`url.join(...)` drops parts of the path if `url` does not end with `/`

[konstantin]

Describe the bug

I'm joining two parts of an URL:

1. One URL with the host and first part of the path
2. One URL with the second part of the path and query params

Expected behaviour: The result path is concatenated from the two parts.
Actual behaviour: The result path is concatenated but drops the last part of the first URL if the first URL does not endwith a "/"

To Reproduce

import pytest
from yarl import URL


@pytest.mark.parametrize(
    "first_url_part",
    [
        pytest.param(URL("https://example.inv/Foo/Bar"), id="w/o trailing slash"),
        pytest.param(URL("https://example.inv/") / "Foo" / "Bar", id="dynamically without trailing slash"),
        pytest.param(URL("https://example.inv/Foo/Bar/"), id="with trailing slash"),
        pytest.param(URL("https://example.inv/") / "Foo" / "Bar/", id="dynamically with trailing slash"),
    ],
)
def test_url_join(first_url_part: URL):
    assert isinstance(first_url_part, URL)
    second_url_part = URL("Baz/asd?xyz=123")
    actual = first_url_part.join(second_url_part)
    assert actual == URL("https://example.inv/Foo/Bar/Baz/asd?xyz=123")
    # for 'w/o trailing slash' case the assertion is not fulfilled
    # Expected :URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123')
    # Actual   :URL('https://example.inv/Foo/Baz/asd?xyz=123') # <-- why is the '/Bar/' part missing?

Expected behavior

The last part of the first URL path should not be dropped.

Test Output

========================= 3 failed, 1 passed in 0.13s =========================
FAILED                     [ 25%]
unittests\urltest.py:4 (test_url_join[w/o trailing slash])
URL('https://example.inv/Foo/Baz/asd?xyz=123') != URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123')

Expected :URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123')
Actual   :URL('https://example.inv/Foo/Baz/asd?xyz=123')
<Click to see difference>

first_url_part = URL('https://example.inv/Foo/Bar')

    @pytest.mark.parametrize(
        "first_url_part",
        [
            pytest.param(URL("https://example.inv/Foo/Bar"), id="w/o trailing slash"),
            pytest.param(URL("https://example.inv/") / "Foo" / "Bar", id="dynamically without trailing slash"),
            pytest.param(URL("https://example.inv/Foo/Bar/"), id="with trailing slash"),
            pytest.param(URL("https://example.inv/") / "Foo" / "Bar/", id="dynamically with trailing slash"),
        ],
    )
    def test_url_join(first_url_part: URL):
        assert isinstance(first_url_part, URL)
        second_url_part = URL("Baz/asd?xyz=123")
        actual = first_url_part.join(second_url_part)
>       assert actual == URL("https://example.inv/Foo/Bar/Baz/asd?xyz=123")
E       AssertionError: assert URL('https://example.inv/Foo/Baz/asd?xyz=123') == URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123')
E        +  where URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123') = URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123')

urltest.py:18: AssertionError
FAILED     [ 50%]
unittests\urltest.py:4 (test_url_join[dynamically without trailing slash])
URL('https://example.inv/Foo/Baz/asd?xyz=123') != URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123')

Expected :URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123')
Actual   :URL('https://example.inv/Foo/Baz/asd?xyz=123')
<Click to see difference>

first_url_part = URL('https://example.inv/Foo/Bar')

    @pytest.mark.parametrize(
        "first_url_part",
        [
            pytest.param(URL("https://example.inv/Foo/Bar"), id="w/o trailing slash"),
            pytest.param(URL("https://example.inv/") / "Foo" / "Bar", id="dynamically without trailing slash"),
            pytest.param(URL("https://example.inv/Foo/Bar/"), id="with trailing slash"),
            pytest.param(URL("https://example.inv/") / "Foo" / "Bar/", id="dynamically with trailing slash"),
        ],
    )
    def test_url_join(first_url_part: URL):
        assert isinstance(first_url_part, URL)
        second_url_part = URL("Baz/asd?xyz=123")
        actual = first_url_part.join(second_url_part)
>       assert actual == URL("https://example.inv/Foo/Bar/Baz/asd?xyz=123")
E       AssertionError: assert URL('https://example.inv/Foo/Baz/asd?xyz=123') == URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123')
E        +  where URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123') = URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123')

urltest.py:18: AssertionError
PASSED                    [ 75%]FAILED        [100%]
unittests\urltest.py:4 (test_url_join[dynamically with trailing slash])
URL('https://example.inv/Foo/Baz/asd?xyz=123') != URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123')

Expected :URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123')
Actual   :URL('https://example.inv/Foo/Baz/asd?xyz=123')
<Click to see difference>

first_url_part = URL('https://example.inv/Foo/Bar')

    @pytest.mark.parametrize(
        "first_url_part",
        [
            pytest.param(URL("https://example.inv/Foo/Bar"), id="w/o trailing slash"),
            pytest.param(URL("https://example.inv/") / "Foo" / "Bar", id="dynamically without trailing slash"),
            pytest.param(URL("https://example.inv/Foo/Bar/"), id="with trailing slash"),
            pytest.param(URL("https://example.inv/") / "Foo" / "Bar/", id="dynamically with trailing slash"),
        ],
    )
    def test_url_join(first_url_part: URL):
        assert isinstance(first_url_part, URL)
        second_url_part = URL("Baz/asd?xyz=123")
        actual = first_url_part.join(second_url_part)
>       assert actual == URL("https://example.inv/Foo/Bar/Baz/asd?xyz=123")
E       AssertionError: assert URL('https://example.inv/Foo/Baz/asd?xyz=123') == URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123')
E        +  where URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123') = URL('https://example.inv/Foo/Bar/Baz/asd?xyz=123')

urltest.py:18: AssertionError

Process finished with exit code 1

Python Version

3.11.0

multidict Version

multidict==6.0.4

yarl Version

yarl==1.9.2

OS

Windows

Additional context

No response

Code of Conduct

• I agree to follow the aio-libs Code of Conduct

[konstantin]

Workaround:

actual = URL(str(first_url_part)+"/").join(second_url_part)

[Julian Berman]

This looks like correct behavior to me at least.

Compare e.g.:

>>> urllib.parse.urljoin("https://example.inv/Foo/Bar", "Baz/asd?xyz=123")
'https://example.inv/Foo/Baz/asd?xyz=123'

Though some web servers or applications treat paths with or without trailing slashes the same, according to RFC3986 they are not necessarily, and joining them if the path doesn't have a trailing slash indeed has different behavior, similar to what happens in your browser with these relative references.