py2rs-review-r4-algo-complexity

name	py2rs-review-r4-algo-complexity
description	[DRAFT] 第 4 轮算法复杂度审查。对一个迁移单元或已收批的 review batch，只在有复杂度分析、理论依据和 benchmark 证据时建议或实施算法改变；默认审查不改生产代码。

R4 Algorithm Complexity Review

R4 只处理算法和复杂度。Rust 代码不因为“看起来更 Rust”就改算法。

Preconditions

• The R0 behavior report exists.
• Earlier required reviews are complete or explicitly deferred.
• Representative input sizes and hot paths are known or can be measured.

Review Focus

• O(n^2) or worse loops on realistic large inputs.
• Repeated parsing, sorting, serialization or regex work.
• Hot-path allocations and string concatenation.
• Missing indexes/maps for repeated lookup.
• Python C-extension behavior that Rust does not automatically beat.
• Memory growth from proposed optimizations.

Evidence Rule

Any algorithm change requires all three:

1. Complexity analysis: old vs new.
2. Theoretical basis: why the new structure improves cost.
3. Benchmark or measured fixture: before/after numbers.

Without the three, write a finding or recommendation; do not change the algorithm.

Workflow

1. Confirm the unit or closed review batch, included unit ids and performance-relevant public behavior.
2. Identify hot paths from code, tests, profiles or realistic workload notes.
3. Classify findings as required, optional or no-change.
4. Run or request benchmarks where practical.
5. Save a report with decision.

Code Changes

Default review-gate mode does not edit production code. If explicitly asked to remediate:

• Apply only changes backed by the evidence rule.
• Keep public behavior stable.
• Record benchmark numbers.
• Rerun the R0 behavior gate after changes.
• Do not approve your own final R4 report.

Report

Use project convention or:

reviews/YYYY-MM-DD-<unit-or-batch-id>-data-algorithm.md

Include:

• Scope reviewed
• Included unit ids and per-unit verdicts (not_required only when the manifest does not require R4 for that unit)
• Hot paths inspected
• Complexity findings
• Benchmark/checks run
• Residual risk
• Promotion decision

Non-Negotiables

• No speculative "probably faster" changes.
• No behavior change without explicit migration-unit policy.
• No benchmark-free algorithm rewrite.