From fa832268b325822d2386b9c7572a84ec38a28a6a Mon Sep 17 00:00:00 2001 From: cflmy Date: Mon, 16 Nov 2026 17:26:22 +0800 Subject: [PATCH] prove: fix issues found in Lean verification round MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - p2_len fixture claim corrected: pure function's trace is always empty, so '长度(轨迹) = 2' was an authoring mistake -> '长度(轨迹) = 0' (Nat-path coverage unchanged); all 7 claim-bearing files now prove - generated Lean headers set linter.unusedVariables/unusedSimpArgs off: machine-emitted artifacts should not trigger human-authoring lints (0 warnings now) - verification driver script + updated results in design doc 12 and library README --- doc/design/prove-cli.md | 11 +++-- doc/verification/skill-proof/README.md | 9 ++-- .../skill-proof/run-lean-check.sh | 49 +++++++++++++++++++ src/prove/emit.rs | 4 ++ tests/prove.rs | 6 +-- 5 files changed, 69 insertions(+), 10 deletions(-) create mode 100755 doc/verification/skill-proof/run-lean-check.sh diff --git a/doc/design/prove-cli.md b/doc/design/prove-cli.md index ac2d39a..f5b5af1 100644 --- a/doc/design/prove-cli.md +++ b/doc/design/prove-cli.md @@ -820,8 +820,13 @@ prove 生成 → `lake env lean` → `proved / unproved / lean-error` 三态分 | 目标 | 结果 | |---|---| | 业务样例 `审批.mq.md`(3 定理:大额必经人工 / 恰两步轨迹 / 必有回执) | `lake env lean` 0 error,全部 **proved** | -| 6 个带定理的测试夹具(p1_theorem / a_deterministic / d_opaque_op / f_continuation / p2_arity / p2_len) | 5 个 **proved**;`p2_len`(计数器)如实 **unproved**——其断言 `长度(轨迹) = 2` 为假命题(函数零操作,轨迹恒空),Lean 拒绝即正确行为 | +| 6 个带定理的测试夹具 + 业务样例 | **7/7 全部 proved**(`p2_len` 夹具的断言已修正:纯函数轨迹恒空,`长度(轨迹) = 2` 是作者笔误,改为 `长度(轨迹) = 0`——Nat 比较路径覆盖不变) | | 全量 Rust 测试 | 499 通过;仅剩 2 个已记录的环境受限失败(python3 缺失 / playwright root 属主),与 prove 改动无关 | -诚实性核对:`p2_len` 的 unproved 正是"绝不 lean 无法证明的 marqdo 可以证明"的落实—— -生成侧把语法合法但语义为假的断言如实交给 Lean,判定侧不会虚报 proved。 +诚实性核对(首轮实测发现后修正):`p2_len` 曾因断言写错(纯函数却断言两步轨迹) +被 Lean 如实拒绝——生成侧把语义为假的断言原样交给 Lean、判定侧不虚报 proved, +正是"绝不 lean 无法证明的 marqdo 可以证明"的落实;修复方向是改对断言,不是放宽判定。 + +生成文件头部带 `set_option linter.unusedVariables false` / +`set_option linter.unusedSimpArgs false`:机器生成产物不应被人写代码的 linter +刷屏( simp 参数是固定词汇表,个别文件用不到属正常)。 diff --git a/doc/verification/skill-proof/README.md b/doc/verification/skill-proof/README.md index 7be4e64..c015d93 100644 --- a/doc/verification/skill-proof/README.md +++ b/doc/verification/skill-proof/README.md @@ -43,7 +43,8 @@ lake env lean .lean # exit 0 且无 error ⇒ 该文件全 ## 实测结论(2026-11-16,Lean 4.35.0 / lake 5.0.0) -- 业务样例(审批.mq.md,3 条定理):`lake env lean` 0 error,全部证明。 -- 10 个测试夹具生成文件:9 个 0 error;`p2_len`(计数器)的 - `长度(轨迹) = 2` 为**假命题**(该函数不产生操作,轨迹恒空)——Lean 如实 - 未证,符合"绝不 lean 无法证明的 marqdo 可以证明"。 +- 业务样例(审批.mq.md,3 条定理):`lake env lean` 0 error 0 warning,全部证明。 +- 6 个带定理的测试夹具 + 业务样例:**7/7 全部 proved**(驱动脚本三态分类: + proved / unproved / lean-error)。 +- 首轮曾发现 `p2_len` 夹具断言写错(纯函数却断言两步轨迹)被 Lean 如实拒绝—— + 修复方向是改对断言,不是放宽判定:"绝不 lean 无法证明的 marqdo 可以证明"。 diff --git a/doc/verification/skill-proof/run-lean-check.sh b/doc/verification/skill-proof/run-lean-check.sh new file mode 100755 index 0000000..19562d2 --- /dev/null +++ b/doc/verification/skill-proof/run-lean-check.sh @@ -0,0 +1,49 @@ +#!/usr/bin/env bash +# run-lean-check.sh — 用真实 Lean 4.35.0 校验 `marqdo prove --emit` 的产物。 +# 用法: run-lean-check.sh ... +# 每个文件: prove 生成 → lake env lean 编译 → 按退出状态分类。 +set -u +ROOT="$(cd "$(dirname "$0")/../.." && pwd)" # /workspace/marqdo +LEAN_BIN="$ROOT/target/toolchain/lean-4.35.0-linux/bin" +PROOF_DIR="$ROOT/target/proof/skill-proof" +MARQDO="$ROOT/target/node/debug/marqdo" +export PATH="$LEAN_BIN:$PATH" + +[ -x "$MARQDO" ] || { echo "marqdo binary missing: $MARQDO" >&2; exit 2; } +[ -f "$PROOF_DIR/.lake/build/lib/lean/MarqdoProof.olean" ] || \ + { (cd "$PROOF_DIR" && lake build) >&2 || exit 2; } + +declare -A RESULT +work="$(mktemp -d)" +trap 'rm -rf "$work"' EXIT +for src in "$@"; do + name="$(basename "$src" .mq.md)" + outdir="$work/$name" + mkdir -p "$outdir" + if ! "$MARQDO" prove "$src" --emit "$outdir" >/dev/null 2>&1; then + RESULT[$src]="emit-failed" + continue + fi + lean_file="$(find "$outdir" -name '*.lean' | head -1)" + [ -n "$lean_file" ] || { RESULT[$src]="no-lean-emitted"; continue; } + cp "$lean_file" "$PROOF_DIR/Generated/_check.lean" + log="$work/$name.leanlog" + (cd "$PROOF_DIR" && lake env lean Generated/_check.lean) >"$log" 2>&1 + rc=$? + rm -f "$PROOF_DIR/Generated/_check.lean" + if [ $rc -eq 0 ] && ! grep -q "error" "$log"; then + RESULT[$src]="proved" + elif grep -q "unsolved goals" "$log"; then + RESULT[$src]="unproved" + else + RESULT[$src]="lean-error" + fi +done + +fail=0 +for src in "$@"; do + status="${RESULT[$src]}" + printf '%-12s %s\n' "$status" "$src" + [ "$status" = "proved" ] || fail=1 +done +exit $fail diff --git a/src/prove/emit.rs b/src/prove/emit.rs index cd23035..94b5b5e 100644 --- a/src/prove/emit.rs +++ b/src/prove/emit.rs @@ -22,6 +22,10 @@ pub fn emit_module(plan: &Plan, module_stem: &str) -> (String, String) { let mut out = String::new(); let _ = writeln!(out, "import {LIBRARY_IMPORT}"); let _ = writeln!(out, "open MarqdoProof"); + // 机器生成的证明产物:关闭面向人写代码的 linter(未用变量/simp 参数是 + // 生成器的固定词汇表带来的正常噪声,不是问题)。 + let _ = writeln!(out, "set_option linter.unusedVariables false"); + let _ = writeln!(out, "set_option linter.unusedSimpArgs false"); let _ = writeln!(out); let _ = writeln!(out, "-- 由 `marqdo prove emit` 生成:确定性输出,请勿手改。"); let _ = writeln!(out, "-- 源:{}(sha256 {})", plan.module, plan.source_sha256); diff --git a/tests/prove.rs b/tests/prove.rs index 6174af7..f59470d 100644 --- a/tests/prove.rs +++ b/tests/prove.rs @@ -773,12 +773,12 @@ import: # 轨迹计数 -## 两步 +## 纯计算 + `金额` | 定理 | 结论 | 策略 | |---|---|---| -| 恰两步 | 长度(轨迹) = 2 | auto | +| 空轨迹 | 长度(轨迹) = 0 | auto | **a = [一步](result.ok(value=1))** **b = [两步](result.ok(value=2))** @@ -790,7 +790,7 @@ import: assert_eq!(r.code, 0, "stderr={}", r.stderr); let lean = read(&s.dir, "out/Generated/count_mq.lean"); assert!( - lean.contains("((«轨迹».length : Nat) = 2)"), + lean.contains("((«轨迹».length : Nat) = 0)"), "Nat 比较:{lean}" ); assert!(!lean.contains("mqEq («轨迹».length"), "不许出现类型不匹配:{lean}");