Benchmarks updated; some tests parametrized

[ml31415]

• Restkit removed from benchmarks as it seems discontinued
• httpx added to benchmarks
• Some test cases fixed or improved using pytest parametrization
• publish.yaml github actions updated

[cyberw]

Nice!

[cyberw]

Should I publish a release?

[ml31415]

I'm still working on it. Probably good to wait one or two more days.

[cyberw]

Ok! I dont mind doing a 2.1.1 soon after, in order to keep 2.0.12->2.1.0 mainly consisting of the dropped <3.9 support and updated llhttp.

[ml31415]

Ok, then feel free to do the release. Just pushed my last changes for today. I hope the action updates in the publish workflow didn't mess up anything.

[cyberw]

Hmm... there seems to be some flakiness with httpbin tests. https://github.com/geventhttpclient/geventhttpclient/actions/runs/8621840724

I have already made the git tag and GH release, but held off on triggering the "publish" action.

[ml31415]

See #198 . It's harmless, but we need some more reliable solution for testing I guess. Even if this might be some temporary overload of httpbin. It's not the first time.

[cyberw]

I triggered the publish. Not sure whether httpbin is flaky or throttling us somehow.

[cyberw]

Publish failed, maybe related to the switch to pyproject.toml? https://github.com/geventhttpclient/geventhttpclient/actions/runs/8622216864/job/23632872268

[ml31415]

Checking dist/UNKNOWN-0.0.0.tar.gz: ERROR InvalidDistribution: Metadata is missing required fields: Name.

Name is present in pyproject.toml. It should not be necessary anymore in setup.py, so I had removed all metadata from there. Initially I had tried to leave the metadata in both files, but then setuptools came up with warnings of overriding the settings. Maybe only name and version need to be present in setup.py?

In general, setuptools are currently some major pain with pyproject.toml compatibility. They seem not really in the mood, to proactively support entries in their pyproject tools section and rather wait for some PEP discussions to end for some python wide standard. That's the only reason why I couldn't remove setup.py and MANIFEST.in. And there are several more such bug reports pending for ages already. The more I have to mess with setuptools, the more I feel tempted to try out some other build tool, poetry or hatchling maybe.

[cyberw]

I think it should be fixable with setuptools too (havent investigated though)

I'm a little skeptical to poetry because of its poor speed and complexity. Also it seems very focused on applications as opposed to libraries. Never tried hatchling, but it does seem to be gaining some traction recently, maybe that is a good thing to try...