##// END OF EJS Templates
MAINT: remove support and testing on Python 3.8 (#14023)...
MAINT: remove support and testing on Python 3.8 (#14023) According to NEP 29 it is now time to remove support for Python 3.8 This commit removes support for Python 3.8 from the codebase, as well as removing the tests for Python 3.8 from the CI workflow. It also updates the `pyproject.toml` file to reflect the removal of Python 3.8 support.

File last commit:

r27098:7410af58
r28229:304d8237 merge
Show More
test_tokenutil.py
141 lines | 3.8 KiB | text/x-python | PythonLexer
"""Tests for tokenutil"""
# Copyright (c) IPython Development Team.
# Distributed under the terms of the Modified BSD License.
import pytest
from IPython.utils.tokenutil import token_at_cursor, line_at_cursor
def expect_token(expected, cell, cursor_pos):
token = token_at_cursor(cell, cursor_pos)
offset = 0
for line in cell.splitlines():
if offset + len(line) >= cursor_pos:
break
else:
offset += len(line)+1
column = cursor_pos - offset
line_with_cursor = "%s|%s" % (line[:column], line[column:])
assert token == expected, "Expected %r, got %r in: %r (pos %i)" % (
expected,
token,
line_with_cursor,
cursor_pos,
)
def test_simple():
cell = "foo"
for i in range(len(cell)):
expect_token("foo", cell, i)
def test_function():
cell = "foo(a=5, b='10')"
expected = 'foo'
# up to `foo(|a=`
for i in range(cell.find('a=') + 1):
expect_token("foo", cell, i)
# find foo after `=`
for i in [cell.find('=') + 1, cell.rfind('=') + 1]:
expect_token("foo", cell, i)
# in between `5,|` and `|b=`
for i in range(cell.find(','), cell.find('b=')):
expect_token("foo", cell, i)
def test_multiline():
cell = '\n'.join([
'a = 5',
'b = hello("string", there)'
])
expected = 'hello'
start = cell.index(expected) + 1
for i in range(start, start + len(expected)):
expect_token(expected, cell, i)
expected = 'hello'
start = cell.index(expected) + 1
for i in range(start, start + len(expected)):
expect_token(expected, cell, i)
def test_multiline_token():
cell = '\n'.join([
'"""\n\nxxxxxxxxxx\n\n"""',
'5, """',
'docstring',
'multiline token',
'""", [',
'2, 3, "complicated"]',
'b = hello("string", there)'
])
expected = 'hello'
start = cell.index(expected) + 1
for i in range(start, start + len(expected)):
expect_token(expected, cell, i)
expected = 'hello'
start = cell.index(expected) + 1
for i in range(start, start + len(expected)):
expect_token(expected, cell, i)
def test_nested_call():
cell = "foo(bar(a=5), b=10)"
expected = 'foo'
start = cell.index('bar') + 1
for i in range(start, start + 3):
expect_token(expected, cell, i)
expected = 'bar'
start = cell.index('a=')
for i in range(start, start + 3):
expect_token(expected, cell, i)
expected = 'foo'
start = cell.index(')') + 1
for i in range(start, len(cell)-1):
expect_token(expected, cell, i)
def test_attrs():
cell = "a = obj.attr.subattr"
expected = 'obj'
idx = cell.find('obj') + 1
for i in range(idx, idx + 3):
expect_token(expected, cell, i)
idx = cell.find('.attr') + 2
expected = 'obj.attr'
for i in range(idx, idx + 4):
expect_token(expected, cell, i)
idx = cell.find('.subattr') + 2
expected = 'obj.attr.subattr'
for i in range(idx, len(cell)):
expect_token(expected, cell, i)
def test_line_at_cursor():
cell = ""
(line, offset) = line_at_cursor(cell, cursor_pos=11)
assert line == ""
assert offset == 0
# The position after a newline should be the start of the following line.
cell = "One\nTwo\n"
(line, offset) = line_at_cursor(cell, cursor_pos=4)
assert line == "Two\n"
assert offset == 4
# The end of a cell should be on the last line
cell = "pri\npri"
(line, offset) = line_at_cursor(cell, cursor_pos=7)
assert line == "pri"
assert offset == 4
@pytest.mark.parametrize(
"c, token",
zip(
list(range(16, 22)) + list(range(22, 28)),
["int"] * (22 - 16) + ["map"] * (28 - 22),
),
)
def test_multiline_statement(c, token):
cell = """a = (1,
3)
int()
map()
"""
expect_token(token, cell, c)