CODEWITHSUNDEEP
pythontransformlexical-analysis
Philip Reynolds
Philip Reynolds

Reputation: 9392

Python - lexical analysis and tokenization

I'm looking to speed along my discovery process here quite a bit, as this is my first venture into the world of lexical analysis. Maybe this is even the wrong path. First, I'll describe my problem:

I've got very large properties files (in the order of 1,000 properties), which when distilled, are really just about 15 important properties and the rest can be generated or rarely ever change.

So, for example:

general {
  name = myname
  ip = 127.0.0.1
}

component1 {
   key = value
   foo = bar
}

This is the type of format I want to create to tokenize something like:

property.${general.name}blah.home.directory = /blah
property.${general.name}.ip = ${general.ip}
property.${component1}.ip = ${general.ip}
property.${component1}.foo = ${component1.foo}

into

property.mynameblah.home.directory = /blah
property.myname.ip = 127.0.0.1
property.component1.ip = 127.0.0.1
property.component1.foo = bar

Lexical analysis and tokenization sounds like my best route, but this is a very simple form of it. It's a simple grammar, a simple substitution and I'd like to make sure that I'm not bringing a sledgehammer to knock in a nail.

I could create my own lexer and tokenizer, or ANTlr is a possibility, but I don't like re-inventing the wheel and ANTlr sounds like overkill.

I'm not familiar with compiler techniques, so pointers in the right direction & code would be most appreciated.

Note: I can change the input format.

Upvotes: 12

Views: 15414

Answers (5)

Matt Anderson
Matt Anderson

Reputation: 19769

There's an excellent article on Using Regular Expressions for Lexical Analysis at effbot.org.

Adapting the tokenizer to your problem:

import re

token_pattern = r"""
(?P<identifier>[a-zA-Z_][a-zA-Z0-9_]*)
|(?P<integer>[0-9]+)
|(?P<dot>\.)
|(?P<open_variable>[$][{])
|(?P<open_curly>[{])
|(?P<close_curly>[}])
|(?P<newline>\n)
|(?P<whitespace>\s+)
|(?P<equals>[=])
|(?P<slash>[/])
"""

token_re = re.compile(token_pattern, re.VERBOSE)

class TokenizerException(Exception): pass

def tokenize(text):
    pos = 0
    while True:
        m = token_re.match(text, pos)
        if not m: break
        pos = m.end()
        tokname = m.lastgroup
        tokvalue = m.group(tokname)
        yield tokname, tokvalue
    if pos != len(text):
        raise TokenizerException('tokenizer stopped at pos %r of %r' % (
            pos, len(text)))

To test it, we do:

stuff = r'property.${general.name}.ip = ${general.ip}'
stuff2 = r'''
general {
  name = myname
  ip = 127.0.0.1
}
'''

print ' stuff '.center(60, '=')
for tok in tokenize(stuff):
    print tok

print ' stuff2 '.center(60, '=')
for tok in tokenize(stuff2):
    print tok

for:

========================== stuff ===========================
('identifier', 'property')
('dot', '.')
('open_variable', '${')
('identifier', 'general')
('dot', '.')
('identifier', 'name')
('close_curly', '}')
('dot', '.')
('identifier', 'ip')
('whitespace', ' ')
('equals', '=')
('whitespace', ' ')
('open_variable', '${')
('identifier', 'general')
('dot', '.')
('identifier', 'ip')
('close_curly', '}')
========================== stuff2 ==========================
('newline', '\n')
('identifier', 'general')
('whitespace', ' ')
('open_curly', '{')
('newline', '\n')
('whitespace', '  ')
('identifier', 'name')
('whitespace', ' ')
('equals', '=')
('whitespace', ' ')
('identifier', 'myname')
('newline', '\n')
('whitespace', '  ')
('identifier', 'ip')
('whitespace', ' ')
('equals', '=')
('whitespace', ' ')
('integer', '127')
('dot', '.')
('integer', '0')
('dot', '.')
('integer', '0')
('dot', '.')
('integer', '1')
('newline', '\n')
('close_curly', '}')
('newline', '\n')

Upvotes: 14

Dmitry Kochkin
Dmitry Kochkin

Reputation: 993

The syntax you provide seems similar to Mako templates engine. I think you could give it a try, it's rather simple API.

Upvotes: 1

Kaleb Pederson
Kaleb Pederson

Reputation: 46479

A simple DFA works well for this. You only need a few states:

  1. Looking for ${
  2. Seen ${ looking for at least one valid character forming the name
  3. Seen at least one valid name character, looking for more name characters or }.

If the properties file is order agnostic, you might want a two pass processor to verify that each name resolves correctly.

Of course, you then need to write the substitution code, but once you have a list of all the names used, the simplest possible implementation is a find/replace on ${name} with its corresponding value.

Upvotes: 4

zdav
zdav

Reputation: 2748

For as simple as your format seems to be, I think a full-on parser/lexer would be way overkill. Seems like a combination of regexes and string manipulation would do the trick.

Another idea is to change the file to something like json or xml and use an existing package.

Upvotes: 2

danben
danben

Reputation: 83250

If you can change the format of the input files, then you could use a parser for an existing format, such as JSON.

However, from your problem statement it sounds like that isn't the case. So if you want to create a custom lexer and parser, use PLY (Python Lex/Yacc). It is easy to use and works the same as lex/yacc.

Here is a link to an example of a calculator built using PLY. Note that everything starting with t_ is a lexer rule - defining a valid token - and everything starting with p_ is a parser rule that defines a production of the grammar.

Upvotes: 1

Related Questions