[JarryShaw]

• Searched for similar pull requests
• Followed the coding style (make pylint, make mypy, make isort)
• make test passes, and a test case covers the change
• Added a changelog entry — N/A — centralised in docs(changelog): shared 1.5.0 changelog — long-lived, merges last (#610, #616, #617, #618, #620) #657

What is the purpose of your pull request?

• fix — corrects a defect

---

Description of your pull request and other information

Part 1 of #2040 (the owner's ruling: cap it). The 3.7 part is reported on the issue.

• crypto and all now cap cryptography below 46, only on PyPy older than 3.11, using a complementary marker pair. PyPI's file lists show 45.0.7 is the newest release with a manylinux pp310 wheel. 46.0.0–46.0.3 have pp310 wheels only for macOS and Windows, and 46.0.4 and later have none. That makes my <=46.0.3 suggestion on the issue wrong for the Linux leg.
• process.rst's pyproject listing (pinned by test_conventions_doc_claims), a sentence in workflows.rst and test_tier_guard's crypto spot check are all updated to match.
• New tests/project/test_pypy310_cryptography_cap_2040.py evaluates both markers for PyPy 3.10, PyPy 3.11 and CPython 3.10. It fails on main's pyproject (2 failed) and passes here. The neighbouring tests: 168 passed.

#2036's EXPECTED entry for the pypy3.10 install annotation goes stale once this lands.

[JarryShaw]

GOOD TO GO at 9a047569a (Sonnet cross-review).

• Wheels: confirmed from PyPI JSON. 45.0.7 has manylinux pp310 wheels; 46.0.x have only macOS and Windows.
• Markers: confirmed complementary across CPython, PyPy and GraalPy (GraalVM), 3.7–3.15. Exactly one line applies on every interpreter.
• conda: util/conda_vendor.py handles both lines.
• Tests: the new test fails on main's pyproject (2 failed). The reviewer's neighbouring set gives 223 passed, 1 skipped.

UNVERIFIED: a real pypy3.10 (KTLO) install, which only CI shows.