# -*- coding: utf-8 -*- """ Pygments RTF formatter tests ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ :copyright: Copyright 2006-2014 by the Pygments team, see AUTHORS. :license: BSD, see LICENSE for details. """ import unittest from string_asserts import StringTests from pygments.util import StringIO from pygments.formatters import RtfFormatter from pygments.lexers.special import TextLexer class RtfFormatterTest(StringTests, unittest.TestCase): foot = (r'\par' '\n' r'}') def _escape(self, string): return(string.replace("\n", r"\n")) def _build_message(self, *args, **kwargs): string = kwargs.get('string', None) t = self._escape(kwargs.get('t', '')) expected = self._escape(kwargs.get('expected', '')) result = self._escape(kwargs.get('result', '')) if string is None: string = (u"The expected output of '{t}'\n" u"\t\tShould be '{expected}'\n" u"\t\tActually outputs '{result}'\n" u"\t(WARNING: Partial Output of Result!)") end = -(len(self._escape(self.foot))) start = end-len(expected) return string.format(t=t, result = result[start:end], expected = expected) def format_rtf(self, t): tokensource = list(TextLexer().get_tokens(t)) fmt = RtfFormatter() buf = StringIO() fmt.format(tokensource, buf) result = buf.getvalue() buf.close() return result def test_rtf_header(self): t = u'' result = self.format_rtf(t) expected = r'{\rtf1\ansi\uc0' msg = (u"RTF documents are expected to start with '{expected}'\n" u"\t\tStarts intead with '{result}'\n" u"\t(WARNING: Partial Output of Result!)".format( expected = expected, result = result[:len(expected)])) self.assertStartsWith(result, expected, msg) def test_rtf_footer(self): t = u'' result = self.format_rtf(t) expected = self.foot msg = (u"RTF documents are expected to end with '{expected}'\n" u"\t\tEnds intead with '{result}'\n" u"\t(WARNING: Partial Output of Result!)".format( expected = self._escape(expected), result = self._escape(result[-len(expected):]))) self.assertEndsWith(result, expected, msg) def test_ascii_characters(self): t = u'a b c d ~' result = self.format_rtf(t) expected = (r'a b c d ~') if not result.endswith(self.foot): return(unittest.skip('RTF Footer incorrect')) msg = self._build_message(t=t, result=result, expected=expected) self.assertEndsWith(result, expected+self.foot, msg) def test_escape_characters(self): t = u'\ {{' result = self.format_rtf(t) expected = (r'\\ \{\{') if not result.endswith(self.foot): return(unittest.skip('RTF Footer incorrect')) msg = self._build_message(t=t, result=result, expected=expected) self.assertEndsWith(result, expected+self.foot, msg) def test_single_characters(self): t = u'â € ¤ каждой' result = self.format_rtf(t) expected = (r'{\u226} {\u8364} {\u164} ' r'{\u1082}{\u1072}{\u1078}{\u1076}{\u1086}{\u1081}') if not result.endswith(self.foot): return(unittest.skip('RTF Footer incorrect')) msg = self._build_message(t=t, result=result, expected=expected) self.assertEndsWith(result, expected+self.foot, msg) def test_double_characters(self): t = u'က 힣 ↕ ↕︎ 鼖' result = self.format_rtf(t) expected = (r'{\u4096} {\u55203} {\u8597} ' r'{\u8597}{\u65038} {\u55422}{\u56859}') if not result.endswith(self.foot): return(unittest.skip('RTF Footer incorrect')) msg = self._build_message(t=t, result=result, expected=expected) self.assertEndsWith(result, expected+self.foot, msg)