the following rules

A -> B C

B -> x D

C -> y

D -> z

the parse tree would be

    A
   / \
  B   C
 / \   \
x   D   y
    |
    z

What is a Visitor and how does it traverse the parse tree?

About

ssCC - A Parser Code Generator

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Java 98.4%
  • Perl 1.6%