prove: P3 — --all 目录模式、T4 差分一致性、库语义对齐、2.8.0 收口
- main.rs:prove --all/--max-files(递归收集,目录不带 --all 拒绝; 聚合 JSON kind=prove-all,整体状态差者胜);--fault-inject offline-claim (调试分歧注入,A3 护栏验证入口) - tests/conformance.rs:T4 差分一致性——5 例纯函数语料(算术/中文标识符/ 分支比较/字符串拼接/i64 边界)解释器求值 == Lean example(simp+try rfl); B4 result.ok 结构字段 access 两侧一致 - MarqdoProof.lean:mqAdd 补文本拼接与 num 混合(与解释器 eval_binary(Add) 对齐——T4 实测抓出的真分歧);mqDisplay 显示串(标量精确,容器一层, 有意简化并记录) - 文档:design §12 P3 落地 + §12.4 延期理由(foreach/assume-guarantee/T3 → 2.9)+ §12.5 权限受限待合入清单;roadmap 状态 → 已实施; 教程 §13.8 prove 章节;skills/marqdo/reference.md 定理表; doc/verification/skill-proof-theory.md(库形式化说明); doc/roadmap/prove-release-2.8.0.md(双语发版说明 + CHANGELOG/README/ skills/marqdo-dev prove 任务待合入原文) - Cargo.toml/Cargo.lock:2.7.1 → 2.8.0 实测:全量 516 通过(仅 2 个已记录环境基线失败);prove 23 + prove_lean 10 + corpus 40 夹具 + conformance 6 例全绿;biz 定理 proved、假定理反例、 --all 聚合与护栏目录模式同样生效。
This commit is contained in:
1 parent
63d60e3ba1
commit
4fae71342d
12 files changed
+1015
-21
No files matched your search
Generated
+14
-15
@@ -1,6 +1,6 @@
|
||||
# This file is automatically @generated by Cargo.
|
||||
# It is not intended for manual editing.
|
||||
version = 4
|
||||
version = 3
|
||||
|
||||
[[package]]
|
||||
name = "adler2"
|
||||
@@ -143,9 +143,9 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "bumpalo"
|
||||
version = "3.20.3"
|
||||
version = "3.20.4"
|
||||
source = "registry+https://github.com/rust-lang/crates.io-index"
|
||||
checksum = "72f5acc6cb2ba439de613abc23857ec3d78374d8ed5ac84e9d11336e87da8649"
|
||||
checksum = "87a0cf66fd1eb2fb7b822cb78370d82303750809bc6ba855f3f58e517d3238db"
|
||||
|
||||
[[package]]
|
||||
name = "byteorder"
|
||||
@@ -582,9 +582,9 @@ checksum = "8f42a60cbdf9a97f5d2305f08a87dc4e09308d1276d28c869c684d7777685682"
|
||||
|
||||
[[package]]
|
||||
name = "js-sys"
|
||||
version = "0.3.106"
|
||||
version = "0.3.107"
|
||||
source = "registry+https://github.com/rust-lang/crates.io-index"
|
||||
checksum = "7883d941dae510fb2d978fc3fe018c71c9e2892fd38854de3e8b92c2e5ad9cc5"
|
||||
checksum = "8dc09d2154098bbf4ef54eab6b117a3c24cbbffe94046b16ee6d6d4510584d3a"
|
||||
dependencies = [
|
||||
"cfg-if",
|
||||
"futures-util",
|
||||
@@ -615,7 +615,7 @@ checksum = "0ceec5bc11778974d1bcb055b18002eba7f4b3518b6a0081b3af5f21666da9ad"
|
||||
|
||||
[[package]]
|
||||
name = "marqdo"
|
||||
version = "2.7.1"
|
||||
version = "2.8.0"
|
||||
dependencies = [
|
||||
"anyhow",
|
||||
"assert_cmd",
|
||||
@@ -1456,12 +1456,11 @@ checksum = "ccf3ec651a847eb01de73ccad15eb7d99f80485de043efb2f370cd654f4ea44b"
|
||||
|
||||
[[package]]
|
||||
name = "wasm-bindgen"
|
||||
version = "0.2.129"
|
||||
version = "0.2.130"
|
||||
source = "registry+https://github.com/rust-lang/crates.io-index"
|
||||
checksum = "9bb54f33acc68fd454578d9820b0bde1a1a3d17aa17bb7b6595806d02886d409"
|
||||
checksum = "a0181c14ddd6c7b8b0b0132a5f1ea2f32be0d893cd3ea97b6e3b28886b8a3d1c"
|
||||
dependencies = [
|
||||
"cfg-if",
|
||||
"once_cell",
|
||||
"rustversion",
|
||||
"wasm-bindgen-macro",
|
||||
"wasm-bindgen-shared",
|
||||
@@ -1469,9 +1468,9 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "wasm-bindgen-macro"
|
||||
version = "0.2.129"
|
||||
version = "0.2.130"
|
||||
source = "registry+https://github.com/rust-lang/crates.io-index"
|
||||
checksum = "2e29d0c35b16e224a7eeb5cd2d25e3e1968fbd65604117b44d3b789d00ee8535"
|
||||
checksum = "6e931a02a39abdbf66966fc35902b5ed58344750a70885bed4ebb3d8eaf28ed9"
|
||||
dependencies = [
|
||||
"quote",
|
||||
"wasm-bindgen-macro-support",
|
||||
@@ -1479,9 +1478,9 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "wasm-bindgen-macro-support"
|
||||
version = "0.2.129"
|
||||
version = "0.2.130"
|
||||
source = "registry+https://github.com/rust-lang/crates.io-index"
|
||||
checksum = "6f501a8bc3719dba86ef8ae4728879c08001bea749eb1333ac5b91e040e2a6b7"
|
||||
checksum = "2ca4b54280e455bd831e33a6b8e734baf432fe63fbc35c42ac14ce67a043a53a"
|
||||
dependencies = [
|
||||
"bumpalo",
|
||||
"proc-macro2",
|
||||
@@ -1492,9 +1491,9 @@ dependencies = [
|
||||
|
||||
[[package]]
|
||||
name = "wasm-bindgen-shared"
|
||||
version = "0.2.129"
|
||||
version = "0.2.130"
|
||||
source = "registry+https://github.com/rust-lang/crates.io-index"
|
||||
checksum = "23f0c9c52aa7cd7d77769a4cfe2a9adb1b331f489a41d912ce14513d5ab995c6"
|
||||
checksum = "3b1f9c3798725016c6ee5f2f57869ec0831e8bbc0ab0310a86e787e97f41389e"
|
||||
dependencies = [
|
||||
"unicode-ident",
|
||||
]
|
||||
|
||||
+1
-1
@@ -1,6 +1,6 @@
|
||||
[package]
|
||||
name = "marqdo"
|
||||
version = "2.7.1"
|
||||
version = "2.8.0"
|
||||
edition = "2021"
|
||||
license = "Apache-2.0"
|
||||
description = "Marqdo — Markdown markers as an interpreted programming language"
|
||||
|
||||
+33
-1
@@ -2,7 +2,7 @@
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| 状态 | **实施中(Partial)**:P0 全部、P1 Lean 判定已落地(见 §12);P2 离线符号执行、P3 目录模式进行中。路线图:[roadmap/prove.md](../roadmap/prove.md) |
|
||||
| 状态 | **已实施(Implemented · 2.8.0)**:P0–P3 落地记录见 §12;遗留项(foreach、跨模块 assume-guarantee、假设表 T3)显式延期至 2.9,理由见 §12.4。路线图:[roadmap/prove.md](../roadmap/prove.md) |
|
||||
| 日期 | 2026-11-13 |
|
||||
| 层归属 | **Language**:定理表角色 + `check --proof`(§7,[ADR 0010](../adr/0010-skill-proof-boundary.md))· **Runtime**:`prove` 管线(§4–§6、§8)· **Document**:`verification/skill-proof` 证据 |
|
||||
| 相关 | [core-surface.md](core-surface.md)(复用 `table` 构造,不新增标记)· [layers.md](layers.md)(准入)· [markdown-mapping-v0.3.md](markdown-mapping-v0.3.md) · [ADR 0010](../adr/0010-skill-proof-boundary.md) · [verification/replay-policy](../../verification/replay-policy/README.md)(Lean 伴随判定的先例) |
|
||||
@@ -894,3 +894,35 @@ CI:`.github/workflows/proof.yml` —— 装 Lean 4.35.0 → 底座校验 →
|
||||
**已记录的文法限制**:分支条件尚不支持字段访问(`access(审核, "ok")` 作条件
|
||||
解析失败)——F4 的"路径条件含自由原子"场景目前由单元测试覆盖,语料库中以
|
||||
分支后的结构化结论代替;文法放开后 prove 侧无需改动(offline.cond_atom 已支持)。
|
||||
|
||||
### P3 组合与产品化(2026-11-17)
|
||||
|
||||
| 任务 | 状态 | 实现 |
|
||||
|---|---|---|
|
||||
| `--all` 目录模式 + `--max-files` | ✅ | 递归收集(跳过 target/dist/node_modules/.git/.marqdo,上限 200);目录不带 --all → 拒绝;聚合 JSON `{"kind":"prove-all","status":…,"files":[…]}`;整体状态差者胜(A6 测试:假 lean 分歧在目录模式同样触发护栏、整体退 1) |
|
||||
| 差分一致性 T4(B 组) | ✅ | `tests/conformance.rs`:5 个纯函数语料(算术/中文标识符/分支比较/字符串拼接/i64 边界)解释器求值 == Lean `example … = Val.… := by simp`;B4 result.ok 结构字段读取两侧一致。B2(除零/模零)保持已记录偏差:解释器为运行时错误,P1 库模型返回 `.int 0`(MarqdoProof.lean 注释) |
|
||||
| 调试注入 `--fault-inject offline-claim` | ✅ | 等价于 `MARQDO_PROOF_INJECT_DIVERGENCE=1`(A3 分歧护栏验证入口) |
|
||||
| 跨函数组合 | 部分 | 同模块直线纯函数**内联**(≤24 语句、无分支、effect=pure)已落地;跨模块调用=不透明操作(效应未知,护栏兜底);**assume-guarantee 后置条件复用 → 2.9**(依赖假设表 T3) |
|
||||
| `foreach` | 延期 | 维持 `proof.foreach_pending` 拒绝;有限列表 fold 编码需要库级归纳引理,冒进会先于 Lean 关闭路径、破坏单调性护栏可信度(§12.4) |
|
||||
| 文档 | ✅ | 教程 §13.8(prove 章节)、skills/marqdo/reference.md(定理表)、本设计文档状态行、roadmap 落地记录;README/CHANGELOG/doc/releases 因仓库权限(root 属主)待人工合入,内容已在 §12.5 |
|
||||
|
||||
#### §12.4 延期理由(2.9)
|
||||
|
||||
1. **foreach**:可证子集刻意无循环(返回路径有限性是反例枚举与 Lean 分支
|
||||
拆解的共同根基)。fold 编码需 `List.length` 归纳引理入 MarqdoProof.lean,
|
||||
且离线判定器需要"有限列表"的证据语法——两件事都没有通过 40 夹具语料库
|
||||
的单调性验证,不该在发布版里冒险。
|
||||
2. **跨模块 assume-guarantee**:需要假设表(T3)承载被调者后置条件,并把
|
||||
假设作为 `hypothesis` 关闭理由暴露给证书。当前跨模块调用是不透明操作,
|
||||
诚实且护栏可兜底;半吊子假设复用会制造"看起来证了其实靠假设"的绿灯。
|
||||
3. **分支条件字段访问**(`access(审核, "ok")` 作条件):解析器文法限制,
|
||||
prove 侧已就绪(offline cond_atom 支持任意键),放开即生效。
|
||||
|
||||
#### §12.5 待仓库权限合入的内容
|
||||
|
||||
以下文件 root 属主、当前工作目录不可写(uid=1000):
|
||||
|
||||
- `CHANGELOG.md`:v2.8.0 条目(见 `doc/roadmap/prove-release-2.8.0.md`)
|
||||
- `README.md`:版本行 2.7.1→2.8.0 + prove 简介(同上文件附文案)
|
||||
- `doc/releases/v2.8.0.md`:双语发布说明(同上文件附全文)
|
||||
- `skills/marqdo-dev/index.mq.md`:`prove` 任务(契约见本节 §12.3 表)
|
||||
@@ -0,0 +1,133 @@
|
||||
# v2.8.0 发布说明 · Release Notes(prove)
|
||||
|
||||
> 仓库权限说明:本文件由 2.8.0 收口时的自动化会话写入 `doc/roadmap/`
|
||||
> (`CHANGELOG.md`、`README.md`、`doc/releases/`、`skills/marqdo-dev/` 为
|
||||
> root 属主、运行账号不可写)。内容已定稿,合入即可;§4 为待合入原文。
|
||||
|
||||
## 1. 一句话
|
||||
|
||||
`marqdo prove` 正式发布:知识文件的定理表 → Lean 4 证明 → 自动判定。
|
||||
没有 Lean 工具链也能跑(离线符号执行,诚实降级);有 Lean 时全链路证明,
|
||||
假定理给出路径见证反例,离线与 Lean 分歧时显式报错永不落绿。
|
||||
|
||||
`marqdo prove` ships: theorem tables in knowledge files compile to Lean 4
|
||||
and are judged automatically. It runs without a Lean toolchain (offline
|
||||
symbolic execution, honest degradation); with Lean, fake theorems get
|
||||
counterexamples and any offline/Lean divergence fails loudly.
|
||||
|
||||
## 2. 亮点(Highlights)
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| 定理表 | `| 定理 | 结论 | 前提 | 策略 |`,结论文法:`结果.ok`、`包含(轨迹,…)`、`长度(轨迹)`、参数比较,且/或/非组合 |
|
||||
| 六段管线 | 装载 → 降级 → 定型 → 生成(Lean 4.35)→ 判定(离线先行 + Lean 复核)→ 证书 |
|
||||
| 离线判定 | F1–F5 片段符号执行:路径枚举、轨迹形状、构造器结构、自由原子二分、区间字面重述与反例见证(毫秒级) |
|
||||
| 单调性护栏 | 离线 `checked` 但 Lean 证不出(或反例被证出)→ `error[proof.divergence]` 退 1——绝不让离线判定比 Lean 更乐观 |
|
||||
| 诚实状态 | `proved > checked > unverified > rejected > unproved > error`;`checked ≠ proved`;无 Lean 时状态如实降级(咨询性退 0),`--require-lean` CI 严格模式退 2 |
|
||||
| 反例 | 假定理给出区间端点见证(如 `金额 = 10001`)并指回定理表行号 |
|
||||
| 证书缓存 | 复合键(源码⊕计划⊕库⊕Lean 版本⊕CLI 版本),`.marqdo/prove/`,命中跳过判定;`--no-cache` 绕过 |
|
||||
| 目录模式 | `prove --all 目录`(`--max-files` 默认 200):逐文件出证,整体取最差,聚合 JSON |
|
||||
| 差分一致性 | T4 语料:解释器求值 == Lean 语义(算术/中文标识符/分支/拼接/i64 边界/结构字段) |
|
||||
| 工具链 | `--lean` 显式指定 → `MARQDO_PROOF_LEAN` → PATH;内嵌判定库按内容哈希缓存构建,无需 lake/elan |
|
||||
|
||||
## 3. 使用(Usage)
|
||||
|
||||
```bash
|
||||
marqdo prove 审批.mq.md --json # 判定 + 单个 JSON 报告
|
||||
marqdo prove emit 审批.mq.md # 只生成 Lean 模块 + plan.json
|
||||
marqdo prove audit 审批.mq.md # 只审查可证子集(不写盘)
|
||||
marqdo prove --all skills/ --json # 目录聚合(整体取最差)
|
||||
marqdo prove clean # 清空 .marqdo/prove
|
||||
```
|
||||
|
||||
测试:`cargo test --test prove`(无 Lean 全绿)·
|
||||
`MARQDO_PROOF_LEAN=<lean> cargo test --test prove_lean --test prove_corpus --test conformance`(有 Lean 全链路 + 40 夹具单调性 + T4 一致性)。
|
||||
|
||||
## 4. 待合入原文(需 root 属主文件权限)
|
||||
|
||||
### 4.1 `CHANGELOG.md` 追加
|
||||
|
||||
```markdown
|
||||
## v2.8.0 — 2026-11-17
|
||||
|
||||
`marqdo prove` 正式发布:定理表 → Lean 4 证明 → 自动判定。离线符号执行
|
||||
(F1–F5)无工具链可用,`checked` 永远弱于 `proved`;单调性护栏让判定器
|
||||
缺陷显式失败;假定理给出反例见证;证书缓存与目录模式面向智能体批量出证。
|
||||
|
||||
`marqdo prove` ships: theorem tables compile to Lean 4 and are judged
|
||||
automatically. Offline symbolic execution works without a toolchain,
|
||||
monotonicity guardrails fail loudly on judger defects, fake theorems get
|
||||
counterexample witnesses, and certificate caching plus directory mode
|
||||
support agent-scale proving.
|
||||
|
||||
详见 [双语发布说明](doc/roadmap/prove-release-2.8.0.md)。
|
||||
```
|
||||
|
||||
### 4.2 `README.md` 修改
|
||||
|
||||
版本行(第 19–21 行)替换为:
|
||||
|
||||
```markdown
|
||||
当前 CLI **2.8.0** 新增 `marqdo prove`:知识文件的定理表编译为 Lean 4 并自动
|
||||
判定真假——没有 Lean 工具链也能离线符号执行(诚实降级,绝不冒充证明),
|
||||
有 Lean 时全链路证明并给出反例见证。参见[发布说明](doc/roadmap/prove-release-2.8.0.md)。
|
||||
|
||||
CLI **2.8.0** adds `marqdo prove`: theorem tables in knowledge files compile
|
||||
to Lean 4 and are judged automatically — offline symbolic execution without
|
||||
a toolchain (honest degradation, never impersonating a proof), full proofs
|
||||
with counterexample witnesses when Lean is available.
|
||||
```
|
||||
|
||||
「愿景与期许」清单追加第 7 条:
|
||||
|
||||
```markdown
|
||||
7. **知识可证明**:知识文件内的定理表 → Lean 4 定理 → 自动判定
|
||||
(`marqdo prove`);离线符号执行与 Lean 单调一致,分歧显式报错。
|
||||
```
|
||||
|
||||
### 4.3 `skills/marqdo-dev/index.mq.md` 追加 `prove` 任务
|
||||
|
||||
```markdown
|
||||
## prove
|
||||
|
||||
| 参数 | 类型 | 默认 | 说明 |
|
||||
|---|---|---|---|
|
||||
| path | text | "." | 单个 `.mq.md` 或目录(目录需要 all=True) |
|
||||
| all | bool | False | 目录模式:递归收集 .mq.md 逐文件出证,整体取最差状态 |
|
||||
| max_files | num | 200 | 目录收集上限 |
|
||||
| require_lean | bool | False | CI 严格模式:未判定状态退 2 |
|
||||
| json | bool | False | 单个 JSON 对象输出(智能体可读契约) |
|
||||
|
||||
| 返回 | 类型 | 说明 |
|
||||
|---|---|---|
|
||||
| result | map | Result:真实 `marqdo prove` 退出状态、标准输出与错误输出 |
|
||||
|
||||
证明知识文件的定理表:先离线符号执行(毫秒级 checked,不是 proved),
|
||||
有 Lean 工具链时由 Lean 复核升级 proved;假定理给出路径见证反例;
|
||||
离线与 Lean 分歧触发 proof.divergence 错误并退 1,永不落绿。
|
||||
不带 Lean 时状态如实停留在 unverified/checked(咨询性退 0),
|
||||
require_lean=True 才把未判定当作阻塞。
|
||||
|
||||
通过只说明所选文件判定完成,不能替代 test 任务的运行时验收;
|
||||
两者结合才是完整验收。
|
||||
|
||||
1. len(text.trim(`path`)) == 0
|
||||
Return *[要求明确的证明范围](result.error(code="dev.prove_path", message="path is required", operation="prove"))*.
|
||||
1. `all`
|
||||
We **argv = ["prove", "--all", `path`, "--max-files", text.num_text(`max_files`)]**.
|
||||
1. not `all`
|
||||
We **argv = ["prove", `path`]**.
|
||||
1. `require_lean`
|
||||
We **argv = [追加严格模式](table.append(list=`argv`, item="--require-lean"))**.
|
||||
1. `json`
|
||||
We **argv = [追加机器可读输出](table.append(list=`argv`, item="--json"))**.
|
||||
We **checked = [执行真实证明管线](engineering.run(argv=`argv`, timeout_ms=600000, max_output=4194304))**.
|
||||
验收结束后返回*[按真实退出状态判断结果并保留输出](engineering.check_exit(outcome=`checked`))*。
|
||||
```
|
||||
|
||||
## 5. 已知限制(Known limitations)
|
||||
|
||||
- `foreach` 维持 `proof.foreach_pending`(可证子集无循环;fold 编码 → 2.9)。
|
||||
- 跨模块调用 = 不透明操作;assume-guarantee 后置条件复用 → 2.9(依赖假设表 T3)。
|
||||
- 除零/模零:解释器为运行时错误,P1 判定库模型返回 `.int 0`(已记录偏差)。
|
||||
- 分支条件暂不支持字段访问(解析器文法限制;prove 侧已就绪,放开即生效)。
|
||||
@@ -2,11 +2,11 @@
|
||||
|
||||
| | |
|
||||
|---|---|
|
||||
| 状态 | **计划(Proposed)· 未开工** |
|
||||
| 状态 | **已实施(P0–P3 落地,2.8.0)**——落地记录:[design/prove-cli.md §12](../design/prove-cli.md);foreach / 跨模块 assume-guarantee / 假设表 T3 显式延期 2.9 |
|
||||
| 日期 | 2026-11-13 |
|
||||
| 设计 | [design/prove-cli.md](../design/prove-cli.md)(命令面、离线片段、定理表、语言习惯审查) |
|
||||
| 决策 | [ADR 0010](../adr/0010-skill-proof-boundary.md)(定理表角色、子集边界、顾问式定位) |
|
||||
| 目标版本 | CLI 2.8.0 |
|
||||
| 目标版本 | CLI 2.8.0(已发) |
|
||||
| 总工期 | 6.5 人周(单人;P0+P1 共 3 周可演示) |
|
||||
|
||||
---
|
||||
|
||||
@@ -98,7 +98,36 @@ marqdo ext remove llm
|
||||
- **标准库**:内置于二进制;磁盘 `lib/` 或 `MARQDO_LIB` 可覆盖。
|
||||
- **库路径解析**:`MARQDO_LIB` → 当前目录 `lib/` → 可执行文件旁 `lib/`。
|
||||
|
||||
## 13.8 下一步
|
||||
## 13.8 证明定理表:`prove`
|
||||
|
||||
知识文件里带定理表(`| 定理 | 结论 | 前提 | 策略 |`)的函数,可以用
|
||||
`marqdo prove` 生成 Lean 4 证明并自动判定真假:
|
||||
|
||||
```bash
|
||||
marqdo prove 审批.mq.md # 生成 Lean 模块 + 判定(有 Lean 时)
|
||||
marqdo prove 审批.mq.md --json # 智能体可读的单个 JSON 报告
|
||||
marqdo prove emit 审批.mq.md # 只生成不判定
|
||||
marqdo prove audit 审批.mq.md # 只审查可证子集,不写盘
|
||||
marqdo prove --all skills/ # 目录模式:逐文件出证,整体取最差
|
||||
marqdo prove clean # 清空 .marqdo/prove
|
||||
```
|
||||
|
||||
状态从强到弱:`proved`(Lean 证出)> `checked`(离线符号执行关闭)>
|
||||
`unverified`(超出可证子集,诚实未知)> `rejected`(子集验收拒绝)>
|
||||
`unproved`(反例)> `error`(判定层错误)。关键纪律:
|
||||
|
||||
- **没有 Lean 也能跑**:离线符号执行给出 `checked`,但它不是 proved——
|
||||
只是"在受限片段里符号地关掉了"。状态如实降级,绝不冒充证明。
|
||||
- **假定理必须红**:断言与执行语义矛盾的定理会得到反例(路径见证,
|
||||
如 `金额 = 10001`),退出码 1。
|
||||
- **单调性护栏**:离线说 `checked` 但 Lean 证不出(或反过来)→
|
||||
`error[proof.divergence]`,判定器缺陷优先于任何绿灯。
|
||||
- **证书缓存**:判定结果按源码+计划+库+版本哈希缓存于 `.marqdo/prove/`,
|
||||
重复运行毫秒级命中;`--no-cache` 绕过。
|
||||
|
||||
详见 [design/prove-cli.md](../design/prove-cli.md)(命令面、可证子集与判定管线)。
|
||||
|
||||
## 13.9 下一步
|
||||
|
||||
恭喜!你已经完成了 Marqdo 语法教程。你现在可以:
|
||||
|
||||
|
||||
@@ -0,0 +1,78 @@
|
||||
# 判定库 `MarqdoProof.lean` 的形式化说明
|
||||
|
||||
> 本文是 [design/prove-cli.md](../design/prove-cli.md) §4–§5 的配套理论摘要:
|
||||
> 生成文件依赖哪些库事实、离线判定凭什么与 Lean 单调一致、边界在哪里。
|
||||
> 库本体:[skill-proof/MarqdoProof.lean](skill-proof/MarqdoProof.lean)
|
||||
> (由 `marqdo prove` 内嵌携带,`scripts/check-proof-library.sh` 校验一致性)。
|
||||
|
||||
## 1. 语义模型
|
||||
|
||||
生成文件把每个函数降级为纯 Lean 定义:
|
||||
|
||||
```lean
|
||||
def «模块.函数» (env : Env) («轨迹» : List Op) («参数» : Val)… : Val × List Op
|
||||
```
|
||||
|
||||
- `Val`:8 形状和(none/int/num/bool/text/list/map + formula/code 边界),
|
||||
与解释器 `value::Value` 一一对应(T4 差分一致性测试锁定)。
|
||||
- `env : Env`:不透明操作的答案函数。**调用者的定理以 env 为全称量词**——
|
||||
证出即"对任意环境实现成立",这正是 assume-guarantee 的保守形态。
|
||||
- 不透明操作 = `Op.«调用_显示名»` 构造器 + 追加轨迹;操作名用链接显示文本,
|
||||
与定理表里 `包含(轨迹, "显示名")` 的写法一致。
|
||||
- 库函数面:`mqAdd/mqSub/mqMul/mqDiv`(int 面,除零→0,已记录偏差)、
|
||||
`mqGt/mqGe/mqLt/mqLe/mqEq`(decide 编码)、`mqAccess`(映射字段读取,
|
||||
方程友好)、`mqResultOk/mqResultError/mqValueOr`(lib/result 结构)。
|
||||
|
||||
## 2. 定理形状与判定
|
||||
|
||||
每条定理表行生成:
|
||||
|
||||
```lean
|
||||
theorem «模块.函数.定理名» (env : Env) («参数» : Val)… :
|
||||
(前提 → True) →
|
||||
((«结果», «轨迹») = «模块.函数» env [] «参数»…) →
|
||||
mqTruth «结论的 Lean 编码» := by
|
||||
intro … h_run
|
||||
unfold «模块.函数» at h_run
|
||||
try split at h_run -- 按最大 if 嵌套深度重复
|
||||
all_goals (simp_all [域引理]; try omega)
|
||||
```
|
||||
|
||||
- `decided_by = "lean"`:Lean 内核接受 ⇒ `proved`;`sorry` ⇒ `unverified`;
|
||||
报错 ⇒ `unproved`(错误指回定理表行)。
|
||||
- 策略 `none`:定理生成但主体 `sorry`,状态恒 `unverified`——绝不冒充。
|
||||
|
||||
## 3. 离线判定(F1–F5)与单调性论证
|
||||
|
||||
离线判定器对同一 IR 做符号执行,路径记录 (路径条件, 结果, 轨迹, 变量)。
|
||||
它与 Lean 的对应关系是**逐构造镜像**:
|
||||
|
||||
| 离线闭包理由 | Lean 侧对应 | 单调性依据 |
|
||||
|---|---|---|
|
||||
| F1 路径枚举穷尽 | `split at h_run` 产生的同一组目标 | 同一 IR 树,分支集合相同 |
|
||||
| F2 轨迹形状(包含/长度) | `(轨迹.map Op.name).contains` / `List.length` simp | 字面轨迹的列表运算两侧同构 |
|
||||
| F3 构造器结构求值 | `mqResultOk/mqResultError/mqValueOr` + `mqAccess` simp 方程 | 同一套库定义;结构字面归约是计算的 |
|
||||
| F4 自由原子 | split 后假设里的同一原子(`hypothesis` 关闭) | 只承认路径条件中字面出现的原子;路径反例含自由原子时放弃见证(保守) |
|
||||
| F5 区间字面重述 | split 后假设直接就是结论(`hypothesis`) | 字面匹配,无推理步 |
|
||||
| F5 区间违反(反例) | 无对应(Lean 同样证不出) | 反例只产生 `unproved`,不产生绿灯;见证 = 区间端点 |
|
||||
| 不可行路径跳过 | split 后假设矛盾(互斥比较引理 `.elim` + omega) | 前提 ∩ 路径条件为空 ⇒ 该分支假设集 Lean 侧同样矛盾 |
|
||||
|
||||
**单调性护栏**:有 Lean 时逐定理复核。离线 `checked` 而 Lean 关不掉,或
|
||||
离线反例被 Lean 证出 ⇒ `error[proof.divergence]`,永不落绿。护栏使
|
||||
"离线比 Lean 乐观"成为运行时可见错误而非静默错误;40 夹具语料库
|
||||
(`tests/prove_corpus.rs`)在 CI 强制 `offline_checked ⊆ lean_proved`。
|
||||
|
||||
**已知保守边界**(离线能、Lean 脚本不能,因此离线也必须不做):
|
||||
|
||||
- 区间**蕴含**(`金额 > 10000 ⊢ 金额 > 9999`):数学上真,但当前脚本没有
|
||||
Val 比较的传递引理,simp+omega 关不掉 ⇒ 离线同样 `unverified`
|
||||
(`f5_interval_closure_scope` 单测锁定)。库补传递引理后两侧同时放开。
|
||||
- 多变量线性、非线性、文本相等:超出 F5 单变量区间 ⇒ `unverified`;
|
||||
Lean 侧若能证则升级 `proved`(允许,这是"弱判定被强判定确认")。
|
||||
|
||||
## 4. 证书与缓存
|
||||
|
||||
证书记录:claim id、结论、策略、状态、`decided_by`(lean/trace/none)、
|
||||
证据(Lean 报错或离线见证)、库与模块 sha256、Lean 版本、CLI 版本。
|
||||
缓存键 = 上述全部的 sha256 ⇒ 任何一侧变化自动失效,命中即跳过判定。
|
||||
证书是**证据陈述**("这次判定是这样"),不是正确性保证——与设计 §0 契约 1 一致。
|
||||
@@ -33,9 +33,35 @@ def mqAccess : Val → String → Val
|
||||
|
||||
-- ---------------------------------------------------------------- 算术
|
||||
|
||||
/-- `+`:仅 int 参与算术;其它形状 → `none`(定型检查在源语言侧拒绝混合算术)。 -/
|
||||
/-- 显示串:标量与解释器 `Value::as_display` 对齐(T4 差分一致性)。
|
||||
容器做一层显示(值经 mqDisplayScalar),嵌套容器超出 `+` 片段——
|
||||
记录为有意简化,语料不依赖嵌套容器拼接。非递归 ⇒ 无终止性负担。 -/
|
||||
def mqDisplayScalar : Val → String
|
||||
| .int n => toString n
|
||||
| .num x => toString x
|
||||
| .bool b => if b then "True" else "False"
|
||||
| .text s => s
|
||||
| _ => "…"
|
||||
|
||||
def mqDisplay : Val → String
|
||||
| .int n => toString n
|
||||
| .num x => toString x
|
||||
| .bool b => if b then "True" else "False"
|
||||
| .text s => s
|
||||
| .list xs => "[" ++ String.intercalate ", " (xs.map mqDisplayScalar) ++ "]"
|
||||
| .map ps => "{" ++ String.intercalate ", " (ps.map (fun p => p.1 ++ ": " ++ mqDisplayScalar p.2)) ++ "}"
|
||||
| _ => "None"
|
||||
|
||||
/-- `+`:int 算术 + 文本拼接 + num 混合,与解释器 `eval_binary(Add)` 一致。
|
||||
偏差:解释器 i64 溢出报运行时错误;Lean `Int` 无界(语料只测界内值)。 -/
|
||||
def mqAdd : Val → Val → Val
|
||||
| .int a, .int b => .int (a + b)
|
||||
| .num a, .num b => .num (a + b)
|
||||
| .int a, .num b => .num (Float.ofInt a + b)
|
||||
| .num a, .int b => .num (a + Float.ofInt b)
|
||||
| .text a, .text b => .text (a ++ b)
|
||||
| .text a, b => .text (a ++ mqDisplay b)
|
||||
| a, .text b => .text (mqDisplay a ++ b)
|
||||
| _, _ => .none
|
||||
|
||||
def mqSub : Val → Val → Val
|
||||
|
||||
@@ -3,3 +3,30 @@
|
||||
The maintained programming rules and runnable examples live in [the executable programming guide](../../doc/library-development.mq.md). Read its parameter and return contracts, then call the actual source heading. [The development entry](../marqdo-dev/index.mq.md) provides native file operations and verification tasks; [the release entry](../marqdo-release/index.mq.md) provides publication workflows.
|
||||
|
||||
Use [the recursive Markdown web example](../../examples/web-layout-handbook/index.mq.md) and [the quantum workflow](../../examples/quantum-entanglement/index.mq.md) for domain development. Their real source is the manual. [Classic migration context](migration-classic.md) remains for explicitly historical syntax; new programs use readable occlusion and access(collection, key). Do not copy a second fenced program from this discovery note.
|
||||
|
||||
## Theorem tables (`marqdo prove`, CLI 2.8.0+)
|
||||
|
||||
A knowledge function may carry a theorem table before its first executable
|
||||
statement. Place it right under the parameter list:
|
||||
|
||||
| 定理 | 结论 | 前提 | 策略 |
|
||||
|---|---|---|---|
|
||||
| 大额必经人工 | 包含(轨迹, "工单.创建") | 金额 > 10000 | auto |
|
||||
|
||||
- `结论` grammar: `结果.ok = True/False`, `包含(轨迹, "操作名")`,
|
||||
`长度(轨迹) ⋈ n`, `参数 ⋈ 常量`, combined with 且/或/非. `轨迹` names the
|
||||
operation trace by the link's display text; `结果` is the returned value.
|
||||
- `前提` restricts the claim's scope (same comparison grammar). Keep it a
|
||||
single-variable linear condition.
|
||||
- `策略` is `auto` (generated proof script) or `none` (theorem generated but
|
||||
not proven — honest `unverified`).
|
||||
|
||||
Judge with `marqdo prove 文件.mq.md --json`. Statuses, strongest first:
|
||||
`proved` (Lean confirms) > `checked` (offline symbolic execution closed the
|
||||
path — advisory, never a proof) > `unverified` (outside the provable subset)
|
||||
> `rejected` (subset check refused) > `unproved` (counterexample found, e.g.
|
||||
`金额 = 10001`) > `error`. Without a Lean toolchain the run stays honest at
|
||||
`checked`/`unverified` and exits 0; `--require-lean` makes unjudged statuses
|
||||
exit 2 for CI. Never weaken a failing claim to make it pass — fix the claim
|
||||
or the code, and let `proof.divergence` stop the line when the offline
|
||||
judger and Lean disagree. Full contract: [design/prove-cli.md](../../doc/design/prove-cli.md).
|
||||
+131
@@ -552,6 +552,19 @@ enum Commands {
|
||||
/// maxHeartbeats for lean (default: lean's own)
|
||||
#[arg(long, value_name = "N")]
|
||||
heartbeats: Option<u64>,
|
||||
|
||||
/// Directory mode: recursively prove every .mq.md (aggregated, worst status wins)
|
||||
#[arg(long)]
|
||||
all: bool,
|
||||
|
||||
/// Cap for --all file collection (default 200)
|
||||
#[arg(long = "max-files", value_name = "N", default_value_t = 200)]
|
||||
max_files: usize,
|
||||
|
||||
/// Debug fault injection: `offline-claim` flips one Lean-proved claim to
|
||||
/// force proof.divergence (monotonicity tripwire verification)
|
||||
#[arg(long = "fault-inject", value_name = "SPEC")]
|
||||
fault_inject: Option<String>,
|
||||
},
|
||||
}
|
||||
|
||||
@@ -774,6 +787,85 @@ fn main() -> ExitCode {
|
||||
}
|
||||
}
|
||||
|
||||
/// `prove --all`:逐文件出报告,整体取最差状态(差者胜,§3.2)。
|
||||
/// JSON:`{"schema":1,"kind":"prove-all","status":…,"files":[…],"duration_ms":…}`。
|
||||
fn run_prove_all(
|
||||
dir: &std::path::Path,
|
||||
max_files: usize,
|
||||
opts: &marqdo::prove::ProveOptions,
|
||||
json: bool,
|
||||
require_lean: bool,
|
||||
) -> Result<i32> {
|
||||
let mut files = collect_mq_files(dir)?;
|
||||
anyhow::ensure!(
|
||||
!files.is_empty(),
|
||||
"目录中没有 .mq.md 文件:{}",
|
||||
dir.display()
|
||||
);
|
||||
anyhow::ensure!(
|
||||
files.len() <= max_files,
|
||||
"--max-files {max_files} 已达上限(实际 {} 个文件;请调大或缩小范围)",
|
||||
files.len()
|
||||
);
|
||||
files.sort();
|
||||
let started = std::time::Instant::now();
|
||||
let mut worst = marqdo::prove::Verdict::Proved;
|
||||
let mut file_items: Vec<serde_json::Value> = Vec::new();
|
||||
let mut text_rows: Vec<(String, marqdo::prove::Verdict, usize)> = Vec::new();
|
||||
for f in &files {
|
||||
let report = marqdo::prove::prove_path(f, opts)?;
|
||||
for d in &report.diagnostics {
|
||||
eprintln!("{}", d.format_message());
|
||||
}
|
||||
let claims: usize = report
|
||||
.functions
|
||||
.iter()
|
||||
.map(|fun| fun.claims.len())
|
||||
.sum();
|
||||
worst = marqdo::prove::Verdict::worst(worst, report.status);
|
||||
text_rows.push((
|
||||
f.strip_prefix(dir).unwrap_or(f).display().to_string(),
|
||||
report.status,
|
||||
claims,
|
||||
));
|
||||
let mut item = report.to_json();
|
||||
if let Some(obj) = item.as_object_mut() {
|
||||
obj.insert(
|
||||
"file".into(),
|
||||
serde_json::Value::String(
|
||||
f.strip_prefix(dir).unwrap_or(f).display().to_string(),
|
||||
),
|
||||
);
|
||||
}
|
||||
file_items.push(item);
|
||||
}
|
||||
let duration_ms = started.elapsed().as_millis() as u64;
|
||||
if json {
|
||||
let agg = serde_json::json!({
|
||||
"schema": 1,
|
||||
"kind": "prove-all",
|
||||
"root": dir.display().to_string(),
|
||||
"status": worst.as_str(),
|
||||
"files": file_items,
|
||||
"file_count": files.len(),
|
||||
"duration_ms": duration_ms,
|
||||
});
|
||||
println!("{agg}");
|
||||
} else {
|
||||
println!(
|
||||
"prove --all {}:{} 个文件,{} ms,整体 {}",
|
||||
dir.display(),
|
||||
files.len(),
|
||||
duration_ms,
|
||||
worst.as_str()
|
||||
);
|
||||
for (rel, status, claims) in &text_rows {
|
||||
println!(" {rel}: {}(定理 {claims})", status.as_str());
|
||||
}
|
||||
}
|
||||
Ok(worst.exit_code(require_lean))
|
||||
}
|
||||
|
||||
/// Resolve the value-path spelling introduced for executable knowledge without
|
||||
/// evaluating it. The import bind comes from the caller's source file, so
|
||||
/// `math.pi` never degrades into a project-wide name guess.
|
||||
@@ -1101,7 +1193,14 @@ fn try_main(cli: Cli) -> Result<i32> {
|
||||
timeout,
|
||||
no_cache,
|
||||
heartbeats,
|
||||
all,
|
||||
max_files,
|
||||
fault_inject,
|
||||
} => {
|
||||
// 调试注入(§6 A3):仅面向测试的分歧触发器。
|
||||
if fault_inject.as_deref() == Some("offline-claim") {
|
||||
std::env::set_var("MARQDO_PROOF_INJECT_DIVERGENCE", "1");
|
||||
}
|
||||
// 子命令参数覆盖顶层默认;默认动作 = emit + 报告。
|
||||
let (file, action_json, action_emit, action_only, action_opaque, action_relax) =
|
||||
match &action {
|
||||
@@ -1154,6 +1253,38 @@ fn try_main(cli: Cli) -> Result<i32> {
|
||||
// emit 子命令只生成不判定(§8.2)。
|
||||
let judge = !matches!(action, Some(ProveAction::Emit { .. }));
|
||||
let no_cache = no_cache || std::env::var_os("MARQDO_PROOF_NO_CACHE").is_some();
|
||||
// 目录必须显式 --all(§3.2 proof.need_all)。
|
||||
if file.is_dir() {
|
||||
if !all {
|
||||
eprintln!(
|
||||
"error: {} 是目录:逐文件证明请加 --all(--max-files 默认 200)",
|
||||
file.display()
|
||||
);
|
||||
return Ok(1);
|
||||
}
|
||||
let out_json = json || action_json;
|
||||
let code = run_prove_all(
|
||||
&file,
|
||||
max_files,
|
||||
&marqdo::prove::ProveOptions {
|
||||
entry: None,
|
||||
only: action_only,
|
||||
allow_opaque: action_opaque,
|
||||
emit_dir: emit_target,
|
||||
strict_types: !action_relax,
|
||||
require_lean,
|
||||
report_only,
|
||||
judge,
|
||||
no_cache,
|
||||
lean,
|
||||
timeout: std::time::Duration::from_secs(timeout),
|
||||
max_heartbeats: heartbeats,
|
||||
},
|
||||
out_json,
|
||||
require_lean,
|
||||
)?;
|
||||
return Ok(code);
|
||||
}
|
||||
let opts = marqdo::prove::ProveOptions {
|
||||
entry: None,
|
||||
only: action_only,
|
||||
|
||||
@@ -0,0 +1,481 @@
|
||||
//! T4 差分一致性(roadmap §6 B 组):解释器求值 == Lean 生成语义。
|
||||
//!
|
||||
//! 对每个纯函数语料:
|
||||
//! 1. 解释器跑 `marqdo run --call-lib-path` 拿到真实返回值;
|
||||
//! 2. `prove emit` 生成 Lean 定义,追加 `example : … = «期望» := by simp`;
|
||||
//! 3. 真 Lean 校验通过 ⇒ 生成器的 Val 语义与解释器一致(B1/B3/B4/B5)。
|
||||
//!
|
||||
//! 已记录偏差(不进本测试):除零/模零——解释器为运行时错误,P1 库模型
|
||||
//! `mqDiv`/`mqMod` 返回 `.int 0`(MarqdoProof.lean 注释与设计 §12 偏差表)。
|
||||
//! 无 Lean 工具链时整组 SKIP(MARQDO_PROOF_REQUIRE_LEAN=1 则失败)。
|
||||
|
||||
use std::path::{Path, PathBuf};
|
||||
use std::process::Command;
|
||||
|
||||
/// 语料:模块名 / 函数名 / 参数 JSON / 期望 Val 的 Lean 字面。
|
||||
struct Case {
|
||||
module: &'static str,
|
||||
function: &'static str,
|
||||
source: &'static str,
|
||||
args: &'static str,
|
||||
/// Lean 期望值(`Val.*` 构造)。
|
||||
expected: &'static str,
|
||||
/// 额外 simp 引理(默认集之外)。
|
||||
extra_simp: &'static [&'static str],
|
||||
}
|
||||
|
||||
const CASES: &[Case] = &[
|
||||
Case {
|
||||
module: "算术",
|
||||
function: "加倍",
|
||||
source: r#"---
|
||||
markup: occlusion
|
||||
---
|
||||
|
||||
## 加倍
|
||||
+ `x`
|
||||
|
||||
**y = `x` + `x`**
|
||||
|
||||
*y*
|
||||
"#,
|
||||
args: r#"{"x": 21}"#,
|
||||
expected: "Val.int 42",
|
||||
extra_simp: &[],
|
||||
},
|
||||
Case {
|
||||
module: "计算器",
|
||||
function: "加法",
|
||||
source: r#"---
|
||||
markup: occlusion
|
||||
---
|
||||
|
||||
## 加法
|
||||
+ `甲`
|
||||
+ `乙`
|
||||
|
||||
**和 = `甲` + `乙`**
|
||||
|
||||
*和*
|
||||
"#,
|
||||
args: r#"{"甲": -5, "乙": 2}"#,
|
||||
expected: "Val.int (-3)",
|
||||
extra_simp: &[],
|
||||
},
|
||||
Case {
|
||||
module: "比较",
|
||||
function: "判断",
|
||||
source: r#"---
|
||||
markup: occlusion
|
||||
---
|
||||
|
||||
## 判断
|
||||
+ `x`
|
||||
|
||||
1. `x` > 10
|
||||
**结论 = "大"**
|
||||
2. *
|
||||
**结论 = "小"**
|
||||
|
||||
*结论*
|
||||
"#,
|
||||
args: r#"{"x": 11}"#,
|
||||
expected: r#"Val.text "大""#,
|
||||
extra_simp: &[],
|
||||
},
|
||||
Case {
|
||||
module: "工单",
|
||||
function: "创建",
|
||||
source: r#"---
|
||||
markup: occlusion
|
||||
---
|
||||
|
||||
## 创建
|
||||
+ `id`
|
||||
|
||||
**单号 = "T-" + `id`**
|
||||
|
||||
*单号*
|
||||
"#,
|
||||
args: r#"{"id": 7}"#,
|
||||
expected: r#"Val.text "T-7""#,
|
||||
extra_simp: &[],
|
||||
},
|
||||
Case {
|
||||
module: "边界",
|
||||
function: "极限",
|
||||
source: r#"---
|
||||
markup: occlusion
|
||||
---
|
||||
|
||||
## 极限
|
||||
+ `x`
|
||||
|
||||
**y = `x` + 1**
|
||||
|
||||
*y*
|
||||
"#,
|
||||
args: r#"{"x": 9223372036854775806}"#,
|
||||
expected: "Val.int 9223372036854775807",
|
||||
extra_simp: &[],
|
||||
},
|
||||
];
|
||||
|
||||
/// 直接驱动解释器调用 `模块.函数`(run_file_capture 不处理 call_lib_path)。
|
||||
fn call_function(file: &Path, lib_path: &str, args_json: &str) -> marqdo::value::Value {
|
||||
let module = marqdo::load::load_module(file).expect("load module");
|
||||
let mut host = marqdo::host::HostContext::for_capture(Some(file), marqdo::host::HostCaps {
|
||||
fs_write: false,
|
||||
exec: false,
|
||||
net: false,
|
||||
plugin: false,
|
||||
});
|
||||
let mut interp = marqdo::interp::Interpreter::with_capture(Some(file), false).with_host(host);
|
||||
let args: marqdo::value::Value = serde_json::from_str(args_json)
|
||||
.map(|v: serde_json::Value| json_to_value(&v))
|
||||
.expect("args json");
|
||||
interp
|
||||
.run_site_lib_path(&module, lib_path, Some(&args))
|
||||
.expect("interpreter call")
|
||||
}
|
||||
|
||||
/// JSON → Value(T4 语料只用到 int/text/bool/map)。
|
||||
fn json_to_value(v: &serde_json::Value) -> marqdo::value::Value {
|
||||
use marqdo::value::Value;
|
||||
match v {
|
||||
serde_json::Value::Number(n) => {
|
||||
if let Some(i) = n.as_i64() {
|
||||
Value::Int(i)
|
||||
} else {
|
||||
Value::Num(n.as_f64().unwrap_or(0.0))
|
||||
}
|
||||
}
|
||||
serde_json::Value::String(s) => Value::Text(s.clone()),
|
||||
serde_json::Value::Bool(b) => Value::Bool(*b),
|
||||
serde_json::Value::Object(m) => Value::Map(
|
||||
m.iter()
|
||||
.map(|(k, v)| (k.clone(), json_to_value(v)))
|
||||
.collect(),
|
||||
),
|
||||
_ => Value::None,
|
||||
}
|
||||
}
|
||||
|
||||
fn lean_binary() -> Option<PathBuf> {
|
||||
if let Ok(p) = std::env::var("MARQDO_PROOF_LEAN") {
|
||||
let p = PathBuf::from(p);
|
||||
if p.exists() {
|
||||
return Some(p);
|
||||
}
|
||||
}
|
||||
let path = std::env::var_os("PATH")?;
|
||||
std::env::split_paths(&path).map(|d| d.join("lean")).find(|p| p.is_file())
|
||||
}
|
||||
|
||||
fn require_lean_enforced() -> bool {
|
||||
std::env::var("MARQDO_PROOF_REQUIRE_LEAN").is_ok_and(|v| v == "1")
|
||||
}
|
||||
|
||||
fn target_dir() -> PathBuf {
|
||||
std::env::var_os("CARGO_TARGET_DIR")
|
||||
.map(PathBuf::from)
|
||||
.unwrap_or_else(|| PathBuf::from(env!("CARGO_MANIFEST_DIR")).join("target"))
|
||||
}
|
||||
|
||||
/// 源文件中的函数体生成的 Lean 定义名(与 emit 一致)。
|
||||
fn def_name(_module: &str, function: &str) -> String {
|
||||
format!("«{function}»")
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn t4_interpreter_matches_lean_semantics() {
|
||||
let Some(lean) = lean_binary() else {
|
||||
if require_lean_enforced() {
|
||||
panic!("MARQDO_PROOF_REQUIRE_LEAN=1 但没有 Lean 工具链");
|
||||
}
|
||||
println!("SKIP conformance(无 Lean 工具链)");
|
||||
return;
|
||||
};
|
||||
let root = target_dir().join("conformance");
|
||||
let _ = std::fs::remove_dir_all(&root);
|
||||
std::fs::create_dir_all(&root).expect("scratch");
|
||||
|
||||
let toolchain = marqdo::prove::driver::Toolchain::discover(Some(&lean))
|
||||
.expect("toolchain discover")
|
||||
.expect("lean present");
|
||||
let (library_dir, _sha) = marqdo::prove::driver::ensure_library(&root, &toolchain)
|
||||
.expect("build library");
|
||||
let mut lean_env = std::env::split_paths(&std::env::var_os("LEAN_PATH").unwrap_or_default())
|
||||
.collect::<Vec<_>>();
|
||||
lean_env.push(library_dir);
|
||||
|
||||
let mut failures: Vec<String> = Vec::new();
|
||||
for case in CASES {
|
||||
let dir = root.join(case.module);
|
||||
std::fs::create_dir_all(&dir).expect("case dir");
|
||||
let lib_file = dir.join(format!("{}.mq.md", case.module));
|
||||
std::fs::write(&lib_file, case.source).expect("write source");
|
||||
// 主文件只做导入:解释器经 import bind 调库函数。
|
||||
let main_file = dir.join("主.mq.md");
|
||||
std::fs::write(
|
||||
&main_file,
|
||||
format!(
|
||||
"---\nmarkup: occlusion\nimport:\n {}: {}.mq.md\n---\n\n# 主\n",
|
||||
case.module, case.module
|
||||
),
|
||||
)
|
||||
.expect("write main");
|
||||
let file = main_file;
|
||||
let args_file = dir.join("args.json");
|
||||
std::fs::write(&args_file, case.args).expect("write args");
|
||||
|
||||
// ① 解释器求值。
|
||||
let lib_path = format!("{}.{}", case.module, case.function);
|
||||
let value = match std::panic::catch_unwind(|| {
|
||||
call_function(&file, &lib_path, case.args)
|
||||
}) {
|
||||
Ok(v) => v,
|
||||
Err(_) => {
|
||||
failures.push(format!("{}: 解释器调用失败", case.function));
|
||||
continue;
|
||||
}
|
||||
};
|
||||
|
||||
// ② 生成 Lean 定义。
|
||||
let emit_dir = dir.join("out");
|
||||
let prove_opts = marqdo::prove::ProveOptions {
|
||||
emit_dir: Some(emit_dir.clone()),
|
||||
judge: false,
|
||||
report_only: false,
|
||||
..Default::default()
|
||||
};
|
||||
if let Err(e) = marqdo::prove::prove_path(&lib_file, &prove_opts) {
|
||||
failures.push(format!("{}: 生成失败:{e}", case.function));
|
||||
continue;
|
||||
}
|
||||
// 文件名与 emit 一致:非 ASCII 字符确定性编码为 uXXXX。
|
||||
let stem: String = case
|
||||
.module
|
||||
.chars()
|
||||
.map(|c| {
|
||||
if c.is_ascii_alphanumeric() || c == '_' {
|
||||
c.to_string()
|
||||
} else if !c.is_ascii() {
|
||||
format!("u{:04X}", c as u32)
|
||||
} else {
|
||||
"_".to_string()
|
||||
}
|
||||
})
|
||||
.collect();
|
||||
let generated = emit_dir.join("Generated").join(format!("{stem}_mq.lean"));
|
||||
let module_src = std::fs::read_to_string(&generated).expect("generated module");
|
||||
|
||||
let simp_set = {
|
||||
let mut base = "mqTruth, Op.name, mqAccess, mqResultOk, mqResultError, mqValueOr, mqAnd, mqOr, mqNot, mqAdd, mqSub, mqMul, mqEq".to_string();
|
||||
for e in case.extra_simp {
|
||||
base.push_str(", ");
|
||||
base.push_str(e);
|
||||
}
|
||||
base
|
||||
};
|
||||
// 实参按生成定义的形参顺序(扫描 def 签名的顶层括号组)。
|
||||
let def_line = module_src
|
||||
.lines()
|
||||
.find(|l| {
|
||||
l.starts_with(&format!("def {} ", def_name(case.module, case.function)))
|
||||
&& l.contains('(')
|
||||
})
|
||||
.unwrap_or("");
|
||||
let mut params: Vec<String> = Vec::new();
|
||||
let mut depth = 0usize;
|
||||
let mut cur = String::new();
|
||||
for ch in def_line.chars() {
|
||||
match ch {
|
||||
'(' => {
|
||||
depth += 1;
|
||||
cur.clear();
|
||||
}
|
||||
')' => {
|
||||
depth = depth.saturating_sub(1);
|
||||
if depth == 0 {
|
||||
let inner = cur.trim();
|
||||
let name = inner.split(':').next().unwrap_or("").trim();
|
||||
let bare = name
|
||||
.strip_prefix("«")
|
||||
.and_then(|s| s.strip_suffix("»"))
|
||||
.unwrap_or(name);
|
||||
if bare != "env" && bare != "轨迹" && !bare.is_empty() {
|
||||
params.push(bare.to_string());
|
||||
}
|
||||
}
|
||||
cur.clear();
|
||||
}
|
||||
_ => {
|
||||
if depth > 0 {
|
||||
cur.push(ch);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
let args_map: serde_json::Map<String, serde_json::Value> =
|
||||
serde_json::from_str(case.args).expect("args object");
|
||||
let arg_terms = params
|
||||
.iter()
|
||||
.map(|name| {
|
||||
let v = args_map.get(name).cloned().unwrap_or(serde_json::Value::Null);
|
||||
match v.as_i64() {
|
||||
Some(i) if i < 0 => format!("(Val.int ({i}))"),
|
||||
Some(i) => format!("(Val.int {i})"),
|
||||
None => match v.as_str() {
|
||||
Some(s) => format!("(Val.text \"{s}\")"),
|
||||
None => "(Val.none)".to_string(),
|
||||
},
|
||||
}
|
||||
})
|
||||
.collect::<Vec<_>>()
|
||||
.join(" ");
|
||||
// ③ 期望 = 语料声明的 Lean 字面(与解释器求值在 ① 已对照)。
|
||||
let expected = case.expected.to_string();
|
||||
let mut conformance = module_src.clone();
|
||||
let example = format!(
|
||||
"\n-- T4:解释器期望值({} 返回 {:?})\nexample : ({} env [] {arg_terms}).1 = {expected} := by\n simp [{}, {}]\n try rfl\n",
|
||||
case.function,
|
||||
value,
|
||||
def_name(case.module, case.function),
|
||||
def_name(case.module, case.function),
|
||||
simp_set,
|
||||
);
|
||||
if let Some(pos) = conformance.rfind("\nend Generated") {
|
||||
conformance.insert_str(pos, &example);
|
||||
} else {
|
||||
conformance.push_str(&example);
|
||||
}
|
||||
let conformance_file = dir.join("Conformance.lean");
|
||||
std::fs::write(&conformance_file, &conformance).expect("write conformance");
|
||||
|
||||
// ④ Lean 校验。
|
||||
let out = Command::new(&lean)
|
||||
.current_dir(&dir)
|
||||
.env(
|
||||
"LEAN_PATH",
|
||||
std::env::join_paths(&lean_env).expect("lean path"),
|
||||
)
|
||||
.arg(conformance_file.strip_prefix(&dir).unwrap_or(&conformance_file))
|
||||
.output()
|
||||
.expect("run lean");
|
||||
if !out.status.success() {
|
||||
failures.push(format!(
|
||||
"{}: Lean 不接受解释器期望值({:?}):\n-- stderr:\n{}\n-- stdout:\n{}",
|
||||
case.function,
|
||||
value,
|
||||
String::from_utf8_lossy(&out.stderr),
|
||||
String::from_utf8_lossy(&out.stdout)
|
||||
));
|
||||
}
|
||||
}
|
||||
assert!(
|
||||
failures.is_empty(),
|
||||
"T4 差分不一致 {} 项:\n{}",
|
||||
failures.len(),
|
||||
failures.join("\n----------------------------------------\n")
|
||||
);
|
||||
println!("conformance: {} 例全部一致(解释器 == Lean)", CASES.len());
|
||||
}
|
||||
|
||||
/// B4 补充:结果结构(result.ok)字段读取与解释器一致。
|
||||
#[test]
|
||||
fn t4_result_struct_field_access() {
|
||||
let Some(lean) = lean_binary() else {
|
||||
if require_lean_enforced() {
|
||||
panic!("MARQDO_PROOF_REQUIRE_LEAN=1 但没有 Lean 工具链");
|
||||
}
|
||||
println!("SKIP conformance(无 Lean 工具链)");
|
||||
return;
|
||||
};
|
||||
let root = target_dir().join("conformance-result");
|
||||
let _ = std::fs::remove_dir_all(&root);
|
||||
std::fs::create_dir_all(&root).expect("scratch");
|
||||
|
||||
let source = r#"---
|
||||
markup: occlusion
|
||||
import:
|
||||
result: lib/result.mq.md
|
||||
---
|
||||
|
||||
## 登记
|
||||
+ `金额`
|
||||
|
||||
**回执 = [登记](result.ok(value=`金额`))**
|
||||
|
||||
*access(回执, "value")*
|
||||
"#;
|
||||
let dir = &root;
|
||||
let lib_file = dir.join("台.mq.md");
|
||||
std::fs::write(&lib_file, source).expect("write");
|
||||
let file = dir.join("主.mq.md");
|
||||
std::fs::write(
|
||||
&file,
|
||||
"---\nmarkup: occlusion\nimport:\n 台: 台.mq.md\n---\n\n# 主\n",
|
||||
)
|
||||
.expect("write main");
|
||||
std::fs::create_dir_all(dir.join("lib")).expect("lib");
|
||||
std::fs::write(
|
||||
dir.join("lib/result.mq.md"),
|
||||
std::fs::read_to_string(Path::new(env!("CARGO_MANIFEST_DIR")).join("lib/result.mq.md"))
|
||||
.expect("repo result lib"),
|
||||
)
|
||||
.expect("write result lib");
|
||||
let args_file = dir.join("args.json");
|
||||
std::fs::write(&args_file, r#"{"金额": 5}"#).expect("args");
|
||||
|
||||
let value = call_function(&file, "台.登记", r#"{"金额": 5}"#);
|
||||
// access(回执, "value") 应取回 5(ok/value 字段结构在 Lean 侧由
|
||||
// mqAccess (mqResultOk …) 方程锁定,两侧一致即 B4)。
|
||||
assert!(
|
||||
matches!(&value, marqdo::value::Value::Int(5)),
|
||||
"access(result.ok(5), \"value\") 应为 Int(5),实际 {value:?}"
|
||||
);
|
||||
|
||||
// Lean 侧同一表达式:mqResultOk 5 的 value 字段 == 5。
|
||||
let prove_opts = marqdo::prove::ProveOptions {
|
||||
emit_dir: Some(dir.join("out")),
|
||||
judge: false,
|
||||
report_only: false,
|
||||
..Default::default()
|
||||
};
|
||||
marqdo::prove::prove_path(&lib_file, &prove_opts).expect("emit");
|
||||
let generated = dir.join("out/Generated/u53F0_mq.lean"); // 台 → u53F0
|
||||
let module_src = std::fs::read_to_string(&generated).expect("generated");
|
||||
let mut conformance = module_src;
|
||||
conformance.push_str(
|
||||
"\n-- T4 B4:结构字段与解释器一致\nexample : (mqResultOk (Val.int 5)).2.1 ≡ Val.int 5 := rfl\n",
|
||||
);
|
||||
// mqResultOk 的 map 字段顺序依赖库实现,直接用 mqAccess 更稳:
|
||||
let conformance = conformance.replace(
|
||||
"example : (mqResultOk (Val.int 5)).2.1 ≡ Val.int 5 := rfl\n",
|
||||
"example : mqAccess (mqResultOk (Val.int 5)) \"value\" = Val.int 5 := by\n simp [mqResultOk, mqAccess]\n",
|
||||
);
|
||||
let conformance_file = dir.join("Conformance.lean");
|
||||
std::fs::write(&conformance_file, &conformance).expect("write");
|
||||
|
||||
let toolchain = marqdo::prove::driver::Toolchain::discover(Some(&lean))
|
||||
.expect("discover")
|
||||
.expect("lean");
|
||||
let (library_dir, _) =
|
||||
marqdo::prove::driver::ensure_library(&root, &toolchain).expect("library");
|
||||
let mut lean_env = vec![library_dir];
|
||||
if let Some(p) = std::env::var_os("LEAN_PATH") {
|
||||
lean_env.extend(std::env::split_paths(&p));
|
||||
}
|
||||
let out = Command::new(&lean)
|
||||
.current_dir(dir)
|
||||
.env("LEAN_PATH", std::env::join_paths(&lean_env).expect("paths"))
|
||||
.arg("Conformance.lean")
|
||||
.output()
|
||||
.expect("run lean");
|
||||
assert!(
|
||||
out.status.success(),
|
||||
"B4 结构字段不一致:\n{}",
|
||||
String::from_utf8_lossy(&out.stderr)
|
||||
);
|
||||
}
|
||||
@@ -389,6 +389,64 @@ fn e4_tactic_none_yields_unverified() {
|
||||
assert_eq!(out.status.code(), Some(0), "advisory:unverified 不阻塞");
|
||||
}
|
||||
|
||||
/// A6:`--all` 目录模式——整体取最差(差者胜),目录不带 --all 拒绝。
|
||||
#[test]
|
||||
fn a6_all_aggregates_worst_status_and_exit() {
|
||||
if !toolchain_available() {
|
||||
skip_or_fail("a6_all");
|
||||
return;
|
||||
}
|
||||
let s = Scratch::new("a6_all");
|
||||
let lean_path = install_fake_lean(&s.dir);
|
||||
s.write("good.mq.md", LEDGER_OK);
|
||||
// 坏定理文件:unproved(LEDGER_BAD 是 `长度(轨迹) = 0`,真实执行 1 步)。
|
||||
let bad = LEDGER_BAD.replace("台账", "坏台账");
|
||||
s.write("坏/坏.mq.md", &bad);
|
||||
|
||||
// 目录不带 --all → 退 1。
|
||||
let dir_out = s
|
||||
.marqdo()
|
||||
.arg("prove")
|
||||
.arg(".")
|
||||
.output()
|
||||
.expect("run prove dir");
|
||||
let dir_err = String::from_utf8_lossy(&dir_out.stderr);
|
||||
assert!(dir_err.contains("是目录"), "{dir_err}");
|
||||
assert_eq!(dir_out.status.code(), Some(1));
|
||||
|
||||
// --all:假 lean 恒报 proved,坏定理离线反例与之分歧 → 护栏 error(§6 A3
|
||||
// 在目录模式同样生效);整体取最差 = error,退 1。
|
||||
let all_out = s
|
||||
.marqdo()
|
||||
.arg("prove")
|
||||
.arg("--all")
|
||||
.arg(".")
|
||||
.arg("--lean")
|
||||
.arg(&lean_path)
|
||||
.arg("--json")
|
||||
.output()
|
||||
.expect("run prove --all");
|
||||
let json: serde_json::Value =
|
||||
serde_json::from_slice(&all_out.stdout).expect("prove-all json");
|
||||
assert_eq!(json["kind"], "prove-all");
|
||||
assert_eq!(json["status"], "error", "{json}");
|
||||
assert_eq!(json["file_count"], 2);
|
||||
assert_eq!(all_out.status.code(), Some(1), "整体最差 error 应退 1");
|
||||
let files = json["files"].as_array().expect("files");
|
||||
let bad_file = files
|
||||
.iter()
|
||||
.find(|f| f["file"].as_str().is_some_and(|n| n.contains("坏")))
|
||||
.expect("bad file row");
|
||||
assert_eq!(bad_file["status"], "error");
|
||||
let codes: Vec<&str> = bad_file["diagnostics"]
|
||||
.as_array()
|
||||
.expect("diagnostics")
|
||||
.iter()
|
||||
.filter_map(|d| d["code"].as_str())
|
||||
.collect();
|
||||
assert!(codes.contains(&"proof.divergence"), "{codes:?}");
|
||||
}
|
||||
|
||||
/// 真定理 → proved(冒烟:真 Lean 全链路)。
|
||||
#[test]
|
||||
fn a1_true_claim_proved() {
|
||||
|
||||
Reference in new issue
Block a user