test: add PR #37 register allocation regressions

[EthanZero2Hero]

Summary

• Add dedicated regression tests for PR feat: topic17 v1.3 regression allocator optimization + scenario analysis + docs #37 register allocation changes.
• Cover spill/reload handling without virtual-register leaks.
• Verify spilled register redefinitions are written back before reload.
• Verify exact register classification, including a_temp versus a0.
• Run the regression test explicitly in CI.

Validation

The regression cases were validated against the fetched PR #37 v1.5 implementation. CI will run:

python3.12 -m pytest tests/test_pr37_regression.py -v --tb=short

[github-actions]

🤖 AI Code Review

共审查 2 个变更文件

📁 .github/workflows/ci.yml

🔴 Hardcoded python version — Line 48: python3.12 硬编码，与工作流其他步骤可能不一致（例如 matrix 或系统默认版本）。建议使用 python 或从环境变量/矩阵中获取版本。

🟡 Missing test file validation — 未确认 tests/test_pr37_regression.py 是否存在，若文件缺失则 CI 会失败。建议添加 if: steps.check-test-file.outputs.exists == 'true' 或提前检查文件。

🟡 Step name not reusable — 名称包含 “PR #37”，但此步骤将永久保留在 CI 中，不适用于未来其他 PR。建议改为 “Run register-allocation regression tests”。

💭 No shared pytest configuration — 直接调用 pytest 可能跳过项目级别的 pyproject.toml 或 pytest.ini 配置（如插件、标记、超时）。建议省略参数，让测试继承项目默认配置。

💭 Step placement — 新步骤位于 Generate test visualization page 之前，但后者仅限 main 分支，而新步骤无条件运行。若本意在 main 上才运行，应添加 if: github.ref == 'refs/heads/main' 或保留为全局（当前逻辑一致）。

---

📁 tests/test_pr37_regression.py

🟡 Suggestion: Use explicit assertions instead of relying on StopIteration — In test_spilled_vreg_redefinition_is_written_back_before_reload, if redefine or writeback are not found, next() raises StopIteration with a cryptic error message. Consider using pytest.fail() with a descriptive message or assert with a condition.

🟡 Suggestion: Tighten regex to avoid potential false positives — The regex (?<![A-Za-z0-9_])v[0-9]+(?![A-Za-z0-9_]) is used to check for leaked vregs. It correctly excludes comments, but if a physical register were named like v0 (unlikely), it would be flagged. Consider using a more specific pattern like \bv[0-9]+\b if the assembler supports word boundaries.

💭 Nit: Test relies on hardcoded comment strings — "store redefined v0" and "reload v0" are implementation details. If the allocator's comment format changes (e.g., "store v0 (redefined)"), the test will break. Consider matching on "store redefined v0" as a substring or using a more robust heuristic.

💭 Nit: Partial duplication of test blocks — _pressure_block() and the block in test_spilled_vreg_redefinition... share similar structure. Consider extracting a helper to reduce duplication, but this is minor.

---