@kernel-sig/console 0.1.0

package/backend/app/integrations/atomgit/models.py

@@ -0,0 +1,311 @@
+"""AtomGit API 响应的数据模型。
+
+字段严格对应 **实测响应**（见 `tests/fixtures/atomgit/`），而非官方文档 ——
+文档描述的是 GitCode 企业版超集，与 atomgit.com 实例存在出入。
+
+实测发现的结构差异（与 GitHub 风格 API 的直觉不同）：
+- ``pulls/{n}/files`` 的 ``patch`` 是**对象**（含 ``diff``/``old_path``/``too_large``），
+  不是字符串
+- commit 的 message 嵌套在 ``commit.{author,committer,message}``
+- ``mergeable_state`` 是含冲突检测明细的**对象**，不是状态字符串
+- 时间字段在"未发生"时返回空字符串而非 null
+- Issue 的 ``number`` 是字符串，PR 的是整数
+- PR 对象不含评审关系，评审状态由 ``labels`` 承载
+"""
+
+from datetime import datetime
+from typing import Annotated, Any, ClassVar
+
+from pydantic import BaseModel, BeforeValidator, ConfigDict, Field
+
+
+def _blank_to_none(value: Any) -> Any:
+    """AtomGit 用空字符串表示"时间未设置"，需归一化为 None。"""
+    if isinstance(value, str) and not value.strip():
+        return None
+    return value
+
+
+def _str_to_int(value: Any) -> Any:
+    """Issue 编号以字符串返回，统一转为整数以与 PR 共用编号空间。"""
+    if isinstance(value, str) and value.isdigit():
+        return int(value)
+    return value
+
+
+def _wrap_string_as_diff(value: Any) -> Any:
+    """``patch`` 可能是对象，也可能是纯 diff 字符串，统一成对象形态。"""
+    if isinstance(value, str):
+        return {"diff": value}
+    return value
+
+
+OptionalDateTime = Annotated[datetime | None, BeforeValidator(_blank_to_none)]
+NumericId = Annotated[int, BeforeValidator(_str_to_int)]
+
+
+class AtomGitBase(BaseModel):
+    model_config = ConfigDict(extra="ignore", populate_by_name=True)
+
+
+class AtomGitUser(AtomGitBase):
+    login: str
+    id: Any = None
+    name: str | None = None
+    avatar_url: str | None = None
+
+
+class AtomGitLabel(AtomGitBase):
+    id: Any = None
+    name: str
+    color: str | None = None
+    description: str | None = None
+
+
+class AtomGitBranchRef(AtomGitBase):
+    ref: str | None = None
+    sha: str | None = None
+    label: str | None = None
+
+
+class AtomGitMergeCheck(AtomGitBase):
+    """``mergeable_state`` 实测为合并前置检查的明细对象。
+
+    比单一的 mergeable 布尔值信息量大得多 —— 可直接用于关注项规则，
+    例如 ``conflict_passed == False`` 表示存在合并冲突。
+    """
+
+    model_config = ConfigDict(extra="allow")
+
+    merge_request_id: int | None = None
+    state: bool | None = None
+    status_without_user_auth: bool | None = None
+    conflict_passed: bool | None = None
+    branch_missing_passed: bool | None = None
+    non_ff_passed: bool | None = None
+    mr_state_passed: bool | None = None
+    merged_by_user_passed: bool | None = None
+
+
+class AtomGitPullRequest(AtomGitBase):
+    """PR 列表与详情共用。详情接口的字段是列表接口的子集扩充。"""
+
+    number: int
+    id: Any = None
+    iid: int | None = None
+    project_id: Any = None
+    title: str
+    body: str | None = None
+    state: str
+    draft: bool = False
+    locked: bool = False
+
+    user: AtomGitUser | None = None
+    labels: list[AtomGitLabel] = Field(default_factory=list)
+    assignees: list[AtomGitUser] = Field(default_factory=list)
+    testers: list[AtomGitUser] = Field(default_factory=list)
+    approval_reviewers: list[AtomGitUser] = Field(default_factory=list)
+
+    target_branch: str | None = None
+    source_branch: str | None = None
+    source_project_id: Any = None
+
+    added_lines: int = 0
+    removed_lines: int = 0
+    # AtomGit 的 notes 字段直接给出评论数，可用于变更检测
+    notes: int = 0
+
+    mergeable: bool | None = None
+    mergeable_state: AtomGitMergeCheck | None = None
+    can_merge_check: bool | None = None
+    close_related_issue: Any = None
+
+    created_at: OptionalDateTime = None
+    updated_at: OptionalDateTime = None
+    merged_at: OptionalDateTime = None
+    closed_at: OptionalDateTime = None
+
+    html_url: str | None = None
+    web_url: str | None = None
+
+    head: AtomGitBranchRef | None = None
+    base: AtomGitBranchRef | None = None
+
+    @property
+    def label_names(self) -> list[str]:
+        return [label.name for label in self.labels]
+
+    @property
+    def author_login(self) -> str | None:
+        return self.user.login if self.user else None
+
+    @property
+    def permalink(self) -> str | None:
+        """规范 PR 链接。html_url 实测指向 /merge_requests/，故优先 web_url。"""
+        return self.web_url or self.html_url
+
+    @property
+    def has_merge_conflict(self) -> bool:
+        """是否冲突。优先用明细对象，回退到 mergeable 布尔值。"""
+        if self.mergeable_state is not None and self.mergeable_state.conflict_passed is not None:
+            return not self.mergeable_state.conflict_passed
+        return self.mergeable is False
+
+
+class AtomGitIssue(AtomGitBase):
+    number: NumericId
+    id: Any = None
+    title: str
+    body: str | None = None
+    state: str
+    comments: int = 0
+
+    # AtomGit 原生结构化字段，比用标签猜测更可靠
+    issue_type: str | None = None
+    issue_state: str | None = None
+    priority: int | None = None
+    issue_priority_detail: dict[str, Any] | None = None
+    issue_type_detail: dict[str, Any] | None = None
+    issue_state_detail: dict[str, Any] | None = None
+
+    labels: list[AtomGitLabel] = Field(default_factory=list)
+    user: AtomGitUser | None = None
+    assignee: AtomGitUser | None = None
+    assignees: list[AtomGitUser] = Field(default_factory=list)
+
+    created_at: OptionalDateTime = None
+    updated_at: OptionalDateTime = None
+    finished_at: OptionalDateTime = None
+    html_url: str | None = None
+
+    @property
+    def label_names(self) -> list[str]:
+        return [label.name for label in self.labels]
+
+    @property
+    def priority_label(self) -> str | None:
+        if self.issue_priority_detail:
+            title = self.issue_priority_detail.get("title")
+            if isinstance(title, str):
+                return title
+        return None
+
+
+class AtomGitComment(AtomGitBase):
+    """PR 与 Issue 的评论。
+
+    实测字段含 ``comment_type`` 与 ``discussion_id``；
+    后者可用于把评论归入同一条讨论线程。
+    """
+
+    id: Any
+    body: str = ""
+    comment_type: str | None = None
+    discussion_id: str | None = None
+    user: AtomGitUser | None = None
+    created_at: OptionalDateTime = None
+    updated_at: OptionalDateTime = None
+
+    @property
+    def author_login(self) -> str | None:
+        return self.user.login if self.user else None
+
+
+class AtomGitFilePatch(AtomGitBase):
+    """``pulls/{n}/files`` 中 ``patch`` 字段的实际结构。"""
+
+    diff: str | None = None
+    old_path: str | None = None
+    new_path: str | None = None
+    a_mode: str | None = None
+    b_mode: str | None = None
+    new_file: bool = False
+    deleted_file: bool = False
+    renamed_file: bool = False
+    too_large: bool = False
+    added_lines: int = 0
+    removed_lines: int = 0
+
+
+class AtomGitFile(AtomGitBase):
+    filename: str
+    status: str | None = None
+    additions: int = 0
+    deletions: int = 0
+    changes: int = 0
+    sha: str | None = None
+    blob_url: str | None = None
+    raw_url: str | None = None
+    patch: Annotated[AtomGitFilePatch | None, BeforeValidator(_wrap_string_as_diff)] = None
+
+    # 二进制文件的判定依据：Git 在 diff 中给出的是这个标记而非文本差异
+    BINARY_MARKERS: ClassVar[tuple[str, ...]] = (
+        "Binary files differ",
+        "GIT binary patch",
+    )
+
+    @property
+    def diff_text(self) -> str | None:
+        return self.patch.diff if self.patch else None
+
+    @property
+    def is_binary(self) -> bool:
+        text = self.diff_text
+        if text is None:
+            return False
+        return any(marker in text for marker in self.BINARY_MARKERS)
+
+
+class AtomGitCommitAuthor(AtomGitBase):
+    login: str | None = None
+    name: str | None = None
+    email: str | None = None
+    date: OptionalDateTime = None
+
+
+class AtomGitCommitInner(AtomGitBase):
+    """commit 的元信息嵌套层，与 GitHub API 一致。"""
+
+    message: str = ""
+    author: AtomGitCommitAuthor | None = None
+    committer: AtomGitCommitAuthor | None = None
+
+
+class AtomGitCommit(AtomGitBase):
+    sha: str
+    commit: AtomGitCommitInner = Field(default_factory=AtomGitCommitInner)
+    author: AtomGitCommitAuthor | None = None
+    committer: AtomGitCommitAuthor | None = None
+    parents: Any = None
+    html_url: str | None = None
+
+    @property
+    def short_sha(self) -> str:
+        return self.sha[:8]
+
+    @property
+    def message(self) -> str:
+        return self.commit.message
+
+    @property
+    def subject(self) -> str:
+        return self.message.split("\n", 1)[0].strip()
+
+    @property
+    def author_email(self) -> str | None:
+        source = self.commit.author or self.author
+        return source.email if source else None
+
+
+class AtomGitRepository(AtomGitBase):
+    id: Any = None
+    name: str
+    full_name: str
+    human_name: str | None = None
+    description: str | None = None
+    default_branch: str | None = None
+    html_url: str | None = None
+    web_url: str | None = None
+    open_issues_count: int | None = None
+    stargazers_count: int | None = None
+    forks_count: int | None = None

package/backend/app/integrations/llm/prompts.py

@@ -0,0 +1,222 @@
+"""AI 任务的 Prompt 模板。
+
+关键设计：Prompt 中内嵌 **openEuler 的具体规则**（KABI 约束、backport 要求、
+补丁格式），而不是让它"通用地评审一段代码"。通用提示只会产出
+"建议增加错误处理""注意边界条件"这类无法据以行动的话。
+
+模板用 ``{name}`` 占位，由调用方填充。
+"""
+
+from __future__ import annotations
+
+# ---------------------------------------------------------------------------
+# 领域规则：所有 Prompt 共享的背景
+# ---------------------------------------------------------------------------
+
+DOMAIN_RULES = """\
+你正在协助 openEuler Kernel SIG 的维护者处理内核补丁。以下规则是硬约束，
+判断时必须遵守：
+
+【KABI 兼容性】openEuler 对已发布内核承诺 KABI 稳定。以下变更属破坏性，
+除非有明确理由否则不应合入：
+  - 删除 EXPORT_SYMBOL / EXPORT_SYMBOL_GPL
+  - 修改已导出函数的签名或返回类型
+  - 结构体新增字段但不是加在末尾，或删除/重排已有字段
+  - 修改位域布局、联合体成员、枚举值
+  - 变更 UAPI 头文件中的接口定义
+
+【补丁格式】每个提交必须包含：
+  - Signed-off-by 链（作者与经手人）
+  - 标题子系统前缀，形如 "net: "、"mm/hugetlb: "，长度不超过 72 字符
+  - backport 补丁需注明来源：from <version> commit <sha> 或 CVE 编号
+  - 正文空行分隔，说明改动的必要性而不只是描述做了什么
+
+【禁止提交】二进制文件（.ko/.bin/.o/.pdf 等）。固件应提交至 linux-firmware。
+
+【Backport 要求】回合上游补丁时必须逐行对照原补丁。若因内核版本差异需要
+调整，必须在提交信息中说明差异原因。openEuler 的定制代码不得被覆盖。
+
+【评审原则】报告问题前先确认该问题在真实调用路径上可触发。
+不要报告纯风格偏好。指出具体文件与行号。"""
+
+
+# ---------------------------------------------------------------------------
+# 任务模板
+# ---------------------------------------------------------------------------
+
+SUMMARIZE_SYSTEM = f"""{DOMAIN_RULES}
+
+你的任务：为这个补丁写一段中文摘要，帮助维护者在 30 秒内判断是否需要深入阅读。
+
+输出 JSON：
+{{
+  "summary": "3-5 句中文摘要，说明这个补丁做了什么、为什么需要",
+  "changes": ["主要变更点，每条一句话"],
+  "risk_areas": ["需要重点看的地方，如并发、错误路径、兼容性；没有则给空数组"],
+  "confidence": 0.0 到 1.0 的数字
+}}
+
+只输出 JSON。不确定的地方如实说明，不要编造。"""
+
+SUMMARIZE_USER = """\
+PR 标题：{title}
+目标分支：{target_branch}
+作者：{author}
+变更统计：+{added_lines}/-{removed_lines} 行
+
+提交信息：
+{commit_messages}
+
+变更文件：
+{file_list}
+
+提交正文：
+{body}"""
+
+
+RISK_REVIEW_SYSTEM = f"""{DOMAIN_RULES}
+
+你的任务：审查这个内核补丁，找出**真实存在**的缺陷。
+
+必须做到：
+- 只报告你能给出具体文件与代码位置的缺陷
+- 不报告纯风格问题、命名偏好、注释缺失
+- 若无法判断某处是否为缺陷，归入 questions 而非 risks
+- 宁可少报也不要误报：维护者被误报消耗的注意力是真实成本
+
+输出 JSON：
+{{
+  "risks": [
+    {{
+      "severity": "blocker|critical|warning|info",
+      "category": "kabi|memory|concurrency|error_handling|security|correctness|backport",
+      "file": "文件路径",
+      "location": "函数名或行号范围",
+      "issue": "问题是什么",
+      "rationale": "为什么这是问题，什么情况下会触发",
+      "suggestion": "建议怎么改"
+    }}
+  ],
+  "questions": ["需要作者澄清的问题"],
+  "kabi_impact": {{"breaking": true 或 false, "detail": "说明"}},
+  "patch_format_issues": ["补丁格式问题，如缺少 Signed-off-by"],
+  "summary": "整体评价，2-3 句",
+  "confidence": 0.0 到 1.0
+}}
+
+只输出 JSON。"""
+
+RISK_REVIEW_USER = """\
+PR 标题：{title}
+目标分支：{target_branch}
+
+提交信息：
+{commit_messages}
+
+补丁内容：
+{diff}"""
+
+
+BACKPORT_VERIFY_SYSTEM = f"""{DOMAIN_RULES}
+
+你的任务：核对一个 backport 补丁与它的上游原始补丁是否等价。
+
+逐行对照，找出所有差异。对每处差异判断属于哪一类：
+  - "harmless"：上下文偏移导致的，语义无变化
+  - "adaptation"：因内核版本差异做的必要调整
+  - "divergence"：语义上有实质差异，需要作者说明
+  - "missing"：上游补丁中的改动在 backport 里缺失
+
+输出 JSON：
+{{
+  "equivalent": true 或 false,
+  "divergences": [
+    {{
+      "kind": "harmless|adaptation|divergence|missing",
+      "location": "文件与大致位置",
+      "upstream": "上游补丁此处的内容",
+      "backport": "backport 此处的内容",
+      "assessment": "你的判断"
+    }}
+  ],
+  "openEuler_code_affected": "是否触及 openEuler 定制代码",
+  "summary": "2-3 句总结",
+  "confidence": 0.0 到 1.0
+}}
+
+只输出 JSON。若未提供上游补丁，在 summary 中说明无法核对并令 equivalent 为 null。"""
+
+BACKPORT_VERIFY_USER = """\
+Backport 补丁标题：{title}
+目标分支：{target_branch}
+
+上游原始补丁：
+{upstream_patch}
+
+openEuler backport 补丁：
+{backport_patch}"""
+
+
+CLASSIFY_SYSTEM = """\
+你在为 openEuler 内核仓库的 PR 做分类。规则引擎已尝试但未能给出结论，
+需要你根据标题、描述与提交信息判断。
+
+类别取值（必须选其一）：
+  cve           — 修复安全漏洞（出现 CVE 编号）
+  backport      — 回合上游补丁，含 LTS / stable 分支的补丁
+  driver_new    — 新增驱动或控制器支持
+  driver_update — 驱动版本升级
+  soc_support   — 处理器 / SoC 厂商支持，如飞腾、兆芯、龙芯、海光、鲲鹏
+  out_of_tree   — openEuler 自研特性，不在上游主线中，
+                  如 urma、ub、etmem、enfs、syscare
+  bugfix        — 修复功能缺陷
+  feature       — 新增功能
+  perf          — 性能优化
+  refactor      — 重构，不改变外部行为
+  docs          — 文档
+  cleanup       — 清理，如删除死代码、修正拼写
+
+判断要点：
+- 提交信息里可能有作者自填的 `category:` 与 `<X> inclusion` 标注，
+  它们比标题可靠，请优先采信
+- 标题里的 `arm64:`、`x86:` 是子系统前缀，不代表"处理器支持"；
+  只有明确指向某家 SoC 厂商才算
+- 拿不准时宁可给 unknown，也不要在两个类别之间硬猜
+
+输出 JSON：
+{{
+  "kind": "上面十二选一",
+  "subsystem": "所属子系统，如 net / mm / fs / arm64 / drm",
+  "confidence": 0.0 到 1.0,
+  "reason": "判断依据，一句话"
+}}
+
+只输出 JSON。"""
+
+CLASSIFY_USER = """\
+标题：{title}
+
+描述：
+{body}
+
+变更文件：
+{file_list}
+
+提交信息：
+{commit_messages}"""
+
+
+TEMPLATES: dict[str, tuple[str, str]] = {
+    "summarize": (SUMMARIZE_SYSTEM, SUMMARIZE_USER),
+    "risk_review": (RISK_REVIEW_SYSTEM, RISK_REVIEW_USER),
+    "backport_verify": (BACKPORT_VERIFY_SYSTEM, BACKPORT_VERIFY_USER),
+    "classify": (CLASSIFY_SYSTEM, CLASSIFY_USER),
+}
+
+
+def get_template(task: str) -> tuple[str, str]:
+    """取内置模板。任务无效时明确报错，而非回落到某个默认模板。"""
+    try:
+        return TEMPLATES[task]
+    except KeyError as exc:
+        raise KeyError(f"未知任务 {task!r}，可用：{', '.join(sorted(TEMPLATES))}") from exc