[charliermarsh]

We parse build-system.requires as PEP 508 requirements in the build backend, reserving distribution-URL classification for uv's direct-build validation. The frontend supplies the build requirements before invoking the backend, so the backend does not need Git parsing or archive-extension inference for those URLs.

This intentionally accepts some build-system URLs that the backend previously rejected, such as uv_build @ hg+https://example.com/uv-build or an archive download endpoint without a filename extension. Requirement syntax and project dependency parsing remain unchanged. Direct builds retain the richer URL parser and its existing errors.

On Linux x86_64 with Rust 1.99.0 and the stripped minimal-size profile, uv-build decreases from 3,075,808 to 3,029,232 bytes (46,576 bytes, 1.51%) against main at e37c1f0c6.

All 68 backend tests, strict clippy, and formatting pass. A regression test exercises successful backend metadata generation and rejection by direct-build validation for the same Mercurial URL. Comparisons across 180 URL cases, 756 marker cases, and the CLI/archive corpus found only the intended build-system URL validation changes; previously successful builds produce identical output.

This absorbs the standalone-binary size benefit of #2753, so their savings should not be added together.

[codspeed]

Merging this PR will not alter performance

⚠️ 8 benchmarks spent significant time in system calls

System calls cannot be consistently instrumented, so they are not included in the measure, which understates the real cost. Please switch to the Walltime instrument to accurately measure system calls.

Measurement and system calls

• discover_workspace_from_all_members_with_excludes
• install_wheel_many_files
• discover_workspace_from_all_members
• resolve_warm_airflow
• resolve_warm_jupyter
• run_python_version_synthetic_workspace
• resolve_warm_jupyter_universal
• unpack_sdist_many_files

✅ 69 untouched benchmarks
⏩ 35 skipped benchmarks¹

---

_{Comparing charlie/codex-uv-build-requirement-urls (bcbfc13) with main (e37c1f0)}

Footnotes

1. 35 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. ↩

[astral-automations-bot]

Promoted to #22577.