|
|
"""Test lexers module"""
|
|
|
|
|
|
# Copyright (c) IPython Development Team.
|
|
|
# Distributed under the terms of the Modified BSD License.
|
|
|
|
|
|
from unittest import TestCase
|
|
|
from pygments import __version__ as pygments_version
|
|
|
from pygments.token import Token
|
|
|
from pygments.lexers import BashLexer
|
|
|
|
|
|
from .. import lexers
|
|
|
|
|
|
pyg214 = tuple(int(x) for x in pygments_version.split(".")[:2]) >= (2, 14)
|
|
|
|
|
|
|
|
|
class TestLexers(TestCase):
|
|
|
"""Collection of lexers tests"""
|
|
|
def setUp(self):
|
|
|
self.lexer = lexers.IPythonLexer()
|
|
|
self.bash_lexer = BashLexer()
|
|
|
|
|
|
def testIPythonLexer(self):
|
|
|
fragment = '!echo $HOME\n'
|
|
|
bash_tokens = [
|
|
|
(Token.Operator, '!'),
|
|
|
]
|
|
|
bash_tokens.extend(self.bash_lexer.get_tokens(fragment[1:]))
|
|
|
ipylex_token = list(self.lexer.get_tokens(fragment))
|
|
|
assert bash_tokens[:-1] == ipylex_token[:-1]
|
|
|
|
|
|
fragment_2 = "!" + fragment
|
|
|
tokens_2 = [
|
|
|
(Token.Operator, '!!'),
|
|
|
] + bash_tokens[1:]
|
|
|
assert tokens_2[:-1] == list(self.lexer.get_tokens(fragment_2))[:-1]
|
|
|
|
|
|
fragment_2 = '\t %%!\n' + fragment[1:]
|
|
|
tokens_2 = [
|
|
|
(Token.Text, '\t '),
|
|
|
(Token.Operator, '%%!'),
|
|
|
(Token.Text, '\n'),
|
|
|
] + bash_tokens[1:]
|
|
|
assert tokens_2 == list(self.lexer.get_tokens(fragment_2))
|
|
|
|
|
|
fragment_2 = 'x = ' + fragment
|
|
|
tokens_2 = [
|
|
|
(Token.Name, 'x'),
|
|
|
(Token.Text, ' '),
|
|
|
(Token.Operator, '='),
|
|
|
(Token.Text, ' '),
|
|
|
] + bash_tokens
|
|
|
assert tokens_2[:-1] == list(self.lexer.get_tokens(fragment_2))[:-1]
|
|
|
|
|
|
fragment_2 = 'x, = ' + fragment
|
|
|
tokens_2 = [
|
|
|
(Token.Name, 'x'),
|
|
|
(Token.Punctuation, ','),
|
|
|
(Token.Text, ' '),
|
|
|
(Token.Operator, '='),
|
|
|
(Token.Text, ' '),
|
|
|
] + bash_tokens
|
|
|
assert tokens_2[:-1] == list(self.lexer.get_tokens(fragment_2))[:-1]
|
|
|
|
|
|
fragment_2 = 'x, = %sx ' + fragment[1:]
|
|
|
tokens_2 = [
|
|
|
(Token.Name, 'x'),
|
|
|
(Token.Punctuation, ','),
|
|
|
(Token.Text, ' '),
|
|
|
(Token.Operator, '='),
|
|
|
(Token.Text, ' '),
|
|
|
(Token.Operator, '%'),
|
|
|
(Token.Keyword, 'sx'),
|
|
|
(Token.Text, ' '),
|
|
|
] + bash_tokens[1:]
|
|
|
if tokens_2[7] == (Token.Text, " ") and pyg214: # pygments 2.14+
|
|
|
tokens_2[7] = (Token.Text.Whitespace, " ")
|
|
|
assert tokens_2[:-1] == list(self.lexer.get_tokens(fragment_2))[:-1]
|
|
|
|
|
|
fragment_2 = 'f = %R function () {}\n'
|
|
|
tokens_2 = [
|
|
|
(Token.Name, 'f'),
|
|
|
(Token.Text, ' '),
|
|
|
(Token.Operator, '='),
|
|
|
(Token.Text, ' '),
|
|
|
(Token.Operator, '%'),
|
|
|
(Token.Keyword, 'R'),
|
|
|
(Token.Text, ' function () {}\n'),
|
|
|
]
|
|
|
assert tokens_2 == list(self.lexer.get_tokens(fragment_2))
|
|
|
|
|
|
fragment_2 = '\t%%xyz\n$foo\n'
|
|
|
tokens_2 = [
|
|
|
(Token.Text, '\t'),
|
|
|
(Token.Operator, '%%'),
|
|
|
(Token.Keyword, 'xyz'),
|
|
|
(Token.Text, '\n$foo\n'),
|
|
|
]
|
|
|
assert tokens_2 == list(self.lexer.get_tokens(fragment_2))
|
|
|
|
|
|
fragment_2 = '%system?\n'
|
|
|
tokens_2 = [
|
|
|
(Token.Operator, '%'),
|
|
|
(Token.Keyword, 'system'),
|
|
|
(Token.Operator, '?'),
|
|
|
(Token.Text, '\n'),
|
|
|
]
|
|
|
assert tokens_2[:-1] == list(self.lexer.get_tokens(fragment_2))[:-1]
|
|
|
|
|
|
fragment_2 = 'x != y\n'
|
|
|
tokens_2 = [
|
|
|
(Token.Name, 'x'),
|
|
|
(Token.Text, ' '),
|
|
|
(Token.Operator, '!='),
|
|
|
(Token.Text, ' '),
|
|
|
(Token.Name, 'y'),
|
|
|
(Token.Text, '\n'),
|
|
|
]
|
|
|
assert tokens_2[:-1] == list(self.lexer.get_tokens(fragment_2))[:-1]
|
|
|
|
|
|
fragment_2 = ' ?math.sin\n'
|
|
|
tokens_2 = [
|
|
|
(Token.Text, ' '),
|
|
|
(Token.Operator, '?'),
|
|
|
(Token.Text, 'math.sin'),
|
|
|
(Token.Text, '\n'),
|
|
|
]
|
|
|
assert tokens_2[:-1] == list(self.lexer.get_tokens(fragment_2))[:-1]
|
|
|
|
|
|
fragment = ' *int*?\n'
|
|
|
tokens = [
|
|
|
(Token.Text, ' *int*'),
|
|
|
(Token.Operator, '?'),
|
|
|
(Token.Text, '\n'),
|
|
|
]
|
|
|
assert tokens == list(self.lexer.get_tokens(fragment))
|
|
|
|
|
|
fragment = '%%writefile -a foo.py\nif a == b:\n pass'
|
|
|
tokens = [
|
|
|
(Token.Operator, '%%writefile'),
|
|
|
(Token.Text, ' -a foo.py\n'),
|
|
|
(Token.Keyword, 'if'),
|
|
|
(Token.Text, ' '),
|
|
|
(Token.Name, 'a'),
|
|
|
(Token.Text, ' '),
|
|
|
(Token.Operator, '=='),
|
|
|
(Token.Text, ' '),
|
|
|
(Token.Name, 'b'),
|
|
|
(Token.Punctuation, ':'),
|
|
|
(Token.Text, '\n'),
|
|
|
(Token.Text, ' '),
|
|
|
(Token.Keyword, 'pass'),
|
|
|
(Token.Text, '\n'),
|
|
|
]
|
|
|
if tokens[10] == (Token.Text, "\n") and pyg214: # pygments 2.14+
|
|
|
tokens[10] = (Token.Text.Whitespace, "\n")
|
|
|
assert tokens[:-1] == list(self.lexer.get_tokens(fragment))[:-1]
|
|
|
|
|
|
fragment = '%%timeit\nmath.sin(0)'
|
|
|
tokens = [
|
|
|
(Token.Operator, '%%timeit\n'),
|
|
|
(Token.Name, 'math'),
|
|
|
(Token.Operator, '.'),
|
|
|
(Token.Name, 'sin'),
|
|
|
(Token.Punctuation, '('),
|
|
|
(Token.Literal.Number.Integer, '0'),
|
|
|
(Token.Punctuation, ')'),
|
|
|
(Token.Text, '\n'),
|
|
|
]
|
|
|
|
|
|
fragment = '%%HTML\n<div>foo</div>'
|
|
|
tokens = [
|
|
|
(Token.Operator, '%%HTML'),
|
|
|
(Token.Text, '\n'),
|
|
|
(Token.Punctuation, '<'),
|
|
|
(Token.Name.Tag, 'div'),
|
|
|
(Token.Punctuation, '>'),
|
|
|
(Token.Text, 'foo'),
|
|
|
(Token.Punctuation, '<'),
|
|
|
(Token.Punctuation, '/'),
|
|
|
(Token.Name.Tag, 'div'),
|
|
|
(Token.Punctuation, '>'),
|
|
|
(Token.Text, '\n'),
|
|
|
]
|
|
|
assert tokens == list(self.lexer.get_tokens(fragment))
|
|
|
|