Syntactic analysis

@production(prod, priority=None)

Use this decorator to declare a grammar production:

class MyParser(LRParser, ReLexer):
    @production('E -> E "+" E')
    def sum(self):
        pass

See the Production syntax section.

The priority argument may be specified to declare that the production has the same priority as an existing token type. Typical use for unary minus:

class MyParser(LRParser, ReLexer):
    # omitting productions for binary +, -, * and /
    @production('E -> "-" E', priority='*')
    def minus(self):
        pass

You can also use a token type that has not been declared to the lexer as long as you have declared an explicit priority for it, using one of the associativity decorators:

@leftAssoc('+', '-')
@leftAssoc('*', '/')
@nonAssoc('UNARYMINUS') # Non associative, higher priority than anything else
class MyParser(LRParser, ReLexer):
    @production('E -> "-" E', priority='UNARYMINUS')
    def minus(self):
        pass

exception ptk.parser.ParseError(grammar, tok, state)

Syntax error when parsing.

Variables:token – The unexpected token.

expecting()

Returns a set of tokens types that would have been valid in input.

ptk.parser.leftAssoc(*operators)

Class decorator for left associative operators. Use this to decorate your Parser class. Operators passed as argument are assumed to have the same priority. The later you declare associativity, the higher the priority; so the following code

@leftAssoc('+', '-')
@leftAssoc('*', '/')
class MyParser(LRParser):
    # ...

declares ‘+’ and ‘-‘ to be left associative, with the same priority. ‘*’ and ‘/’ are also left associative, with a higher priority than ‘+’ and ‘-‘.

See also the priority argument to production().

ptk.parser.rightAssoc(*operators)

Class decorator for right associative operators. Same remarks as leftAssoc().

ptk.parser.nonAssoc(*operators)

Class decorator for non associative operators. Same remarks as leftAssoc().

class ptk.parser.LRParser

LR(1) parser. This class is intended to be used with a lexer class derived from LexerBase, using inheritance; it overrides LexerBase.newToken() so you must inherit from the parser first, then the lexer:

class MyParser(LRParser, ReLexer):
    # ...

newSentence(sentence)

This is called when the start symbol has been reduced.

Parameters:sentence – The value associated with the start symbol.

class ptk.async_parser.AsyncLRParser

This class works like LRParser but supports asynchronous methods (new in Python 3.5). You must use AsyncLexer in conjuction with it:

class Parser(AsyncLRParser, AsyncLexer):

# ...

And only use AsyncLexer.asyncFeed() to feed it the input stream. Semantic actions must be asynchronous methods as well. When the start symbol is reduced, the asyncNewSentence() method is awaited.

Production syntax

Basics

The productions specified through the production() decorator must be specified in a variant of BNF; for example

class Parser(LRParser, ReLexer):
    @production('E -> E plus E')
    def binaryop_sum(self):
        pass

     @production('E -> E minus E')
     def binaryop_minus(self):
         pass

Here non terminal symbols are uppercase and terminals (token types) are lowercase, but this is only a convention.

When you don’t need separate semantic actions you can group several productions by using either the ‘|’ symbol:

class Parser(LRParser, ReLexer):
    @production('E -> E plus E | E minus E')
    def binaryop(self):
        pass

Or decorating the same method several times:

class Parser(LRParser, ReLexer):
    @production('E -> E plus E')
    @production('E -> E minus E')
    def binaryop(self):
        pass

Semantic values

The semantic value associated with a production is the return value of the decorated method. Values for items on the right side of the production are not passed to the method by default; you have to use a specific syntax to associate each item with a name, which will then be used as the name of a keyword argument passed to the method. The name must be specified between brackets after the item, for instance:

class Parser(LRParser, ReLexer):
    @production('E -> E<left> plus E<right>')
    def sum(self, left, right):
        return left + right

You can thus use alternatives and default argument values to slightly change the action’s behavior depending on the actual matched production:

class Parser(LRParser, ReLexer):
    @production('SYMNAME -> identifier<value> | identifier<value> left_bracket identifier<name> right_bracket')
    def symname(self, value, name=None):
        if name is None:
            # First form, name not specified
        else:
            # Second form

Litteral tokens

A litteral token name may appear in a production, between double quotes. This allows you to skip declaring “simple” tokens at the lexer level.

class Parser(LRParser, ReLexer):
    @production('E -> E "+" E')
    def sum(self):
        pass

Note

Those tokens are considered “declared” after the ones explicitely declared through the token() decorator. This may be important because of the disambiguation rules; see the notes for the token() decorator.

Litteral tokens may be named as well.

Repeat operators

A nonterminal in the right side of a production may be immediately followed by a repeat operator among “*”, “+” and ”?”, which have the same meaning as in regular expressions. Note that this is only syntactic sugar; under the hood additional productions are generated.

A -> B?

is equivalent to

A ->
A -> B

The semantic value is None if the empty production was applied, or the semantic value of B if the ‘A -> B’ production was applied.

A -> B*

is equivalent to

A ->
A -> L_B
L_B -> B
L_B -> L_B B

The semantic value is a list of semantic values for B. ‘+’ works the same way except for the empty production, so the list cannot be empty.

Additionally, the ‘*’ and ‘+’ operators may include a separator specification, which is a symbol name or litteral token between parens:

A -> B+("|")

is equivalent to

A -> L_B
L_B -> B
L_B -> L_B "|" B

The semantic value is still a list of B values; there is no way to get the values for the separators.

Wrapping it up

Fully functional parser for a four-operations integer calculator:

@leftAssoc('+', '-')
@leftAssoc('*', '/')
class Parser(LRParser, ReLexer):
    @token('[1-9][0-9]*')
    def number(self, tok):
        tok.value = int(tok.value)

    @production('E -> number<n>')
    def litteral(self, n):
        return n

    @production('E -> "-" E<val>', priority='*')
    def minus(self, val):
        return -val

    @production('E -> "(" E<val> ")"')
    def paren(self, val):
        return val

    @production('E -> E<left> "+"<op> E<right>')
    @production('E -> E<left> "-"<op> E<right>')
    @production('E -> E<left> "*"<op> E<right>')
    @production('E -> E<left> "/"<op> E<right>')
    def binaryop(self, left, op, right):
        return {
            '+': operator.add,
            '-': operator.sub,
            '*': operator.mul,
            '/': operator.floordiv
            }[op](left, right)

Parsing lists of integers separated by commas:

class Parser(LRParser, ReLexer):
    @token('[1-9][0-9]*')
    def number(self, tok):
        tok.value = int(tok.value)
    @production('LIST -> number*(",")<values>')
    def integer_list(self, values):
        print('Values are: %s' % values)

Conflict resolution rules

Conflict resolution rules are the same as those used by Yacc/Bison. A shift/reduce conflict is resolved by choosing to shift. A reduce/reduce conflict is resolved by choosing the reduction associated with the first declared production. leftAssoc(), rightAssoc(), nonAssoc() and the priority argument to production() allows you to explicitely disambiguate.

Asynchronous lexer/parser

The AsyncLexer and AsyncLRParser classes allow you to parse an input stream asynchronously. Since this uses the new asynchronous method syntax introduced in Python 3.5, it’s only available with this version of Python.

The basic idea is that the production methods are asynchronous. Feed the input stream one byte/char at a time by awaiting on AsyncLexer.asyncFeed(). When a token has been recognized unambiguously, this will in turn await on AsyncParser.asyncNewToken(). Semantic actions may then be awaited on as a result.

The samples directory contains the following example of an asynchronous parser:

#!/usr/bin/env python
# -*- coding: UTF-8 -*-

"""

Four operations calculator, asynchronous. Due to various buffering
problems you probably won't see what's the point unless you force
stdin to be noninteractive, e.g.

$ echo '3*4+6' | python3 ./async_calc.py

"""

import six, operator, os, asyncio, sys, codecs

from ptk.lexer import token, AsyncLexer, EOF
from ptk.parser import production, leftAssoc, AsyncLRParser, ParseError


@leftAssoc('+', '-')
@leftAssoc('*', '/')
class Parser(AsyncLRParser, AsyncLexer):
    async def asyncNewSentence(self, result):
        six.print_('== Result:', result)

    # Lexer
    def ignore(self, char):
        return char in [' ', '\t']

    @token(r'[1-9][0-9]*')
    def number(self, tok):
        tok.value = int(tok.value)

    # Parser

    @production('E -> "-" E<value>', priority='*')
    async def minus(self, value):
        six.print_('== Neg: - %d' % value)
        return -value

    @production('E -> "(" E<value> ")"')
    async def paren(self, value):
        return value

    @production('E -> number<number>')
    async def litteral(self, number):
        return number

    @production('E -> E<left> "+"<op> E<right>')
    @production('E -> E<left> "-"<op> E<right>')
    @production('E -> E<left> "*"<op> E<right>')
    @production('E -> E<left> "/"<op> E<right>')
    async def binaryop(self, left, op, right):
        six.print_('Binary operation: %s %s %s' % (left, op, right))
        return {
            '+': operator.add,
            '-': operator.sub,
            '*': operator.mul,
            '/': operator.floordiv
            }[op](left, right)


async def main():
    reader = asyncio.StreamReader()
    await asyncio.get_event_loop().connect_read_pipe(lambda: asyncio.StreamReaderProtocol(reader), sys.stdin)
    decoder = codecs.getincrementaldecoder('utf_8')()

    parser = Parser()

    while True:
        byte = await reader.read(1)
        if not byte:
            break
        char = decoder.decode(byte)
        if char:
            if char == '\n':
                char = EOF
            else:
                six.print_('Input char: "%s"' % repr(char))
            await parser.asyncFeed(char)


loop = asyncio.get_event_loop()
loop.run_until_complete(main())

Asynchronous lexer/parser using Deferreds

The DeferredLexer and DeferredLRParser work the same as AsyncLexer and AsyncLRParser, but use Twisted’s Deferred objects instead of Python 3.5-like asynchronous methods. The special methods are called DeferredLexer.deferNewToken`and :py:func:`DeferredLRParser.deferNewSentence() and must return Deferred instances. Semantic actions can return either Deferred instances or regular values. See the defer_calc.py sample for details.