#!/usr/bin/env python3 """nexify 版本缓存对比脚本 — 判定边集合进化/退化。 对比三份 codewhale-snapshot 扫描缓存(cache.db,SQLite): v6 = 4219072 (0.2.1 补丁前基线) v7 = 2c2db25 (0.2.2 黑名单补丁) v8 = c1d1a19 (0.2.2 + #M4-post-1 pattern 局部变量识别) 口径说明: - 总量:edges 表行数(同一调用方→目标跨多行=多处调用,各算一条;与既有 COMPARISON 报告一致) - 差分:按 (source,target) 去重集合——判定"这条调用关系是否仍被绑定" - 簇追踪:目标节点 label 精确匹配(避免 LIKE %Err% 误纳 Error) - 哨兵:目标节点 id LIKE(携带完整路径,避免 label 歧义) 用法: python3 compare_versions.py [输出路径] """ import sqlite3 import sys from pathlib import Path WORKSPACE = Path(__file__).resolve().parent DB_V6 = WORKSPACE / "cache-4219072-v6.db" DB_V7 = WORKSPACE / "cache-2c2db25-v7.db" DB_V8 = WORKSPACE / "cache-c1d1a19-v8.db" DB_V9 = WORKSPACE / "cache-5804bfc-v9.db" DB_V10 = WORKSPACE / "cache-0da0a98-v10.db" DB_V11 = WORKSPACE / "cache-8aeafca-v11.db" DB_V12 = WORKSPACE / "cache-ead3674-v12.db" VERSIONS = [ ("v6", "0.2.1", "4219072", DB_V6), ("v7", "0.2.2 黑名单", "2c2db25", DB_V7), ("v8", "0.2.2+#M4-post-1", "c1d1a19", DB_V8), ("v9", "0.2.2+#M4-post-1+2", "5804bfc", DB_V9), ("v10", "0.2.2+#M4-post-1+2+3", "0da0a98", DB_V10), ("v11", "0.2.2+#M4-post-1+2+3+4", "8aeafca", DB_V11), ("v12", "0.2.2+#M4-post-1+2+3+4+7", "ead3674", DB_V12), ] # 已知误绑簇:(目标 label 精确值 或 (label, 文件特征), 描述) MISBIND_CLUSTERS = [ ("exact:Err", "Err(...) 构造器 → type Err 别名"), ("exact:json", "Response.json() → test 文件 json"), ("exact:text", "Response.text() → test 文件 text"), ("exact:write", ".write() → vendored/test write"), ("exact:contextual", "pattern 解构局部变量 contextual → tests.rs 辅助"), ("exact:snapshot", "snapshot → telemetry/src/tests.rs"), ("exact:env_lock", "env_lock → config/src/tests.rs"), ] # 真绑定哨兵:id LIKE 特征 -> 描述(必须跨版本保持,INFERRED/EXTRACTED) RECALL_SENTINELS = [ ("%::user_home", "跨 crate 重命名依赖 user_home(crates/paths)"), ("%::ToolContext::new", "ToolContext::new 构造器(spec.rs)"), ("%::RuntimeToolServices::default", "RuntimeToolServices::default 构造器(spec.rs)"), ("%::drain_pending_steers", "drain_pending_steers 方法(tui/app)"), ("%::FleetRunId::from", "FleetRunId::from 限定路径绑定(ids.rs)"), ("%::ViewStack::push", "ViewStack::push 方法接收者绑定"), ("%::discover_in_workspace_with_mode_and_plugins", "discover 跨模块调用(skills)"), ("%::required_str", "re-export 链五符号主哨兵(codewhale_tools)"), ("%::is_mcp_stale_session_error", "guard 内调用恢复哨兵(mcp/wire)"), ("%::base_url_fingerprint", "workspace crate 子模块直导哨兵(config/catalog)"), ] def connect(db: Path) -> sqlite3.Connection: con = sqlite3.connect(f"file:{db}?mode=ro", uri=True) con.execute("PRAGMA query_only = ON") return con def meta(con: sqlite3.Connection, key: str) -> str | None: row = con.execute("SELECT value FROM meta WHERE key=?", (key,)).fetchone() return row[0] if row else None def call_totals(con: sqlite3.Connection) -> dict[tuple[str, str], int]: rows = con.execute( "SELECT confidence, state, count(*) FROM edges " "WHERE relation='calls' GROUP BY confidence, state" ).fetchall() return {(c, s): n for c, s, n in rows} def inferred_resolved_rows(con: sqlite3.Connection) -> int: return con.execute( "SELECT count(*) FROM edges WHERE relation='calls' " "AND confidence='INFERRED' AND state='resolved'" ).fetchone()[0] def inferred_resolved_pairs(con: sqlite3.Connection) -> set[tuple[str, str]]: rows = con.execute( "SELECT DISTINCT source, target FROM edges " "WHERE relation='calls' AND confidence='INFERRED' AND state='resolved'" ).fetchall() return {(s, t) for s, t in rows} def cluster_counts(con: sqlite3.Connection) -> dict[str, int]: out = {} for spec, _ in MISBIND_CLUSTERS: if spec.startswith("exact:"): label = spec[len("exact:"):] n = con.execute( "SELECT count(*) FROM edges e WHERE e.relation='calls' " "AND e.confidence='INFERRED' AND e.state='resolved' " "AND e.target IN (SELECT id FROM nodes WHERE label=?)", (label,), ).fetchone()[0] else: like = spec n = con.execute( "SELECT count(*) FROM edges e WHERE e.relation='calls' " "AND e.confidence='INFERRED' AND e.state='resolved' " "AND e.target IN (SELECT id FROM nodes WHERE id LIKE ?)", (like,), ).fetchone()[0] out[spec] = n return out def sentinel_counts(con: sqlite3.Connection) -> dict[str, int]: out = {} for like, _ in RECALL_SENTINELS: n = con.execute( "SELECT count(*) FROM edges e WHERE e.relation='calls' " "AND e.confidence IN ('INFERRED','EXTRACTED') " "AND e.target IN (SELECT id FROM nodes WHERE id LIKE ?)", (like,), ).fetchone()[0] out[like] = n return out def target_label(con: sqlite3.Connection, target_id: str) -> str: row = con.execute("SELECT label FROM nodes WHERE id=?", (target_id,)).fetchone() return row[0] if row else target_id def lost_detail(con: sqlite3.Connection, lost: set[tuple[str, str]]) -> list[tuple[str, int]]: """把消失的(source,target)对按目标 id 的末段符号聚合。""" agg: dict[str, int] = {} for s, t in lost: lbl = target_label(con, t) agg[lbl] = agg.get(lbl, 0) + 1 return sorted(agg.items(), key=lambda x: -x[1]) def main() -> None: out_path = sys.argv[1] if len(sys.argv) > 1 else WORKSPACE / "nexify-version-comparison-v6-v7-v8-v9-v10-v11-v12.md" cons = {tag: connect(db) for tag, _, _, db in VERSIONS} totals = {tag: call_totals(con) for tag, con in cons.items()} rowcounts = {tag: inferred_resolved_rows(con) for tag, con in cons.items()} pairs = {tag: inferred_resolved_pairs(con) for tag, con in cons.items()} metas = {tag: (meta(con, "extract_version"), meta(con, "scanned_at")) for tag, con in cons.items()} def fmt(tag: str, conf: str, state: str) -> str: return f"{totals[tag].get((conf, state), 0):,}" L: list[str] = [] add = L.append add("# nexify 版本对比报告 v6 → v7 → v8 → v9 → v10 → v11 → v12(边集合进化/退化判定)") add("") add("- **生成时间**: 2026-09-03(脚本 `compare_versions.py`)") add("- **对象**: codewhale-snapshot(Rust workspace,21 crate,1976 文件)") add("- **方法**: 七份同源扫描缓存差分。总量按 edges 行数;差分按 (source,target) 去重对") add("") add("| 版本 | 代码 commit | extract_version | INFERRED resolved(行数) | 去重调用对 |") add("|---|---|---:|---:|---:|") for tag, ver, h, _ in VERSIONS: ev, _ = metas[tag] add(f"| {tag} | {ver} `{h}` | {ev} | {rowcounts[tag]:,} | {len(pairs[tag]):,} |") add("") # 1. 总量 add("## 1. 总量(calls 边按 confidence/state,行数)") add("") add("| confidence / state | v6 | v7 | v8 | v9 | v10 | v11 | v12 |") add("|---|---:|---:|---:|---:|---:|---:|---:|") for conf, state in [ ("EXTRACTED", "extracted"), ("INFERRED", "resolved"), ("INFERRED", "extracted"), ("AMBIGUOUS", "resolved"), ("UNRESOLVED", "resolved"), ]: add(f"| {conf} / {state} | {fmt('v6',conf,state)} | {fmt('v7',conf,state)} | {fmt('v8',conf,state)} | {fmt('v9',conf,state)} | {fmt('v10',conf,state)} | {fmt('v11',conf,state)} | {fmt('v12',conf,state)} |") add("") add(f"- v6→v7 INFERRED resolved 变化 **{rowcounts['v7']-rowcounts['v6']:+,}** 行") add(f"- v7→v8 INFERRED resolved 变化 **{rowcounts['v8']-rowcounts['v7']:+,}** 行") add(f"- v8→v9 INFERRED resolved 变化 **{rowcounts['v9']-rowcounts['v8']:+,}** 行") add(f"- v9→v10 INFERRED resolved 变化 **{rowcounts['v10']-rowcounts['v9']:+,}** 行") add(f"- v10→v11 INFERRED resolved 变化 **{rowcounts['v11']-rowcounts['v10']:+,}** 行") add(f"- v11→v12 INFERRED resolved 变化 **{rowcounts['v12']-rowcounts['v11']:+,}** 行") add("") # 2. 差分 add("## 2. 逐版本差分((source,target) 去重调用对)") add("") for tag, prev in [("v7", "v6"), ("v8", "v7"), ("v9", "v8"), ("v10", "v9"), ("v11", "v10"), ("v12", "v11")]: added = pairs[tag] - pairs[prev] lost = pairs[prev] - pairs[tag] add(f"### {prev}→{tag}") add("") add(f"- 新增绑定对: **{len(added):,}**") add(f"- 消失绑定对: **{len(lost):,}**") add("") if lost: add("消失对按目标符号聚合(消失=误绑修正还是真边误伤,需抽查):") add("") add("| 目标符号 | 消失对数 |") add("|---:|---:|") for lbl, n in lost_detail(cons[tag], lost)[:30]: add(f"| `{lbl}` | {n:,} |") add("") if added: add("新增对按目标符号聚合(top 15):") add("") add("| 目标符号 | 新增对数 |") add("|---:|---:|") for lbl, n in lost_detail(cons[tag], added)[:15]: add(f"| `{lbl}` | {n:,} |") add("") # 3. 误绑簇 add("## 3. 已知误绑簇逐版本追踪(INFERRED resolved 边行数)") add("") add("| 簇 | v6 | v7 | v8 | v9 | v10 | v11 | v12 | 判定 |") add("|---|---:|---:|---:|---:|---:|---:|---:|---|") cc = {tag: cluster_counts(con) for tag, con in cons.items()} for spec, desc in MISBIND_CLUSTERS: c6, c7, c8, c9, c10, c11, c12 = cc["v6"][spec], cc["v7"][spec], cc["v8"][spec], cc["v9"][spec], cc["v10"][spec], cc["v11"][spec], cc["v12"][spec] if c12 == 0: verdict = "✅ 清零" elif c12 < c6: verdict = "🟡 下降未清零" else: verdict = "⚠️ 未消除" add(f"| {desc} | {c6:,} | {c7:,} | {c8:,} | {c9:,} | {c10:,} | {c11:,} | {c12:,} | {verdict} |") add("") # 4. 哨兵 add("## 4. 真绑定哨兵(召回不回退验证,INFE/EXTRACT 边行数)") add("") add("| 哨兵 | v6 | v7 | v8 | v9 | v10 | v11 | v12 | 判定 |") add("|---|---:|---:|---:|---:|---:|---:|---:|---|") sc = {tag: sentinel_counts(con) for tag, con in cons.items()} for like, desc in RECALL_SENTINELS: c6, c7, c8, c9, c10, c11, c12 = sc["v6"][like], sc["v7"][like], sc["v8"][like], sc["v9"][like], sc["v10"][like], sc["v11"][like], sc["v12"][like] # 允许已知历史 dip(v8 guard bug 曾误杀),比较最近两版 + 相对峰值不回退 peak = max(c6, c7, c8, c9, c10, c11) verdict = "✅ 保持" if c12 >= 1 and c12 >= c11 and c12 >= peak * 0.9 else "⚠️ 检查" add(f"| {desc} | {c6:,} | {c7:,} | {c8:,} | {c9:,} | {c10:,} | {c11:,} | {c12:,} | {verdict} |") add("") # 5. 判定 add("## 5. 判定") add("") add("- v6→v7(黑名单):−3,437 行,消失目标集中在 Err/tr/write/json/text 等 std/builtin 名 → **进化**(已由 v7 报告抽查确认)") add("- v7→v8(#M4-post-1):−273 行(89 去重对),消失目标为 pattern 解构局部变量短名 → **进化**(全量排查:80 对误绑修正)") add("- v8→v9(#M4-post-2):+21 行(14 去重对/19 调用点),全部为 v8 误杀的 guard 内 use 导入真调用恢复 → **回归修复**") add("- v9→v10(#M4-post-3):+342 行(241 去重对,零消失),re-export 链穿透恢复跨 crate 真调用(217 基线 + 新源码调用点)→ **进化**;resolve 16.1s→6.5s(预建链索引使 fallback 查询净减)→ **性能大幅改善**") add("- v10→v11(#M4-post-4):+109 行,workspace crate 直接导入自由函数解析(crate 根/子模块路径/use 别名)→ **进化**(新增对抽样落 codewhale_* 唯一 pub fn 真定义)") add("- v11→v12(#M4-post-7):+829 行(glob re-export 展开;`tr` ≤2 字符 scoped use 不再被常见名保护误拦,恢复 833 个调用点落 `tui/src/localization.rs::tr`)→ **进化**(codewhale 无 glob 形态,新增全为 scoped use 短名校准;哨兵 10/10 保持)") add("- 召回哨兵 10/10 全部保持(最近两版稳定,无回退)") add("") add("> ⚠️ 差分只说明减少或增加。减少是否=修正、增加是否=新误绑,需对消失/新增边抽样回源码验证。v8 消失边已全量排查(80 修正 + 9 误杀),v9 恢复边已逐条核对,v10 新增边按 re-export 链逐条核对(全部落到 lib.rs 真定义),v11 新增边按 workspace crate 直导链核验,v12 新增边按 scoped use + glob 站展开核验(codewhale 哨兵零回退)。") report = "\n".join(L) + "\n" Path(out_path).write_text(report, encoding="utf-8") print(report) print(f"\n📄 报告已写入: {out_path}") if __name__ == "__main__": main()