C[omp]ute

From: andrew cooke <andrew@...>

Date: Thu, 27 May 2010 08:16:19 -0400

I have been working on a regular expression package for Python - a pure Python
implementation that will, eventually, be integrated into Lepl.  So far parsing
works (I can now parse almost all of Python's regular expression syntax) and I
have just got a very basic interpreter (effectively NFA) based engine running
some basic matches.  The following tests pass:

class VisitorTest(TestCase):
    
    def test_string(self):
        assert Visitor(Unicode(), parse('abc'), 'abc')
        assert Visitor(Unicode(), parse('abc'), 'abcd')
        assert not Visitor(Unicode(), parse('abc'), 'ab')
        
    def test_dot(self):
        assert Visitor(Unicode(), parse('a.c'), 'abc')
        assert Visitor(Unicode(), parse('...'), 'abcd')
        assert not Visitor(Unicode(), parse('...'), 'ab')

As I hoped, the parsing has been by far the hardest part.  Actually writing an
implementation given the parsed graph and visitor interfaces is pretty easy.

The parser, as implied above, generates a directed graph of nodes that reflect
the different operations supported.  This could then be "compiled" by an
implementation or simply used directly.  For direct use it supports a basic
visitor pattern, where each node implements .visit(visitor, state) and then
"calls back" on the visitor.  So the visitor callback methods are equivalent
to the different bytecodes that would be used in compilation:

class Visitor(object):
    def string(self, next, text, state=None):
    def character(self, next, state=None):
    def start_group(self, next, number, state=None):
    def end_group(self, next, number, state=None):
    def group_reference(self, next, number, state=None):
    def conditional(self, next, number, state=None):
    def split(self, next, state=None):
    def match(self, state=None):
    def dot(self, next, multiline, state=None):
    def start_of_line(self, next, multiline, state=None):
    def end_of_line(self, next, multiline, state=None):
    def lookahead(self, next, sense, forwards, state=None):
    def repeat(self, next, begin, end, state=None):
    def word_boundary(self, next, inverted, state=None):
    def digit(self, next, inverted, state=None):
    def space(self, next, inverted, state=None):
    def word(self, next, inverted, state=None):

(Note that "repeat" is replaced by explicit expansion in the parser if the
appropriate options are set; also, expansion of character classes is handled
by an "alphabet" that is not otherwise explicit here).

Andrew