Some tests fail with some supported httpx versions due to whitespace differences in JSON responses

[musicinmybrain]

I’m working on updating the python-starlette package to 0.42.0 in Fedora.

I see that #2773 pinned httpx to the range >=0.27.0,<0.29.0 due what seems to have been an API incompatibility.

That PR also adjusted the whitespace in some expected JSON response strings, e.g. from {"body": '{"a": "123"}'} to {"body": '{"a":"123"}'} This seems to be correct for the latest httpx, 0.28.1, but the whitespace change isn’t in the httpx 0.27 series, so several tests fail if httpx 0.27.0, 0.27.1, or 0.27.2 is installed. For example:

$ git clone https://github.com/encode/starlette.git
$ cd starlette
$ git checkout 0.42.0
$ uv venv _e
$ . _e/bin/activate
(_e) $ uv pip install -e .[full]
(_e) $ uv pip install -r requirements.txt
(_e) $ python -m pytest
= 867 passed, 2 xfailed in 4.18s =
(_e) $ uv pip install httpx==0.27.2
(_e) $ python -m pytest
[output condensed with most details omitted]
= FAILURES =
_ test_wsgi_post[asyncio] _
>       assert response.json() == {"body": '{"a":"123"}'}
E       assert {'body': '{"a": "123"}'} == {'body': '{"a":"123"}'}
E         
E         Differing items:
E         {'body': '{"a": "123"}'} != {'body': '{"a":"123"}'}
E         Use -v to get more diff

tests/test_requests.py:94: AssertionError
_ test_wsgi_post[trio] _
_ test_request_body[asyncio] _
_ test_request_stream[asyncio] _
_ test_request_body[trio] _
_ test_request_stream[trio] _
= 6 failed, 861 passed, 2 xfailed in 4.02s =

It’s easy enough to work around this downstream in Fedora by reverting just 24de2bf in a patch until our python-httpx is updated to 0.28.x. In any case, this is a testing issue, not an issue with the Starlette library itself. Still, I’m curious whether this should be considered a bug.

[Kludex]

Thanks for the report 🙏

It’s easy enough to work around this downstream in Fedora by reverting just 24de2bf in a patch until our python-httpx is updated to 0.28.x. In any case, this is a testing issue, not an issue with the Starlette library itself. Still, I’m curious whether this should be considered a bug.

It affects the TestClient, which is a public API, and you could say it's a breaking change... Since HTTPX was fine with this change, I think it's all good.

Ref.:

• Sharper CHANGELOG entry. httpx#3448
• Ensure JSON representation is compact. #3363 httpx#3367