Regression: Equality of joined URLs fails to be compared equal with a readymade URL object.

[gmichaeljaison]

Describe the bug

Equality of joined URLs fails to be compared equal with a readymade URL object.

The following code returns False in the latest release version 1.9.1. It used to return True in previous version.

(yarl.URL("https://hello.com") / "world") == yarl.URL("https://hello.com/world")

To Reproduce

(yarl.URL("https://hello.com") / "world") == yarl.URL("https://hello.com/world")

Expected behavior

Should be equal

Logs/tracebacks

(yarl.URL("https://hello.com") / "world") == yarl.URL("https://hello.com/world")

False

Python Version

$ python --version
3.8+

multidict Version

$ python -m pip show multidict
Name: multidict
Version: 6.0.2
Required-by: httpie, yarl

yarl Version

$ python -m pip show yarl
1.9.1

OS

macOS

Additional context

No response

Code of Conduct

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

[enku]

A git bisect reveals the regression was caused by 2dac93f.

[dalazx]

I am seeing a significant regression as well:

>>> from yarl import __version__, URL
>>> __version__
'1.9.1'
>>> (URL("http://0.0.0.0:1234/this/is/a/path")).raw_path
'/this/is/a/path'
>>> (URL("http://0.0.0.0:1234") / "this/is/a/path").raw_path
'this/is/a/path'

the latter example is missing the leading slash.

[achimnol]

I have encountered the same issue:

The test code:

import yarl
from pathlib import Path
import aiohttp
import asyncio
import os

async def main(): -> None
    sock_path = Path("/var/run/docker.sock")
    decoded_path = os.fsdecode(sock_path)
    docker_host = yarl.URL("http://localhost")
    connector = aiohttp.UnixConnector(decoded_path)
    async with aiohttp.ClientSession(connector=connector) as sess:
        url = docker_host / "info"
        print(repr(url))
        print(str(url))
        print(bytes(url))
        print(f"{url.host=}", f"{url.path=}", f"{url.raw_path=}")
        async with sess.get(url) as resp:
            data = await resp.text()
            print(data)

if __name__ == "__main__":
    asyncio.run(main())

[achimnol]

If I try to mangle path like base_url / "/info", it says ValueError: Appending path '/info' starting from slash is forbidden.
I think the path joining rule should applied slightly differently when it is done via the '/' (truediv) operator, at least.

[selcukcihan]

I'd like to report an issue we are experiencing. We have yarl dependency through aiohttp and with version 1.9.1 our tests started failing. The problem I'm seeing is with a change in behaviour of update_query method. In the new version, update_query(None) removes the query string completely.

Version 1.8.2

Python 3.9.16 (main, Mar 15 2023, 15:55:54) 
[Clang 14.0.0 (clang-1400.0.29.202)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import yarl
>>> yarl.__version__
'1.8.2'
>>> url = yarl.URL("http://127.0.0.1:3177/?foo=bar")
>>> url.update_query(None) == url
True

Version 1.9.1

Python 3.9.16 (main, Mar 15 2023, 15:55:54) 
[Clang 14.0.0 (clang-1400.0.29.202)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import yarl
>>> yarl.__version__
'1.9.1'
>>> url = yarl.URL("http://127.0.0.1:3177/?foo=bar")
>>> url.update_query(None) == url
False

[webknjaz]

Hi folks, thanks for reaching out! It would be useful if someone could submit a PR with a failing test for this per https://pganssle-talks.github.io/xfail-lightning/.

I'm at the PyCon Sprints now and might have time to look into this.

@mjpieters is this related to your changes?

[mjpieters]

@mjpieters is this related to your changes?

Looks like it as per git bisect, I'll prioritise this next. Hopefully today.

[mjpieters]

In the new version, update_query(None) removes the query string completely.

That was always the intended, documented behaviour, it was a bug that it didn't actually do this before. Use an empty dictionary if you don't want the query to be reset. See #723

[mjpieters]

If I try to mangle path like base_url / "/info", it says ValueError: Appending path '/info' starting from slash is forbidden. I think the path joining rule should applied slightly differently when it is done via the '/' (truediv) operator, at least.

That was already the behaviour before and not a regression; it was an explicit choice to disallow this.