GeezerVison

Summary

Split the Python 3.14-compatible math.log base handling from #8954.

Use the existing natural-log conversion for both operands, evaluating x before the base. This preserves the large-integer path for integer bases, avoids premature float overflow, and raises ZeroDivisionError when the base logarithm is zero.

Keep log2/log10 and noninteger conversion behavior unchanged. In particular, an exception from __float__ still propagates without retrying __index__.

Extend the existing math snippet with large bases, base one, conversion ordering, integer subclasses, and conversion-error cases. No canonical standard-library assertions change.

Validation

• The expanded snippet fails on main d0dfcb54 with OverflowError for a large integer base.
• The same snippet passes with this patch and on CPython 3.14.7.
• Separate main/patched/reference observations verify the exception type for base one and negative large bases, and the operand conversion order.
• All 6 selected existing logarithm tests in test_math pass unchanged.
• Release workspace test targets compile with --no-run, excluding WASM, venvlauncher, and C API; the separate C-API test target also compiles with --no-run.
• Affected-crate Clippy and repository hooks pass.

Local runtime validation is limited to the focused snippet, observations, and canonical tests above. Broad Rust test targets were compile-checked, not executed.

References: CPython 3.14 log conversion and evaluation order, float division.

AI assistance

• This PR follows the project's AI policy.

OpenAI Codex assisted with auditing the split, implementation, regression coverage, and validation. The runtime does not expose an exact model identifier.