mirror of
https://github.com/jtesta/ssh-audit.git
synced 2024-11-22 10:31:41 +01:00
12f811cb5c
* Remove `native text` converter This was only necessary with Python 2. After Python 2 removal, both functions `to_ntext` and `to_utext` exactly did the same. modified: ssh-audit.py modified: test/test_utils.py * Rename `to_utext` to `to_text` ... as in Python 3 there is only text (and bytes). modified: ssh-audit.py modified: test/test_utils.py
70 lines
2.7 KiB
Python
70 lines
2.7 KiB
Python
import pytest
|
|
|
|
|
|
# pylint: disable=attribute-defined-outside-init
|
|
class TestUtils:
|
|
@pytest.fixture(autouse=True)
|
|
def init(self, ssh_audit):
|
|
self.utils = ssh_audit.Utils
|
|
|
|
def test_to_bytes(self):
|
|
assert self.utils.to_bytes(b'fran\xc3\xa7ais') == b'fran\xc3\xa7ais'
|
|
assert self.utils.to_bytes('fran\xe7ais') == b'fran\xc3\xa7ais'
|
|
# other
|
|
with pytest.raises(TypeError):
|
|
self.utils.to_bytes(123)
|
|
|
|
def test_to_text(self):
|
|
assert self.utils.to_text(b'fran\xc3\xa7ais') == 'fran\xe7ais'
|
|
assert self.utils.to_text('fran\xe7ais') == 'fran\xe7ais'
|
|
# other
|
|
with pytest.raises(TypeError):
|
|
self.utils.to_text(123)
|
|
|
|
def test_is_ascii(self):
|
|
assert self.utils.is_ascii('francais') is True
|
|
assert self.utils.is_ascii('fran\xe7ais') is False
|
|
# other
|
|
assert self.utils.is_ascii(123) is False
|
|
|
|
def test_to_ascii(self):
|
|
assert self.utils.to_ascii('francais') == 'francais'
|
|
assert self.utils.to_ascii('fran\xe7ais') == 'fran?ais'
|
|
assert self.utils.to_ascii('fran\xe7ais', 'ignore') == 'franais'
|
|
with pytest.raises(TypeError):
|
|
self.utils.to_ascii(123)
|
|
|
|
def test_is_print_ascii(self):
|
|
assert self.utils.is_print_ascii('francais') is True
|
|
assert self.utils.is_print_ascii('francais\n') is False
|
|
assert self.utils.is_print_ascii('fran\xe7ais') is False
|
|
# other
|
|
assert self.utils.is_print_ascii(123) is False
|
|
|
|
def test_to_print_ascii(self):
|
|
assert self.utils.to_print_ascii('francais') == 'francais'
|
|
assert self.utils.to_print_ascii('francais\n') == 'francais?'
|
|
assert self.utils.to_print_ascii('fran\xe7ais') == 'fran?ais'
|
|
assert self.utils.to_print_ascii('fran\xe7ais\n') == 'fran?ais?'
|
|
assert self.utils.to_print_ascii('fran\xe7ais', 'ignore') == 'franais'
|
|
assert self.utils.to_print_ascii('fran\xe7ais\n', 'ignore') == 'franais'
|
|
with pytest.raises(TypeError):
|
|
self.utils.to_print_ascii(123)
|
|
|
|
def test_ctoi(self):
|
|
assert self.utils.ctoi(123) == 123
|
|
assert self.utils.ctoi('ABC') == 65
|
|
|
|
def test_parse_int(self):
|
|
assert self.utils.parse_int(123) == 123
|
|
assert self.utils.parse_int('123') == 123
|
|
assert self.utils.parse_int(-123) == -123
|
|
assert self.utils.parse_int('-123') == -123
|
|
assert self.utils.parse_int('abc') == 0
|
|
|
|
def test_unique_seq(self):
|
|
assert self.utils.unique_seq((1, 2, 2, 3, 3, 3)) == (1, 2, 3)
|
|
assert self.utils.unique_seq((3, 3, 3, 2, 2, 1)) == (3, 2, 1)
|
|
assert self.utils.unique_seq([1, 2, 2, 3, 3, 3]) == [1, 2, 3]
|
|
assert self.utils.unique_seq([3, 3, 3, 2, 2, 1]) == [3, 2, 1]
|