Switch benchmarks to real HTTP roundtrips against a localhost WSGI server

[Kludex]

Summary

• The previous suite mixed pure-Python microbenchmarks (URL parsing, header construction) with MockTransport-based "client" benchmarks that short-circuited every transport-layer code path. Optimisations to the network backends (e.g. perf: use memoryview in write() to avoid copies #954) could not be detected because MockTransport never touches httpcore2._backends.
• This PR replaces the suite with six benchmarks that exercise the user-facing path: a tiny in-process WSGI server runs on a random localhost port, and the benchmarks drive a real httpx2.Client against it. The whole stack runs - Client, Transport, connection pool, real sockets - so any optimisation along that path now produces a measurable signal.

Benchmarks

• test_bench_url_parse_and_join - pure CPU, URL parse + join regression canary.
• test_bench_request_build_json - request construction + JSON encode.
• test_bench_client_get_json - GET a moderate JSON response from the real server.
• test_bench_client_post_large - POST 4MB through the full client (exercises socket write).
• test_bench_client_stream_download - stream 4MB from the server (exercises socket read).
• test_bench_client_keepalive_burst - 64 sequential GETs over a kept-alive connection.

Test plan

• CodSpeed walltime CI runs cleanly on this PR
• Re-run perf: use memoryview in write() to avoid copies #954 against main with this suite in place and confirm regressions/improvements at the socket layer now register

AI Disclaimer

This PR was developed with the assistance of either Claude or Codex. I've reviewed and verified the changes.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1ae46818f2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Enable HTTP/1.1 for keep-alive benchmark server

_SilentHandler inherits WSGIRequestHandler without overriding protocol_version, so the benchmark server responds as HTTP/1.0 (connection closed after each response). In this setup, test_bench_client_keepalive_burst cannot actually exercise a kept-alive connection, and the other client benchmarks also include reconnect overhead rather than stable connection reuse, which invalidates the transport/pooling signal this commit is trying to measure.

Useful? React with 👍 / 👎.

[codspeed-hq]

Merging this PR will degrade performance by 92.57%

❌ 1 regressed benchmark
🆕 5 new benchmarks
⏩ 6 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
🆕	WallTime	test_bench_request_build_json	N/A	204.6 ms	N/A
🆕	WallTime	test_bench_url_parse_and_join	N/A	165.8 ms	N/A
🆕	WallTime	test_bench_client_post_large	N/A	149.4 ms	N/A
🆕	WallTime	test_bench_client_get_json	N/A	344.5 ms	N/A
❌	WallTime	test_bench_client_stream_download	28.2 ms	379.3 ms	-92.57%
🆕	WallTime	test_bench_client_keepalive_burst	N/A	235.6 ms	N/A

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing high-level-benchmark-suite (a1d3bd5) with main (8d5f182)}

Footnotes

1. 6 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩