From bd2d1a67de98cacfbe172028ecfb0893abd89895 Mon Sep 17 00:00:00 2001 From: Cursor Agent Date: Mon, 2 Nov 2026 16:03:36 +0000 Subject: [PATCH] fix(git): record untracked new files in diff snapshots git diff HEAD omits files that were never added, so a notes.md created in the worktree left both the failure episode and the success event with an empty scene. Append those files as new-file hunks from the shared snapshot helper, and keep dependency, runtime, and binary noise out of that supplement. Co-authored-by: Steper Lin --- CHANGELOG.md | 3 + DEBUG.md | 17 ++++-- src/evolver/gep/git_ops.py | 114 ++++++++++++++++++++++++++++++++++++- tests/gep/test_solidify.py | 46 +++++++++++++++ tests/test_git_ops.py | 74 ++++++++++++++++++++++++ 5 files changed, 246 insertions(+), 8 deletions(-) diff --git a/CHANGELOG.md b/CHANGELOG.md index acf87ac..6b8c5f2 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -8,6 +8,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0 ## Unreleased +### Fixed +- **失败现场 diff 包含未跟踪新文件（DEBUG #57）**：`capture_diff_snapshot` 在 `git diff HEAD` 之后追加未跟踪新文件的 unified diff。失败侧 `_failure_diff` 与成功侧共用这一 helper。依赖树、缓存、运行态目录、受保护路径和含 NUL 的二进制不进入这份补充。不改固化门，不写 Parent baseline，不开 val 密封。 + ### Docs - **README 对齐软默认与 MCP 口味（英/中）**：soft-accept 为 MCP 默认路径（经验写入 `~/.evolver/workspaces//`，不自动 git commit，不写 Parent baseline；`no_baseline` / `pack_absent` / `unmeasured` / `drop` / `flat` / `regression` 为警告）。硬阻断仍为密钥、`~/.evolver/`、`val_seal`。补开箱 `setup` / `doctor`、soft T0、前 5 tick 核心循环、`pending_solidify`、安装后重启 MCP，以及本 fork 的 clone URL（`linkerlin/evolver.py`）。顺手修正仍写「缺基线即停 / 过门即 commit / drop 仍硬拒绝」的 quickstart、AGENTS.md 与教程句子。 diff --git a/DEBUG.md b/DEBUG.md index dab8d48..52abec3 100644 --- a/DEBUG.md +++ b/DEBUG.md @@ -1082,7 +1082,7 @@ 后再谈归因，截断输出里的百分比进度不是证据；**`lastfailed` 不是线索库， 是考古层**——`--lf` 前先 `--collect-only` 验 id 存活，零命中会静默变成全量。 -### 57. 失败现场 diff 看不见未跟踪文件（MCP 验证脚本实证，未修） +### 57. 失败现场 diff 看不见未跟踪文件（MCP 验证脚本实证，已修） - **症状**：MCP 闭环验证脚本里，宿主新建 `notes.md`（未跟踪）后走 `hypothesis_missing` 拒绝，episode 的 diff 为空，断言失败；把同一改动 @@ -1090,10 +1090,17 @@ - **根因**：`_failure_diff` → `capture_diff_snapshot` 即 `git diff HEAD`， 天然不含未跟踪文件。成功侧走同一 helper、同盲区——行为一致，没有偏袒 任一臂，但两边都可能丢掉"新建文件"类候选的关键现场。 -- **修复**：未修。改快照口径牵连成功侧与既有 diff 断言，需单独一轮（快照 - 未跟踪内容 + 钉住两臂一致 + 全量回归）。 -- **经验**：**验证脚本模拟宿主改动必须落在已跟踪文件上**；凡断言 diff - 内容，先确认文件跟踪状态——`git diff HEAD` 类快照默认看不见新文件。 +- **修复**：`capture_diff_snapshot` 仍先取 `git diff HEAD`（已跟踪文件的 + 暂存与未暂存改动，含已 `git add` 的新文件），再把从未入库的未跟踪新文件 + 收成 `diff --git` 新文件块接在后面。失败侧 `_failure_diff` 与成功侧 + 都走这一 helper，两臂同一口径。补充块跳过爆炸半径不计的路径（`.venv`、 + `node_modules`、`__pycache__`、`.git`、`.env` 等受保护文件）、运行态目录 + （`memory/`、`.evolver/` 等，与 `_is_runtime_state` 同口径）以及文件头 + 含 NUL 的二进制；读入量不超过快照剩余预算。不改固化门，不写 Parent + baseline，不开 val 密封。 +- **经验**：**凡断言 diff 内容，先确认文件是不是未跟踪**——修之前 + `git diff HEAD` 看不见新文件。快照要看见新文件，就改共用 helper， + 不要给失败侧单开一条路。运行态与依赖树不能跟候选正文抢那 2000 字。 ### 58. 外部 MCP 评审 P1 四项（评估会话，全修） diff --git a/src/evolver/gep/git_ops.py b/src/evolver/gep/git_ops.py index 0e77016..2c5e94b 100644 --- a/src/evolver/gep/git_ops.py +++ b/src/evolver/gep/git_ops.py @@ -11,7 +11,7 @@ import time from collections.abc import Sequence from pathlib import Path -from typing import Any +from typing import Any, Final from evolver.solo.git_guard import guard_git_subcommand @@ -95,12 +95,120 @@ def git_list_untracked_files(cwd: Path | str | None = None) -> list[str]: return [line for line in raw.splitlines() if line.strip()] +# Path components that are dependency or cache trees, including nested ones. +# Top-level prefixes are also dropped by is_constraint_counted_path. +_SNAPSHOT_NOISE_PARTS: Final[frozenset[str]] = frozenset( + { + ".git", + "node_modules", + ".venv", + "venv", + "__pycache__", + ".pytest_cache", + } +) +# Top-level runtime heads. Same set as solidify._is_runtime_state: the +# engine's own write domain, not the mutation scene. Only the untracked +# supplement is filtered; `git diff HEAD` is left unchanged. +_SNAPSHOT_RUNTIME_HEADS: Final[frozenset[str]] = frozenset( + { + ".evolver", + ".evolver_settings", + ".evomap", + ".pytest_cache", + "logs", + "memory", + } +) +_BINARY_SNIFF_BYTES: Final = 8192 + + +def _untracked_snapshot_candidate(rel: str) -> bool: + """True when an untracked path belongs in the mutation diff scene.""" + norm = normalize_rel_path(rel) + if not norm or not is_constraint_counted_path(norm): + return False + if norm.startswith("evolver/.config/") or norm.endswith(".pyc"): + return False + parts = norm.split("/") + if parts[0] in _SNAPSHOT_RUNTIME_HEADS: + return False + return all(part not in _SNAPSHOT_NOISE_PARTS for part in parts) + + +def _format_new_file_diff(rel: str, text: str) -> str: + """Unified diff for a file that exists only in the worktree.""" + path = normalize_rel_path(rel) + header = f"diff --git a/{path} b/{path}\nnew file mode 100644\n--- /dev/null\n+++ b/{path}\n" + if text == "": + return header + lines = text.splitlines() + body = "\n".join(f"+{line}" for line in lines) + if text.endswith("\n"): + body += "\n" + else: + body += "\n\\ No newline at end of file\n" + return f"{header}@@ -0,0 +1,{len(lines)} @@\n{body}" + + +def _read_text_prefix(path: Path, limit: int) -> str | None: + """UTF-8 prefix of a regular file, or None when it is missing or binary.""" + if limit <= 0 or not path.is_file(): + return None + try: + with path.open("rb") as handle: + sniff = handle.read(_BINARY_SNIFF_BYTES) + if b"\x00" in sniff: + return None + blob = sniff[:limit] if len(sniff) >= limit else sniff + handle.read(limit - len(sniff)) + except OSError: + return None + return blob.decode("utf-8", errors="replace") + + +def _untracked_diff_text(cwd: Path, *, budget: int) -> str: + """New-file diffs for untracked candidates, capped at ``budget`` chars.""" + if budget <= 0: + return "" + chunks: list[str] = [] + used = 0 + for rel in git_list_untracked_files(cwd): + if used >= budget or not _untracked_snapshot_candidate(rel): + continue + remaining = budget - used + text = _read_text_prefix(cwd / rel, remaining) + if text is None: + continue + chunk = _format_new_file_diff(rel, text) + if len(chunk) > remaining: + chunk = chunk[:remaining] + chunks.append(chunk) + used += len(chunk) + return "".join(chunks) + + def capture_diff_snapshot( cwd: Path | str | None = None, max_chars: int = 80_000, ) -> str: - """Capture current diff vs HEAD, truncated to max_chars.""" - raw = try_run_cmd(["diff", "HEAD"], cwd=cwd) + """Capture the worktree vs HEAD, including untracked new files. + + ``git diff HEAD`` already contains staged and unstaged edits to tracked + paths. A file that was never ``git add``ed is absent from that diff, so + its content is appended as a ``diff --git`` new-file hunk. Failure and + success snapshots share this helper. Noise trees, runtime-state dirs, + protected paths, and binaries are left out of the untracked supplement. + The result is truncated to ``max_chars``. + """ + root = Path(cwd) if cwd is not None else Path.cwd() + raw = try_run_cmd(["diff", "HEAD"], cwd=root) + if len(raw) > max_chars: + return raw[:max_chars] + "\n...[truncated]\n" + extra = _untracked_diff_text(root, budget=max_chars - len(raw)) + if extra: + if raw and not raw.endswith("\n"): + raw += "\n" + raw += extra if len(raw) > max_chars: return raw[:max_chars] + "\n...[truncated]\n" return raw diff --git a/tests/gep/test_solidify.py b/tests/gep/test_solidify.py index 82b6931..5fcaa68 100644 --- a/tests/gep/test_solidify.py +++ b/tests/gep/test_solidify.py @@ -231,6 +231,52 @@ def test_run_validations_timeout(git_ws: Path, monkeypatch: pytest.MonkeyPatch) assert res["results"][0]["stderr"] +def test_hypothesis_rejection_records_untracked_new_file(git_ws: Path) -> None: + """DEBUG #57: a gate reject must keep an untracked new file in the scene. + + The file is rolled back afterwards; the episode is the only copy left. + """ + (git_ws / "notes.md").write_text("untracked-notes-marker\n", encoding="utf-8") + write_state_for_solidify(_last_run()) + result = solidify() + assert result["ok"] is False + assert result["error"] == "hypothesis_missing" + from evolver.gep.episode_record import load_episode, record_round + + stored = record_round(event_id=str(result["event_id"])) + body = load_episode(stored["id"]) + assert body is not None + assert "diff --git a/notes.md b/notes.md" in body["diff"] + assert "untracked-notes-marker" in body["diff"] + assert not (git_ws / "notes.md").exists() + + +def test_success_diff_snapshot_includes_untracked_new_file( + git_ws: Path, + armed_pack: dict[str, Any], + declared_hypothesis: dict[str, Any], +) -> None: + """Success uses the same snapshot helper, so a new file is recorded there too.""" + _ = armed_pack, declared_hypothesis + set_flag("enable_fitness_cascade", False, persist=False) + (git_ws / "notes.md").write_text("untracked-notes-marker\n", encoding="utf-8") + write_state_for_solidify( + _last_run( + mutation={ + "id": "mut_untracked_ok", + "validation": [[sys.executable, "-c", "print(1)"]], + } + ) + ) + result = solidify() + assert result["ok"] is True + evt = json.loads( + (get_gep_assets_dir() / "events.jsonl").read_text(encoding="utf-8").strip().splitlines()[-1] + ) + assert "untracked-notes-marker" in str(evt.get("diff_snapshot") or "") + assert "diff --git a/notes.md" in str(evt.get("diff_snapshot") or "") + + def test_hypothesis_rejection_leaves_a_recordable_scene(git_ws: Path) -> None: """A settled Reject is an engine scene: gene, diff, and the gate verdict.""" (git_ws / "README.md").write_text("changed by the candidate\n", encoding="utf-8") diff --git a/tests/test_git_ops.py b/tests/test_git_ops.py index b3e2380..09e1015 100644 --- a/tests/test_git_ops.py +++ b/tests/test_git_ops.py @@ -89,6 +89,80 @@ def test_is_git_repo_true_and_false(tmp_path: Path) -> None: assert is_git_repo(tmp_path) is True +def _init_repo(path: Path) -> None: + subprocess.run(["git", "init"], cwd=path, check=True, capture_output=True) + subprocess.run( + ["git", "config", "user.email", "t@t.com"], cwd=path, check=True, capture_output=True + ) + subprocess.run(["git", "config", "user.name", "T"], cwd=path, check=True, capture_output=True) + (path / "README.md").write_text("init\n", encoding="utf-8") + subprocess.run(["git", "add", "README.md"], cwd=path, check=True, capture_output=True) + subprocess.run( + ["git", "-c", "commit.gpgsign=false", "commit", "-m", "init"], + cwd=path, + check=True, + capture_output=True, + ) + + +def test_capture_diff_snapshot_includes_untracked_new_file(tmp_path: Path) -> None: + """DEBUG #57: a new file that was never git-added is part of the scene.""" + _init_repo(tmp_path) + (tmp_path / "notes.md").write_text("untracked-notes-marker\n", encoding="utf-8") + snap = capture_diff_snapshot(tmp_path) + assert "diff --git a/notes.md b/notes.md" in snap + assert "+untracked-notes-marker" in snap + + +def test_capture_diff_snapshot_includes_staged_new_file(tmp_path: Path) -> None: + """A new file already in the index stays visible via git diff HEAD.""" + _init_repo(tmp_path) + (tmp_path / "added.py").write_text("staged-new-marker\n", encoding="utf-8") + subprocess.run(["git", "add", "added.py"], cwd=tmp_path, check=True, capture_output=True) + snap = capture_diff_snapshot(tmp_path) + assert "staged-new-marker" in snap + + +def test_capture_diff_snapshot_keeps_tracked_edit_beside_untracked(tmp_path: Path) -> None: + _init_repo(tmp_path) + (tmp_path / "README.md").write_text("tracked-edit-marker\n", encoding="utf-8") + (tmp_path / "notes.md").write_text("untracked-notes-marker\n", encoding="utf-8") + snap = capture_diff_snapshot(tmp_path) + assert "tracked-edit-marker" in snap + assert "untracked-notes-marker" in snap + + +def test_capture_diff_snapshot_skips_noise_and_binaries(tmp_path: Path) -> None: + """Explosion-radius noise and binaries must not crowd the mutation scene.""" + _init_repo(tmp_path) + (tmp_path / "notes.md").write_text("untracked-notes-marker\n", encoding="utf-8") + noise = { + tmp_path / ".venv" / "lib" / "x.py": "noise-venv-marker\n", + tmp_path / "pkg" / "node_modules" / "leftpad.js": "noise-node-modules-marker\n", + tmp_path / "src" / "__pycache__" / "x.pyc": "noise-pycache-marker\n", + tmp_path / "memory" / "evolution_solidify_state.json": "noise-memory-marker\n", + tmp_path / ".evolver" / "gep" / "events.jsonl": "noise-evolver-marker\n", + tmp_path / ".env": "API_KEY=noise-env-marker\n", + } + for path, text in noise.items(): + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text(text, encoding="utf-8") + blob = tmp_path / "payload.bin" + blob.write_bytes(b"\x00noise-binary-marker") + snap = capture_diff_snapshot(tmp_path) + assert "untracked-notes-marker" in snap + for marker in ( + "noise-venv-marker", + "noise-node-modules-marker", + "noise-pycache-marker", + "noise-memory-marker", + "noise-evolver-marker", + "noise-env-marker", + "noise-binary-marker", + ): + assert marker not in snap + + def test_capture_diff_snapshot_truncates(tmp_path: Path) -> None: subprocess.run(["git", "init"], cwd=tmp_path, check=True, capture_output=True) subprocess.run(