rana-cli/wakatime/languages/python.py
2015-09-16 12:59:52 -07:00

120 lines
4.2 KiB
Python

# -*- coding: utf-8 -*-
"""
wakatime.languages.python
~~~~~~~~~~~~~~~~~~~~~~~~~
Parse dependencies from Python code.
:copyright: (c) 2014 Alan Hamlett.
:license: BSD, see LICENSE for more details.
"""
from . import TokenParser
from ..compat import u
class PythonParser(TokenParser):
state = None
parens = 0
nonpackage = False
def parse(self, tokens=[]):
if not tokens and not self.tokens:
self.tokens = self._extract_tokens()
for index, token, content in self.tokens:
self._process_token(token, content)
return self.dependencies
def _process_token(self, token, content):
if u(token).split('.')[-1] == 'Namespace':
self._process_namespace(token, content)
elif u(token).split('.')[-1] == 'Name':
self._process_name(token, content)
elif u(token).split('.')[-1] == 'Word':
self._process_word(token, content)
elif u(token).split('.')[-1] == 'Operator':
self._process_operator(token, content)
elif u(token).split('.')[-1] == 'Punctuation':
self._process_punctuation(token, content)
elif u(token).split('.')[-1] == 'Text':
self._process_text(token, content)
else:
self._process_other(token, content)
def _process_namespace(self, token, content):
if self.state is None:
self.state = content
else:
if content == 'as':
self.nonpackage = True
else:
self._process_import(token, content)
def _process_name(self, token, content):
if self.state is not None:
if self.nonpackage:
self.nonpackage = False
else:
if self.state == 'from':
self.append(content, truncate=True, truncate_to=1)
if self.state == 'from-2' and content != 'import':
self.append(content, truncate=True, truncate_to=1)
elif self.state == 'import':
self.append(content, truncate=True, truncate_to=1)
elif self.state == 'import-2':
self.append(content, truncate=True, truncate_to=1)
else:
self.state = None
def _process_word(self, token, content):
if self.state is not None:
if self.nonpackage:
self.nonpackage = False
else:
if self.state == 'from':
self.append(content, truncate=True, truncate_to=1)
if self.state == 'from-2' and content != 'import':
self.append(content, truncate=True, truncate_to=1)
elif self.state == 'import':
self.append(content, truncate=True, truncate_to=1)
elif self.state == 'import-2':
self.append(content, truncate=True, truncate_to=1)
else:
self.state = None
def _process_operator(self, token, content):
if self.state is not None:
if content == '.':
self.nonpackage = True
def _process_punctuation(self, token, content):
if content == '(':
self.parens += 1
elif content == ')':
self.parens -= 1
self.nonpackage = False
def _process_text(self, token, content):
if self.state is not None:
if content == "\n" and self.parens == 0:
self.state = None
self.nonpackage = False
def _process_other(self, token, content):
pass
def _process_import(self, token, content):
if not self.nonpackage:
if self.state == 'from':
self.append(content, truncate=True, truncate_to=1)
self.state = 'from-2'
elif self.state == 'from-2' and content != 'import':
self.append(content, truncate=True, truncate_to=1)
elif self.state == 'import':
self.append(content, truncate=True, truncate_to=1)
self.state = 'import-2'
elif self.state == 'import-2':
self.append(content, truncate=True, truncate_to=1)
else:
self.state = None
self.nonpackage = False