mirror of
https://github.com/mikf/gallery-dl.git
synced 2024-11-22 18:53:21 +01:00
79c01ec7ae
J joins list elements by calling <separator>.join(list): Example: {f:J - /} -> "a - b - c" (if "f" is ["a", "b", "c"])
315 lines
10 KiB
Python
315 lines
10 KiB
Python
#!/usr/bin/env python3
|
|
# -*- coding: utf-8 -*-
|
|
|
|
# Copyright 2015-2019 Mike Fährmann
|
|
#
|
|
# This program is free software; you can redistribute it and/or modify
|
|
# it under the terms of the GNU General Public License version 2 as
|
|
# published by the Free Software Foundation.
|
|
|
|
import unittest
|
|
import gallery_dl.util as util
|
|
import gallery_dl.exception as exception
|
|
import sys
|
|
import random
|
|
import string
|
|
|
|
|
|
class TestRange(unittest.TestCase):
|
|
|
|
def test_parse_range(self, f=util.RangePredicate.parse_range):
|
|
self.assertEqual(
|
|
f(""),
|
|
[])
|
|
self.assertEqual(
|
|
f("1-2"),
|
|
[(1, 2)])
|
|
self.assertEqual(
|
|
f("-"),
|
|
[(1, sys.maxsize)])
|
|
self.assertEqual(
|
|
f("-2,4,6-8,10-"),
|
|
[(1, 2), (4, 4), (6, 8), (10, sys.maxsize)])
|
|
self.assertEqual(
|
|
f(" - 3 , 4- 4, 2-6"),
|
|
[(1, 3), (4, 4), (2, 6)])
|
|
|
|
def test_optimize_range(self, f=util.RangePredicate.optimize_range):
|
|
self.assertEqual(
|
|
f([]),
|
|
[])
|
|
self.assertEqual(
|
|
f([(2, 4)]),
|
|
[(2, 4)])
|
|
self.assertEqual(
|
|
f([(2, 4), (6, 8), (10, 12)]),
|
|
[(2, 4), (6, 8), (10, 12)])
|
|
self.assertEqual(
|
|
f([(2, 4), (4, 6), (5, 8)]),
|
|
[(2, 8)])
|
|
self.assertEqual(
|
|
f([(1, 1), (2, 2), (3, 6), (8, 9)]),
|
|
[(1, 6), (8, 9)])
|
|
|
|
|
|
class TestPredicate(unittest.TestCase):
|
|
|
|
def test_range_predicate(self):
|
|
dummy = None
|
|
|
|
pred = util.RangePredicate(" - 3 , 4- 4, 2-6")
|
|
for i in range(6):
|
|
self.assertTrue(pred(dummy, dummy))
|
|
with self.assertRaises(exception.StopExtraction):
|
|
bool(pred(dummy, dummy))
|
|
|
|
pred = util.RangePredicate("1, 3, 5")
|
|
self.assertTrue(pred(dummy, dummy))
|
|
self.assertFalse(pred(dummy, dummy))
|
|
self.assertTrue(pred(dummy, dummy))
|
|
self.assertFalse(pred(dummy, dummy))
|
|
self.assertTrue(pred(dummy, dummy))
|
|
with self.assertRaises(exception.StopExtraction):
|
|
bool(pred(dummy, dummy))
|
|
|
|
pred = util.RangePredicate("")
|
|
with self.assertRaises(exception.StopExtraction):
|
|
bool(pred(dummy, dummy))
|
|
|
|
def test_unique_predicate(self):
|
|
dummy = None
|
|
pred = util.UniquePredicate()
|
|
|
|
# no duplicates
|
|
self.assertTrue(pred("1", dummy))
|
|
self.assertTrue(pred("2", dummy))
|
|
self.assertFalse(pred("1", dummy))
|
|
self.assertFalse(pred("2", dummy))
|
|
self.assertTrue(pred("3", dummy))
|
|
self.assertFalse(pred("3", dummy))
|
|
|
|
# duplicates for "text:"
|
|
self.assertTrue(pred("text:123", dummy))
|
|
self.assertTrue(pred("text:123", dummy))
|
|
self.assertTrue(pred("text:123", dummy))
|
|
|
|
def test_build_predicate(self):
|
|
pred = util.build_predicate([])
|
|
self.assertIsInstance(pred, type(lambda: True))
|
|
|
|
pred = util.build_predicate([util.UniquePredicate()])
|
|
self.assertIsInstance(pred, util.UniquePredicate)
|
|
|
|
pred = util.build_predicate([util.UniquePredicate(),
|
|
util.UniquePredicate()])
|
|
self.assertIsInstance(pred, util.ChainPredicate)
|
|
|
|
|
|
class TestISO639_1(unittest.TestCase):
|
|
|
|
def test_code_to_language(self):
|
|
d = "default"
|
|
self._run_test(util.code_to_language, {
|
|
("en",): "English",
|
|
("FR",): "French",
|
|
("xx",): None,
|
|
("" ,): None,
|
|
(None,): None,
|
|
("en", d): "English",
|
|
("FR", d): "French",
|
|
("xx", d): d,
|
|
("" , d): d,
|
|
(None, d): d,
|
|
})
|
|
|
|
def test_language_to_code(self):
|
|
d = "default"
|
|
self._run_test(util.language_to_code, {
|
|
("English",): "en",
|
|
("fRENch",): "fr",
|
|
("xx",): None,
|
|
("" ,): None,
|
|
(None,): None,
|
|
("English", d): "en",
|
|
("fRENch", d): "fr",
|
|
("xx", d): d,
|
|
("" , d): d,
|
|
(None, d): d,
|
|
})
|
|
|
|
def _run_test(self, func, tests):
|
|
for args, result in tests.items():
|
|
self.assertEqual(func(*args), result)
|
|
|
|
|
|
class TestFormatter(unittest.TestCase):
|
|
|
|
kwdict = {
|
|
"a": "hElLo wOrLd",
|
|
"b": "äöü",
|
|
"l": ["a", "b", "c"],
|
|
"u": "%27%3C%20/%20%3E%27",
|
|
"name": "Name",
|
|
"title1": "Title",
|
|
"title2": "",
|
|
"title3": None,
|
|
"title4": 0,
|
|
}
|
|
|
|
def test_conversions(self):
|
|
self._run_test("{a!l}", "hello world")
|
|
self._run_test("{a!u}", "HELLO WORLD")
|
|
self._run_test("{a!c}", "Hello world")
|
|
self._run_test("{a!C}", "Hello World")
|
|
self._run_test("{a!U}", self.kwdict["a"])
|
|
self._run_test("{u!U}", "'< / >'")
|
|
self._run_test("{a!s}", self.kwdict["a"])
|
|
self._run_test("{a!r}", "'" + self.kwdict["a"] + "'")
|
|
self._run_test("{a!a}", "'" + self.kwdict["a"] + "'")
|
|
self._run_test("{b!a}", "'\\xe4\\xf6\\xfc'")
|
|
with self.assertRaises(KeyError):
|
|
self._run_test("{a!q}", "hello world")
|
|
|
|
def test_optional(self):
|
|
self._run_test("{name}{title1}", "NameTitle")
|
|
self._run_test("{name}{title1:?//}", "NameTitle")
|
|
self._run_test("{name}{title1:? **/''/}", "Name **Title''")
|
|
|
|
self._run_test("{name}{title2}", "Name")
|
|
self._run_test("{name}{title2:?//}", "Name")
|
|
self._run_test("{name}{title2:? **/''/}", "Name")
|
|
|
|
self._run_test("{name}{title3}", "NameNone")
|
|
self._run_test("{name}{title3:?//}", "Name")
|
|
self._run_test("{name}{title3:? **/''/}", "Name")
|
|
|
|
self._run_test("{name}{title4}", "Name0")
|
|
self._run_test("{name}{title4:?//}", "Name")
|
|
self._run_test("{name}{title4:? **/''/}", "Name")
|
|
|
|
def test_missing(self):
|
|
replacement = "None"
|
|
|
|
self._run_test("{missing}", replacement)
|
|
self._run_test("{missing.attr}", replacement)
|
|
self._run_test("{missing[key]}", replacement)
|
|
self._run_test("{missing:?a//}", "")
|
|
|
|
self._run_test("{name[missing]}", replacement)
|
|
self._run_test("{name[missing].attr}", replacement)
|
|
self._run_test("{name[missing][key]}", replacement)
|
|
self._run_test("{name[missing]:?a//}", "")
|
|
|
|
def test_missing_custom_default(self):
|
|
replacement = default = "foobar"
|
|
self._run_test("{missing}" , replacement, default)
|
|
self._run_test("{missing.attr}", replacement, default)
|
|
self._run_test("{missing[key]}", replacement, default)
|
|
self._run_test("{missing:?a//}", "a" + default, default)
|
|
|
|
def test_slicing(self):
|
|
v = self.kwdict["a"]
|
|
self._run_test("{a[1:10]}" , v[1:10])
|
|
self._run_test("{a[-10:-1]}", v[-10:-1])
|
|
self._run_test("{a[5:]}" , v[5:])
|
|
self._run_test("{a[50:]}", v[50:])
|
|
self._run_test("{a[:5]}" , v[:5])
|
|
self._run_test("{a[:50]}", v[:50])
|
|
self._run_test("{a[:]}" , v)
|
|
self._run_test("{a[1:10:2]}" , v[1:10:2])
|
|
self._run_test("{a[-10:-1:2]}", v[-10:-1:2])
|
|
self._run_test("{a[5::2]}" , v[5::2])
|
|
self._run_test("{a[50::2]}", v[50::2])
|
|
self._run_test("{a[:5:2]}" , v[:5:2])
|
|
self._run_test("{a[:50:2]}", v[:50:2])
|
|
self._run_test("{a[::]}" , v)
|
|
|
|
def test_maxlen(self):
|
|
v = self.kwdict["a"]
|
|
self._run_test("{a:L5/foo/}" , "foo")
|
|
self._run_test("{a:L50/foo/}", v)
|
|
self._run_test("{a:L50/foo/>50}", " " * 39 + v)
|
|
self._run_test("{a:L50/foo/>51}", "foo")
|
|
self._run_test("{a:Lab/foo/}", "foo")
|
|
|
|
def test_join(self):
|
|
self._run_test("{l:J}" , "abc")
|
|
self._run_test("{l:J,}" , "a,b,c")
|
|
self._run_test("{l:J,/}" , "a,b,c")
|
|
self._run_test("{l:J,/>20}" , " a,b,c")
|
|
self._run_test("{l:J - }" , "a - b - c")
|
|
self._run_test("{l:J - /}" , "a - b - c")
|
|
self._run_test("{l:J - />20}", " a - b - c")
|
|
|
|
self._run_test("{a:J/}" , self.kwdict["a"])
|
|
self._run_test("{a:J, /}" , ", ".join(self.kwdict["a"]))
|
|
|
|
def _run_test(self, format_string, result, default=None):
|
|
formatter = util.Formatter(format_string, default)
|
|
output = formatter.format_map(self.kwdict)
|
|
self.assertEqual(output, result, format_string)
|
|
|
|
|
|
class TestOther(unittest.TestCase):
|
|
|
|
def test_bencode(self):
|
|
self.assertEqual(util.bencode(0), "")
|
|
self.assertEqual(util.bencode(123), "123")
|
|
self.assertEqual(util.bencode(123, "01"), "1111011")
|
|
self.assertEqual(util.bencode(123, "BA"), "AAAABAA")
|
|
|
|
def test_bdecode(self):
|
|
self.assertEqual(util.bdecode(""), 0)
|
|
self.assertEqual(util.bdecode("123"), 123)
|
|
self.assertEqual(util.bdecode("1111011", "01"), 123)
|
|
self.assertEqual(util.bdecode("AAAABAA", "BA"), 123)
|
|
|
|
def test_bencode_bdecode(self):
|
|
for _ in range(100):
|
|
value = random.randint(0, 1000000)
|
|
for alphabet in ("01", "0123456789", string.ascii_letters):
|
|
result = util.bdecode(util.bencode(value, alphabet), alphabet)
|
|
self.assertEqual(result, value)
|
|
|
|
def test_advance(self):
|
|
items = range(5)
|
|
|
|
self.assertCountEqual(
|
|
util.advance(items, 0), items)
|
|
self.assertCountEqual(
|
|
util.advance(items, 3), range(3, 5))
|
|
self.assertCountEqual(
|
|
util.advance(items, 9), [])
|
|
self.assertCountEqual(
|
|
util.advance(util.advance(items, 1), 2), range(3, 5))
|
|
|
|
def test_raises(self):
|
|
self.assertRaises(Exception, util.raises(Exception()))
|
|
|
|
func = util.raises(ValueError(1))
|
|
self.assertRaises(Exception, func)
|
|
self.assertRaises(Exception, func)
|
|
self.assertRaises(Exception, func)
|
|
|
|
def test_combine_dict(self):
|
|
self.assertEqual(
|
|
util.combine_dict({}, {}),
|
|
{})
|
|
self.assertEqual(
|
|
util.combine_dict({1: 1, 2: 2}, {2: 4, 4: 8}),
|
|
{1: 1, 2: 4, 4: 8})
|
|
self.assertEqual(
|
|
util.combine_dict(
|
|
{1: {11: 22, 12: 24}, 2: {13: 26, 14: 28}},
|
|
{1: {11: 33, 13: 39}, 2: "str"}),
|
|
{1: {11: 33, 12: 24, 13: 39}, 2: "str"})
|
|
self.assertEqual(
|
|
util.combine_dict(
|
|
{1: {2: {3: {4: {"1": "a", "2": "b"}}}}},
|
|
{1: {2: {3: {4: {"1": "A", "3": "C"}}}}}),
|
|
{1: {2: {3: {4: {"1": "A", "2": "b", "3": "C"}}}}})
|
|
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main()
|