diff --git a/crates/df-storage/src/crud.rs b/crates/df-storage/src/crud.rs deleted file mode 100644 index 5d920e5..0000000 --- a/crates/df-storage/src/crud.rs +++ /dev/null @@ -1,2212 +0,0 @@ -//! Repository 模式的 CRUD 操作 -//! -//! 为 7 张表各提供一个 Repo 结构体，统一实现 insert / get_by_id / list_all / query / update_field / delete。 - -use std::sync::Arc; - -use rusqlite::{params, Connection, OptionalExtension, Row}; -use tokio::sync::Mutex; - -use df_types::error::{Error, Result}; - -use crate::db::Database; -use crate::models::{ - AiConversationRecord, AiProviderRecord, AiToolExecutionRecord, BranchRecord, IdeaRecord, - KnowledgeEventRecord, KnowledgeRecord, NodeExecutionRecord, ProjectRecord, ReleaseRecord, - TaskRecord, WorkflowRecord, -}; - -/// rusqlite 错误 → df-types `Error::Storage` 统一包装(R-PD-10 DRY, -/// 替代散落的 `.map_err(storage_err)`)。 -fn storage_err(e: E) -> Error { - Error::Storage(e.to_string()) -} - -/// 规范化路径用于比较:canonicalize 解析绝对规范路径(失败降级), -/// 统一正斜杠 + 小写。与 `df_project::scan::normalize_path` **同算法镜像**(df-storage -/// 不依赖 df-project,故独立实现;改动须同步)。防 `C:\a\b` vs `C:/a/b/` 绕过。 -fn normalize_stored_path(p: &str) -> String { - match std::path::Path::new(p).canonicalize() { - Ok(abs) => abs.to_string_lossy().replace('\\', "/").to_lowercase(), - Err(_) => p - .trim_end_matches(['\\', '/']) - .replace('\\', "/") - .to_lowercase(), - } -} - -// ============================================================ -// 知识库 SELECT 列清单(防 COLS 漂移) -// ============================================================ - -/// `knowledges` 表对应 `KnowledgeRecord` 14 个字段的列名(顺序与结构体一致)。 -/// -/// 多处 `search`/`search_vector` 内联 COLS 串的 DRY 收口(CR-260615-03):集中一处定义, -/// 配合下方 `KNOWLEDGE_COL_COUNT` 断言,任一处加列漏改会被测试 `test_knowledge_cols_matches_record` -/// 立即捕获(`knowledge_from_row` 按 name 取列,SELECT 漏列会运行时 rusqlite 报错,故提前断言)。 -const KNOWLEDGE_COLS: &str = "id,kind,title,content,tags,status,confidence,reuse_count,verified,source_project,source_ref,reasoning,created_at,updated_at"; -/// `KnowledgeRecord` 字段数(与上面列清单的逗号分隔项数一致,被测试断言)。 -/// 仅测试期消费(列漂移断言);保留为非 `cfg(test)` 以便测试外的阅读者一眼看到字段数。 -#[cfg_attr(not(test), allow(dead_code))] -const KNOWLEDGE_COL_COUNT: usize = 14; -/// `search_vector` 用的列清单:KNOWLEDGE_COLS + embedding(余弦计算用,不入 KnowledgeRecord)。 -const KNOWLEDGE_COLS_WITH_EMBEDDING: &str = concat!( - "id,kind,title,content,tags,status,confidence,reuse_count,verified,", - "source_project,source_ref,reasoning,created_at,updated_at,embedding" -); - -// ============================================================ -// 辅助宏 — 消除 6 个 Repo 的重复样板 -// ============================================================ - -/// 一次性为 7 张表生成完整的 Repo 结构体及其 CRUD 实现。 -/// -/// 用法: `impl_repo!(RepoName, ModelType, "table_name", from_row_body, insert_body, update_body);` -/// -/// 约束：`$record` 须实现 `Clone`（`update_full` 内部 clone 后丢给 spawn_blocking）。 -macro_rules! impl_repo { - ( - $(#[$meta:meta])* - $name:ident, $record:ident, $table:literal, - from_row => |$row:ident| $from_body:expr, - insert => |$conn:ident, $rec:ident| $insert_body:expr, - update => |$u_conn:ident, $u_rec:ident| $update_body:expr - ) => { - $(#[$meta])* - pub struct $name { - conn: Arc>, - } - - impl $name { - pub fn new(db: &Database) -> Self { - Self { conn: db.conn() } - } - - pub async fn insert(&self, record: $record) -> Result { - let conn = self.conn.clone(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let id = record.id.clone(); - let $conn = &*guard; - let $rec = &record; - $insert_body.map_err(storage_err)?; - Ok(id) - }) - .await - .map_err(storage_err)? - } - - pub async fn get_by_id(&self, id: &str) -> Result> { - let conn = self.conn.clone(); - let id = id.to_owned(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare(&format!("SELECT * FROM {} WHERE id = ?1", $table)) - .map_err(storage_err)?; - let row = stmt - .query_row(params![id], |$row| $from_body) - .optional() - .map_err(storage_err)?; - Ok(row) - }) - .await - .map_err(storage_err)? - } - - pub async fn list_all(&self) -> Result> { - let conn = self.conn.clone(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare(&format!("SELECT * FROM {} ORDER BY created_at DESC", $table)) - .map_err(storage_err)?; - let rows = stmt - .query_map([], |$row| $from_body) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push( - r.map_err(storage_err)?, - ); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } - - pub async fn query(&self, field: &str, value: &str) -> Result> { - // 白名单校验，防止 SQL 注入 - validate_column_name(field, $table)?; - let conn = self.conn.clone(); - let sql = format!("SELECT * FROM {} WHERE {} = ?1 ORDER BY created_at DESC", $table, field); - let value = value.to_owned(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare(&sql) - .map_err(storage_err)?; - let rows = stmt - .query_map(params![value], |$row| $from_body) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push( - r.map_err(storage_err)?, - ); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } - - pub async fn update_field(&self, id: &str, field: &str, value: &str) -> Result { - validate_column_name(field, $table)?; - let conn = self.conn.clone(); - let sql = format!("UPDATE {} SET {} = ?1, updated_at = ?2 WHERE id = ?3", $table, field); - let id = id.to_owned(); - let value = value.to_owned(); - let now = now_millis_str(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let affected = guard - .execute(&sql, params![value, now, id]) - .map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } - - /// 整体更新记录（全可变字段，保留 id 与 created_at），单次原子写 - pub async fn update_full(&self, record: &$record) -> Result { - let conn = self.conn.clone(); - let rec = record.clone(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let $u_conn = &*guard; - let $u_rec = &rec; - let affected = - $update_body.map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } - - pub async fn delete(&self, id: &str) -> Result { - let conn = self.conn.clone(); - let id = id.to_owned(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let affected = guard - .execute(&format!("DELETE FROM {} WHERE id = ?1", $table), params![id]) - .map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } - } - }; -} - -// ============================================================ -// 通用键值设置 — app_settings 表(KV,不适合 impl_repo! 宏) -// ============================================================ - -/// 通用应用设置 KV Repo(表 `app_settings`) -/// -/// 前端 localStorage 迁移目标:存主题/语言/AI 偏好/连接配置等,`value` 为 JSON 字符串。 -/// KV 无固定 schema 记录结构,故不走 `impl_repo!` 宏,手写更直白。 -pub struct SettingsRepo { - conn: Arc>, -} - -impl SettingsRepo { - pub fn new(db: &Database) -> Self { - Self { conn: db.conn() } - } - - /// 取单个 key 的值(JSON 字符串),不存在返回 None - pub async fn get(&self, key: &str) -> Result> { - let conn = self.conn.clone(); - let key = key.to_owned(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let v = guard - .query_row( - "SELECT value FROM app_settings WHERE key = ?1", - params![key], - |row| row.get::<_, String>(0), - ) - .optional() - .map_err(storage_err)?; - Ok(v) - }) - .await - .map_err(storage_err)? - } - - /// 写 key/value(`INSERT OR REPLACE`),刷新 updated_at - pub async fn set(&self, key: &str, value: &str) -> Result { - let conn = self.conn.clone(); - let key = key.to_owned(); - let value = value.to_owned(); - let now = now_millis_str(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let affected = guard - .execute( - "INSERT OR REPLACE INTO app_settings (key, value, updated_at) VALUES (?1, ?2, ?3)", - params![key, value, now], - ) - .map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } - - /// 取全部 key/value - pub async fn get_all(&self) -> Result> { - let conn = self.conn.clone(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare("SELECT key, value FROM app_settings") - .map_err(storage_err)?; - let rows = stmt - .query_map([], |row| Ok((row.get::<_, String>(0)?, row.get::<_, String>(1)?))) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push(r.map_err(storage_err)?); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } - - /// 删除 key,返回是否实际删除 - pub async fn delete(&self, key: &str) -> Result { - let conn = self.conn.clone(); - let key = key.to_owned(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let affected = guard - .execute("DELETE FROM app_settings WHERE key = ?1", params![key]) - .map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } -} - -// ============================================================ -// 列名白名单 — 防止 SQL 注入 -// ============================================================ - -/// 按表返回允许用于 query / update_field 的列名。 -/// -/// 白名单两重作用:① 防注入(列名参数化前校验);② 按表隔离(update_task 误传 -/// projects 的 "name" 会在校验阶段拒绝,而非靠 SQLite "no such column" 兜底报错)。 -/// 专用更新路径的列不列入:knowledges.embedding(set_embedding)、projects.deleted_at -/// (soft_delete/restore)、tasks.deleted_at(soft_delete/restore)。未登记的表返回 None -/// → 放行(仅靠参数化防注入,向后兼容)。 -pub fn allowed_columns_for(table: &str) -> Option<&'static [&'static str]> { - Some(match table { - "ideas" => &[ - "id", "title", "description", "status", "priority", "score", "tags", "source", - "promoted_to", "ai_analysis", "scores", "created_at", "updated_at", - ], - "projects" => &[ - "id", "name", "description", "status", "idea_id", "path", "stack", "created_at", - "updated_at", - ], - "tasks" => &[ - // id/created_at 不列入:主键与创建时间不可通过通用 update_field 改写 - // (防篡改主键/伪造创建时间/跨项目移动)。 - // status 不列入(D-260616-04 status 收口):所有 status 改动走 - // advance_status_atomic(CAS SQL `WHERE id AND status=expected`,唯一 status 写入路径), - // 不经通用 update_field。防 update_task IPC / AI 工具旁路改 status 绕过状态机 - // (can_transition 校验)与 review_rounds 累加。后人勿把 status 补回白名单。 - // review_rounds 不列入:它仅 advance_status_atomic 退回转换 - // (in_review→in_progress / testing→in_review)时原子 +1(收口:仅该专用路径可改, - // update_task/update_field 白名单均不含)。后人勿把 review_rounds 补进白名单, - // 否则破坏「review_rounds 唯一写入路径」收口、引入旁路写导致计数错乱。 - "project_id", "title", "description", "priority", "branch_name", - "assignee", "workflow_def_id", "base_branch", - // output_json:ai_execute 写产出 / ai_self_review 读产出自审 / human_review 展示对象 - // (决策 a:task 中心,产出跟 task 走)。非状态机收口字段,合法可写。 - "output_json", - "updated_at", - // TODO(B-260616-16): project_id 跨表存在性校验待 commands/task.rs 层补。 - // 通用 CRUD 层(db repo)只懂表/列语义,不持有跨表业务约束(查 projects 表存在性)。 - // project_id 当前可在白名单内改写,合法目标存在性由上层命令层校验。 - ], - "releases" => &[ - "id", "project_id", "version", "status", "task_ids", "changelog", "created_at", - "released_at", - ], - "branches" => &[ - "id", "project_id", "task_id", "name", "base", "status", "created_at", "updated_at", - "merged_at", - ], - "workflow_executions" => &[ - "id", "name", "dag_json", "status", "triggered_by", "project_id", "task_id", - "created_at", "completed_at", - ], - "node_executions" => &[ - "id", "workflow_id", "node_id", "node_type", "status", "input_json", "output_json", - "error_message", "started_at", "completed_at", - ], - "ai_providers" => &[ - "id", "name", "provider_type", "api_key", "base_url", "default_model", "models", - "is_default", "config", "created_at", "updated_at", - ], - "ai_conversations" => &[ - "id", "title", "messages", "provider_id", "model", "models", "archived", - "prompt_tokens", "completion_tokens", "created_at", "updated_at", - ], - "ai_tool_executions" => &[ - "id", "conversation_id", "tool_call_id", "tool_name", "arguments", "result", - "status", "risk_level", "requested_at", "executed_at", "decided_by", - ], - "knowledges" => &[ - "id", "kind", "title", "content", "tags", "status", "confidence", "reuse_count", - "verified", "source_project", "source_ref", "reasoning", "created_at", "updated_at", - ], - "knowledge_events" => &[ - "id", "knowledge_id", "event_type", "source_ref", "context_json", "timestamp", - ], - _ => return None, - }) -} - -fn validate_column_name(field: &str, table: &str) -> Result<()> { - match allowed_columns_for(table) { - Some(cols) if cols.contains(&field) => Ok(()), - Some(_) => Err(Error::Storage(format!("表 {} 不允许的字段名: {}", table, field))), - // 未登记表保守拒绝(FR-S6: 原放行 Ok，若未来未登记表走通用查询路径，列名直进字符串拼接即 SQL 注入；与 is_allowed_column 的 None=>false 对齐) - None => Err(Error::Storage(format!("表 {} 未登记列白名单，拒绝防注入", table))), - } -} - -/// 是否允许更新某表的某列(按表隔离白名单)。未登记表返回 false(保守拒绝,防误传)。 -/// -/// 供 AI 工具层等外部调用方复用 CRUD 白名单,避免双份字段列表不同步。 -pub fn is_allowed_column(table: &str, field: &str) -> bool { - match allowed_columns_for(table) { - Some(cols) => cols.contains(&field), - None => false, - } -} - -// ============================================================ -// 时间工具 -// ============================================================ - -fn now_millis_str() -> String { - // 转发 df_types::now_millis(DRY:时间获取统一入口在 df-types,避免本 crate 与 commands 层各写一份 SystemTime 调用) - df_types::now_millis().to_string() -} - -// ============================================================ -// 向量工具 — embedding BLOB 序列化 + 余弦相似度 -// ============================================================ - -/// Vec → 小端字节 BLOB -fn f32s_to_blob(v: &[f32]) -> Vec { - v.iter().flat_map(|f| f.to_le_bytes()).collect() -} - -/// BLOB → Vec(长度非 4 倍数时截断尾部残字节) -fn blob_to_f32s(blob: &[u8]) -> Vec { - blob.chunks_exact(4) - .map(|c| f32::from_le_bytes([c[0], c[1], c[2], c[3]])) - .collect() -} - -/// 余弦相似度(确定性数学,与任何实现结果一致) -fn cosine_similarity(a: &[f32], b: &[f32]) -> f32 { - let dot: f32 = a.iter().zip(b).map(|(x, y)| x * y).sum(); - let norm_a: f32 = a.iter().map(|x| x * x).sum::().sqrt(); - let norm_b: f32 = b.iter().map(|x| x * x).sum::().sqrt(); - dot / (norm_a * norm_b + 1e-8) -} - -// ============================================================ -// from_row 辅助函数 -// ============================================================ - -fn idea_from_row(row: &Row<'_>) -> std::result::Result { - Ok(IdeaRecord { - id: row.get("id")?, - title: row.get("title")?, - description: row.get("description")?, - status: row.get("status")?, - priority: row.get("priority")?, - score: row.get("score")?, - tags: row.get("tags")?, - source: row.get("source")?, - promoted_to: row.get("promoted_to")?, - ai_analysis: row.get("ai_analysis")?, - scores: row.get("scores")?, - created_at: row.get("created_at")?, - updated_at: row.get("updated_at")?, - }) -} - -fn project_from_row(row: &Row<'_>) -> std::result::Result { - Ok(ProjectRecord { - id: row.get("id")?, - name: row.get("name")?, - description: row.get("description")?, - status: row.get("status")?, - idea_id: row.get("idea_id")?, - path: row.get("path")?, - stack: row.get("stack")?, - created_at: row.get("created_at")?, - updated_at: row.get("updated_at")?, - }) -} - -fn task_from_row(row: &Row<'_>) -> std::result::Result { - Ok(TaskRecord { - id: row.get("id")?, - project_id: row.get("project_id")?, - title: row.get("title")?, - description: row.get("description")?, - status: row.get("status")?, - priority: row.get("priority")?, - branch_name: row.get("branch_name")?, - assignee: row.get("assignee")?, - workflow_def_id: row.get("workflow_def_id")?, - base_branch: row.get("base_branch")?, - review_rounds: row.get("review_rounds")?, - output_json: row.get("output_json")?, - created_at: row.get("created_at")?, - updated_at: row.get("updated_at")?, - }) -} - -fn release_from_row(row: &Row<'_>) -> std::result::Result { - Ok(ReleaseRecord { - id: row.get("id")?, - project_id: row.get("project_id")?, - version: row.get("version")?, - status: row.get("status")?, - task_ids: row.get("task_ids")?, - changelog: row.get("changelog")?, - created_at: row.get("created_at")?, - released_at: row.get("released_at")?, - }) -} - -fn workflow_from_row(row: &Row<'_>) -> std::result::Result { - Ok(WorkflowRecord { - id: row.get("id")?, - name: row.get("name")?, - dag_json: row.get("dag_json")?, - status: row.get("status")?, - triggered_by: row.get("triggered_by")?, - project_id: row.get("project_id")?, - task_id: row.get("task_id")?, - created_at: row.get("created_at")?, - completed_at: row.get("completed_at")?, - }) -} - -fn branch_from_row(row: &Row<'_>) -> std::result::Result { - Ok(BranchRecord { - id: row.get("id")?, - project_id: row.get("project_id")?, - task_id: row.get("task_id")?, - name: row.get("name")?, - base: row.get("base")?, - status: row.get("status")?, - created_at: row.get("created_at")?, - updated_at: row.get("updated_at")?, - merged_at: row.get("merged_at")?, - }) -} - -fn node_execution_from_row( - row: &Row<'_>, -) -> std::result::Result { - Ok(NodeExecutionRecord { - id: row.get("id")?, - workflow_id: row.get("workflow_id")?, - node_id: row.get("node_id")?, - node_type: row.get("node_type")?, - status: row.get("status")?, - input_json: row.get("input_json")?, - output_json: row.get("output_json")?, - error_message: row.get("error_message")?, - started_at: row.get("started_at")?, - completed_at: row.get("completed_at")?, - }) -} - -// ============================================================ -// 7 个 Repo 实现 -// ============================================================ - -impl_repo!( - /// 想法表 CRUD - IdeaRepo, - IdeaRecord, - "ideas", - from_row => |row| idea_from_row(row), - insert => |conn, rec| { - conn.execute( - "INSERT INTO ideas (id, title, description, status, priority, score, tags, source, promoted_to, ai_analysis, scores, created_at, updated_at) - VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12, ?13)", - params![ - rec.id, rec.title, rec.description, rec.status, rec.priority, - rec.score, rec.tags, rec.source, rec.promoted_to, rec.ai_analysis, - rec.scores, rec.created_at, rec.updated_at - ], - ) - }, - update => |conn, rec| { - conn.execute( - "UPDATE ideas SET title = ?1, description = ?2, status = ?3, priority = ?4, score = ?5, tags = ?6, source = ?7, promoted_to = ?8, ai_analysis = ?9, scores = ?10, updated_at = ?11 WHERE id = ?12", - params![ - rec.title, rec.description, rec.status, rec.priority, - rec.score, rec.tags, rec.source, rec.promoted_to, rec.ai_analysis, - rec.scores, rec.updated_at, rec.id - ], - ) - } -); - -impl_repo!( - /// 项目表 CRUD - ProjectRepo, - ProjectRecord, - "projects", - from_row => |row| project_from_row(row), - insert => |conn, rec| { - conn.execute( - "INSERT INTO projects (id, name, description, status, idea_id, path, stack, created_at, updated_at) - VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9)", - params![ - rec.id, rec.name, rec.description, rec.status, rec.idea_id, - rec.path, rec.stack, rec.created_at, rec.updated_at - ], - ) - }, - update => |conn, rec| { - conn.execute( - "UPDATE projects SET name = ?1, description = ?2, status = ?3, idea_id = ?4, path = ?5, stack = ?6, updated_at = ?7 WHERE id = ?8", - params![ - rec.name, rec.description, rec.status, rec.idea_id, - rec.path, rec.stack, rec.updated_at, rec.id - ], - ) - } -); - -impl ProjectRepo { - /// 列出未删除项目(deleted_at IS NULL) - pub async fn list_active(&self) -> Result> { - let conn = self.conn.clone(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare("SELECT id, name, description, status, idea_id, path, stack, created_at, updated_at FROM projects WHERE deleted_at IS NULL ORDER BY created_at DESC") - .map_err(storage_err)?; - let rows = stmt - .query_map([], |row| project_from_row(row)) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push(r.map_err(storage_err)?); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } - - /// 软删:标记 deleted_at(进回收站,可恢复)。仅作用于未删项目,返回是否命中。 - pub async fn soft_delete(&self, id: &str) -> Result { - let conn = self.conn.clone(); - let id = id.to_owned(); - let now = now_millis_str(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let affected = guard - .execute( - "UPDATE projects SET deleted_at = ?1, updated_at = ?1 WHERE id = ?2 AND deleted_at IS NULL", - params![now, id], - ) - .map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } - - /// 恢复:清 deleted_at(从回收站还原) - pub async fn restore(&self, id: &str) -> Result { - let conn = self.conn.clone(); - let id = id.to_owned(); - let now = now_millis_str(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let affected = guard - .execute( - "UPDATE projects SET deleted_at = NULL, updated_at = ?1 WHERE id = ?2 AND deleted_at IS NOT NULL", - params![now, id], - ) - .map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } - - /// 查找已绑定该规范化路径的项目(排除 exclude_id 自身)。无冲突返回 None。 - /// - /// DRY(R-PD-11):统一 project.rs::find_binding_conflict 与 tool_registry.rs::bind_dir_to_project - /// 的「列项目→排除自身→按规范化路径比较」逻辑。本 crate 不依赖 df-project,故 norm_path 须由 - /// 调用方先经 `df_project::scan::normalize_path`(内含 canonicalize,防 `C:\a\b` vs `C:/a/b/` 绕过)。 - pub async fn find_path_conflict( - &self, - norm_path: &str, - exclude_id: Option<&str>, - ) -> Result> { - let projects = self.list_active().await?; - Ok(projects.into_iter().find(|p| { - let excluded = exclude_id.is_some_and(|eid| p.id == eid); - !excluded && p.path.as_ref().is_some_and(|pp| { - // 路径规范化由调用方完成目标侧;这里为已存项目路径补规范化 - // (历史路径写法可能不规范,统一规范化比较避免误判/漏判) - normalize_stored_path(pp) == norm_path - }) - })) - } - - /// 列出回收站(deleted_at IS NOT NULL),按更新时间(≈删除时间)降序 - pub async fn list_deleted(&self) -> Result> { - let conn = self.conn.clone(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare("SELECT id, name, description, status, idea_id, path, stack, created_at, updated_at FROM projects WHERE deleted_at IS NOT NULL ORDER BY updated_at DESC") - .map_err(storage_err)?; - let rows = stmt - .query_map([], |row| project_from_row(row)) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push(r.map_err(storage_err)?); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } - - /// 彻底删除:事务级联删 branches→releases→tasks→projects(不可恢复) - /// - /// SQLite 已开 PRAGMA foreign_keys=ON 但表无 ON DELETE CASCADE,ALTER 改不了 FK 约束, - /// 故应用层级联:子表先于父表删,单事务保证一致性。 - pub async fn purge_with_descendants(&self, id: &str) -> Result { - let conn = self.conn.clone(); - let id = id.to_owned(); - tokio::task::spawn_blocking(move || { - let mut guard = conn.blocking_lock(); - let tx = guard.transaction().map_err(storage_err)?; - tx.execute("DELETE FROM branches WHERE project_id = ?1", params![id]) - .map_err(storage_err)?; - tx.execute("DELETE FROM releases WHERE project_id = ?1", params![id]) - .map_err(storage_err)?; - tx.execute("DELETE FROM tasks WHERE project_id = ?1", params![id]) - .map_err(storage_err)?; - let affected = tx - .execute("DELETE FROM projects WHERE id = ?1", params![id]) - .map_err(storage_err)?; - tx.commit().map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } -} - -impl_repo!( - /// 任务表 CRUD - TaskRepo, - TaskRecord, - "tasks", - from_row => |row| task_from_row(row), - insert => |conn, rec| { - conn.execute( - "INSERT INTO tasks (id, project_id, title, description, status, priority, branch_name, assignee, workflow_def_id, base_branch, review_rounds, output_json, created_at, updated_at) - VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12, ?13, ?14)", - params![ - rec.id, rec.project_id, rec.title, rec.description, rec.status, rec.priority, - rec.branch_name, rec.assignee, rec.workflow_def_id, rec.base_branch, - rec.review_rounds, rec.output_json, rec.created_at, rec.updated_at - ], - ) - }, - update => |conn, rec| { - conn.execute( - "UPDATE tasks SET project_id = ?1, title = ?2, description = ?3, status = ?4, priority = ?5, branch_name = ?6, assignee = ?7, workflow_def_id = ?8, base_branch = ?9, review_rounds = ?10, output_json = ?11, updated_at = ?12 WHERE id = ?13", - params![ - rec.project_id, rec.title, rec.description, rec.status, rec.priority, - rec.branch_name, rec.assignee, rec.workflow_def_id, rec.base_branch, - rec.review_rounds, rec.output_json, rec.updated_at, rec.id - ], - ) - } -); - -impl TaskRepo { - /// 列出未删除任务(deleted_at IS NULL)— 对标 ProjectRepo::list_active - /// - /// 显式列出 14 个 TaskRecord 列名(同 ProjectRepo::list_active 写法), - /// 不 SELECT deleted_at:TaskRecord 不带该字段,取了 from_row 会因未知列报错。 - pub async fn list_active(&self) -> Result> { - let conn = self.conn.clone(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare("SELECT id, project_id, title, description, status, priority, branch_name, assignee, workflow_def_id, base_branch, review_rounds, output_json, created_at, updated_at FROM tasks WHERE deleted_at IS NULL ORDER BY created_at DESC") - .map_err(storage_err)?; - let rows = stmt - .query_map([], |row| task_from_row(row)) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push(r.map_err(storage_err)?); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } - - /// 软删:标记 deleted_at(进回收站,可恢复)。仅作用于未删任务,返回是否命中。 - /// 对标 ProjectRepo::soft_delete。 - pub async fn soft_delete(&self, id: &str) -> Result { - let conn = self.conn.clone(); - let id = id.to_owned(); - let now = now_millis_str(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let affected = guard - .execute( - "UPDATE tasks SET deleted_at = ?1, updated_at = ?1 WHERE id = ?2 AND deleted_at IS NULL", - params![now, id], - ) - .map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } - - /// 恢复:清 deleted_at(从回收站还原)。仅作用于已删任务,返回是否命中。 - /// 对标 ProjectRepo::restore。 - pub async fn restore(&self, id: &str) -> Result { - let conn = self.conn.clone(); - let id = id.to_owned(); - let now = now_millis_str(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let affected = guard - .execute( - "UPDATE tasks SET deleted_at = NULL, updated_at = ?1 WHERE id = ?2 AND deleted_at IS NOT NULL", - params![now, id], - ) - .map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } - - /// 原子推进任务状态(任务推进链 F-260616-02 唯一 status 写入路径) - /// - /// 下沉 SQL `WHERE id=? AND status=?expected` 做 CAS(Compare-And-Swap)防 TOCTOU: - /// 并发推进/旁路修改若已改 status,affected_rows==0,本方法返回 None,调用方 - /// (task_advance_node)据此报「状态已变,推进中止」。`review_rounds` 不进通用 - /// update_field 白名单(收口:仅本方法可改 status 与 review_rounds)。 - /// - /// - `expected`:调用方读取的当前 status(状态机校验时的 from),CAS 前置。 - /// - `new_status`:目标 status(状态机 can_transition 已校验合法)。 - /// - `bump_rounds`:退回转换(in_review→in_progress / testing→in_review)传 true, - /// 一并 `review_rounds = review_rounds + 1`(同 UPDATE 原子,避免读改写竞争)。 - /// 前向推进 / 进出 blocked / 进 cancelled 传 false,不动 review_rounds。 - /// - /// 返回:成功推进返回更新后的 TaskRecord;affected==0(状态已变/任务不存在)返回 None。 - pub async fn advance_status_atomic( - &self, - id: &str, - expected: &str, - new_status: &str, - bump_rounds: bool, - ) -> Result> { - let conn = self.conn.clone(); - let id = id.to_owned(); - let expected = expected.to_owned(); - let new_status = new_status.to_owned(); - let now = now_millis_str(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - // CAS:WHERE id AND status=expected 锁定当前态;affected==0 即并发已改动。 - // deleted_at IS NULL:回收站任务(soft_delete 设了 deleted_at)CAS 必败→affected=0, - // 返回 None,杜绝回收站任务被推进(D-02 软删语义收口,一处关闭)。 - let sql = if bump_rounds { - "UPDATE tasks SET status = ?1, review_rounds = review_rounds + 1, updated_at = ?2 \ - WHERE id = ?3 AND status = ?4 AND deleted_at IS NULL" - } else { - "UPDATE tasks SET status = ?1, updated_at = ?2 \ - WHERE id = ?3 AND status = ?4 AND deleted_at IS NULL" - }; - let affected = guard - .execute(sql, params![new_status, now, id, expected]) - .map_err(storage_err)?; - if affected == 0 { - return Ok(None); - } - // 回读更新后的记录(含新 status / 累加后的 review_rounds / 新 updated_at)。 - let mut stmt = guard - .prepare("SELECT id, project_id, title, description, status, priority, branch_name, assignee, workflow_def_id, base_branch, review_rounds, output_json, created_at, updated_at FROM tasks WHERE id = ?1") - .map_err(storage_err)?; - let row = stmt - .query_row(params![id], |row| task_from_row(row)) - .optional() - .map_err(storage_err)?; - Ok(row) - }) - .await - .map_err(storage_err)? - } - - /// 列出回收站(deleted_at IS NOT NULL),按更新时间(≈删除时间)降序。对标 ProjectRepo::list_deleted。 - /// - /// 注:任务表无专用 list_active_by_project 方法,按项目列活跃任务由 commands/task.rs - /// 的 list_tasks 用 list_active 后内存过滤 project_id 实现(任务量小,无需 SQL 下推)。 - pub async fn list_deleted(&self) -> Result> { - let conn = self.conn.clone(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare("SELECT id, project_id, title, description, status, priority, branch_name, assignee, workflow_def_id, base_branch, review_rounds, output_json, created_at, updated_at FROM tasks WHERE deleted_at IS NOT NULL ORDER BY updated_at DESC") - .map_err(storage_err)?; - let rows = stmt - .query_map([], |row| task_from_row(row)) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push(r.map_err(storage_err)?); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } -} - -impl_repo!( - /// 分支表 CRUD - BranchRepo, - BranchRecord, - "branches", - from_row => |row| branch_from_row(row), - insert => |conn, rec| { - conn.execute( - "INSERT INTO branches (id, project_id, task_id, name, base, status, created_at, updated_at, merged_at) - VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9)", - params![ - rec.id, rec.project_id, rec.task_id, rec.name, rec.base, - rec.status, rec.created_at, rec.updated_at, rec.merged_at - ], - ) - }, - update => |conn, rec| { - conn.execute( - "UPDATE branches SET project_id = ?1, task_id = ?2, name = ?3, base = ?4, status = ?5, updated_at = ?6, merged_at = ?7 WHERE id = ?8", - params![ - rec.project_id, rec.task_id, rec.name, rec.base, - rec.status, rec.updated_at, rec.merged_at, rec.id - ], - ) - } -); - -impl_repo!( - /// 发布表 CRUD - ReleaseRepo, - ReleaseRecord, - "releases", - from_row => |row| release_from_row(row), - insert => |conn, rec| { - conn.execute( - "INSERT INTO releases (id, project_id, version, status, task_ids, changelog, created_at, released_at) - VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8)", - params![ - rec.id, rec.project_id, rec.version, rec.status, rec.task_ids, - rec.changelog, rec.created_at, rec.released_at - ], - ) - }, - update => |conn, rec| { - conn.execute( - "UPDATE releases SET project_id = ?1, version = ?2, status = ?3, task_ids = ?4, changelog = ?5, released_at = ?6 WHERE id = ?7", - params![ - rec.project_id, rec.version, rec.status, rec.task_ids, - rec.changelog, rec.released_at, rec.id - ], - ) - } -); - -impl_repo!( - /// 工作流执行表 CRUD - WorkflowRepo, - WorkflowRecord, - "workflow_executions", - from_row => |row| workflow_from_row(row), - insert => |conn, rec| { - conn.execute( - "INSERT INTO workflow_executions (id, name, dag_json, status, triggered_by, project_id, task_id, created_at, completed_at) - VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9)", - params![ - rec.id, rec.name, rec.dag_json, rec.status, rec.triggered_by, - rec.project_id, rec.task_id, rec.created_at, rec.completed_at - ], - ) - }, - update => |conn, rec| { - conn.execute( - "UPDATE workflow_executions SET name = ?1, dag_json = ?2, status = ?3, triggered_by = ?4, project_id = ?5, task_id = ?6, completed_at = ?7 WHERE id = ?8", - params![ - rec.name, rec.dag_json, rec.status, rec.triggered_by, - rec.project_id, rec.task_id, rec.completed_at, rec.id - ], - ) - } -); - -impl_repo!( - /// 节点执行表 CRUD - NodeExecutionRepo, - NodeExecutionRecord, - "node_executions", - from_row => |row| node_execution_from_row(row), - insert => |conn, rec| { - conn.execute( - "INSERT INTO node_executions (id, workflow_id, node_id, node_type, status, input_json, output_json, error_message, started_at, completed_at) - VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10)", - params![ - rec.id, rec.workflow_id, rec.node_id, rec.node_type, rec.status, - rec.input_json, rec.output_json, rec.error_message, rec.started_at, rec.completed_at - ], - ) - }, - update => |conn, rec| { - conn.execute( - "UPDATE node_executions SET workflow_id = ?1, node_id = ?2, node_type = ?3, status = ?4, input_json = ?5, output_json = ?6, error_message = ?7, started_at = ?8, completed_at = ?9 WHERE id = ?10", - params![ - rec.workflow_id, rec.node_id, rec.node_type, rec.status, - rec.input_json, rec.output_json, rec.error_message, rec.started_at, rec.completed_at, rec.id - ], - ) - } -); - -// ============================================================ -// AI 相关 Repo (V3) -// ============================================================ - -fn ai_provider_from_row(row: &Row<'_>) -> std::result::Result { - // model_configs:DB TEXT 列存 JSON 字符串。读 Option 兼容老库 NULL, - // 再经 deserialize_model_configs 解析(老字符串数组/新对象数组/空 → Vec)。 - // 解析失败不致命:降级为空 Vec(防单行坏数据拖垮 list_all)。 - let model_configs: Vec = { - let raw: Option = row.get("model_configs").ok(); - match raw { - None => Vec::new(), - Some(s) => serde_json::from_str::(&format!( - r#"{{"v":{}}}"#, - if s.trim().is_empty() { - "null".to_string() - } else if s.trim_start().starts_with('[') || s.trim_start().starts_with('{') { - s - } else { - // 非 JSON 字面文本(理论不会出现)→ 包装为 JSON 字符串让 deserialize 兜底 - serde_json::to_string(&s).unwrap_or_else(|_| "null".into()) - } - )) - .map(|w| w.v) - .unwrap_or_default(), - } - }; - Ok(AiProviderRecord { - id: row.get("id")?, - name: row.get("name")?, - provider_type: row.get("provider_type")?, - api_key: row.get("api_key")?, - base_url: row.get("base_url")?, - default_model: row.get("default_model")?, - models: row.get("models")?, - model_configs, - is_default: row.get::<_, i32>("is_default")? != 0, - config: row.get("config")?, - created_at: row.get("created_at")?, - updated_at: row.get("updated_at")?, - // F-260614-04: enabled/weight 列老库经 v19 迁移补建,DEFAULT 1 / DEFAULT 50。 - // from_row 按 i32 取列值兼容(SQLite 无真 BOOLEAN),0→false/非0→true。 - enabled: row.get::<_, i32>("enabled").unwrap_or(1) != 0, - // weight 读侧 clamp [0,100]:与 insert/update_full 落库的 `.min(100)` 对齐, - // 防老库(clamp 落地前写入的)或外部直改 DB 产生的越界值污染路由权重语义。 - weight: row.get::<_, i32>("weight") - .unwrap_or(50) - .clamp(0, 100) as u32, - }) -} - -/// from_row 内部辅助:复用 deserialize_model_configs 解析 DB TEXT 列 JSON。 -/// 包一层 { "v": <原始值> } 把任意 JSON 值送进 deserialize_model_configs。 -#[derive(serde::Deserialize)] -struct ModelConfigsWrap { - #[serde(default, deserialize_with = "df_ai_core::model::deserialize_model_configs")] - v: Vec, -} - -fn ai_conversation_from_row(row: &Row<'_>) -> std::result::Result { - Ok(AiConversationRecord { - id: row.get("id")?, - title: row.get("title")?, - messages: row.get("messages")?, - provider_id: row.get("provider_id")?, - model: row.get("model")?, - models: row.get("models")?, - archived: row.get::<_, i32>("archived")? != 0, - pinned: row.get::<_, i32>("pinned")? != 0, - prompt_tokens: row.get("prompt_tokens")?, - completion_tokens: row.get("completion_tokens")?, - created_at: row.get("created_at")?, - updated_at: row.get("updated_at")?, - }) -} - -fn ai_tool_execution_from_row(row: &Row<'_>) -> std::result::Result { - Ok(AiToolExecutionRecord { - id: row.get("id")?, - conversation_id: row.get("conversation_id")?, - tool_call_id: row.get("tool_call_id")?, - tool_name: row.get("tool_name")?, - arguments: row.get("arguments")?, - result: row.get("result")?, - status: row.get("status")?, - risk_level: row.get("risk_level")?, - requested_at: row.get("requested_at")?, - executed_at: row.get("executed_at")?, - decided_by: row.get("decided_by")?, - }) -} - -fn knowledge_from_row(row: &Row<'_>) -> std::result::Result { - Ok(KnowledgeRecord { - id: row.get("id")?, - kind: row.get("kind")?, - title: row.get("title")?, - content: row.get("content")?, - tags: row.get("tags")?, - status: row.get("status")?, - confidence: row.get("confidence")?, - reuse_count: row.get("reuse_count")?, - verified: row.get::<_, i32>("verified")? != 0, - source_project: row.get("source_project")?, - source_ref: row.get("source_ref")?, - reasoning: row.get("reasoning")?, - created_at: row.get("created_at")?, - updated_at: row.get("updated_at")?, - }) -} - -fn knowledge_event_from_row(row: &Row<'_>) -> std::result::Result { - Ok(KnowledgeEventRecord { - id: row.get("id")?, - knowledge_id: row.get("knowledge_id")?, - event_type: row.get("event_type")?, - source_ref: row.get("source_ref")?, - context_json: row.get("context_json")?, - timestamp: row.get("timestamp")?, - }) -} - -impl_repo!( - /// AI 提供商配置表 CRUD - AiProviderRepo, - AiProviderRecord, - "ai_providers", - from_row => |row| ai_provider_from_row(row), - insert => |conn, rec| { - let is_default = if rec.is_default { 1i32 } else { 0i32 }; - // model_configs:Vec → JSON 字符串落 TEXT 列 - let model_configs_json = serde_json::to_string(&rec.model_configs).unwrap_or_else(|_| "[]".into()); - // F-260614-04: enabled/weight 落库(SQLite 无 BOOLEAN,i32 承载)。 - let enabled_i = if rec.enabled { 1i32 } else { 0i32 }; - let weight_i = rec.weight.min(100) as i32; - conn.execute( - "INSERT OR REPLACE INTO ai_providers (id, name, provider_type, api_key, base_url, default_model, models, model_configs, is_default, config, created_at, updated_at, enabled, weight) - VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12, ?13, ?14)", - params![ - rec.id, rec.name, rec.provider_type, rec.api_key, rec.base_url, - rec.default_model, rec.models, model_configs_json, is_default, rec.config, rec.created_at, rec.updated_at, - enabled_i, weight_i - ], - ) - }, - update => |conn, rec| { - let is_default = if rec.is_default { 1i32 } else { 0i32 }; - let model_configs_json = serde_json::to_string(&rec.model_configs).unwrap_or_else(|_| "[]".into()); - let enabled_i = if rec.enabled { 1i32 } else { 0i32 }; - let weight_i = rec.weight.min(100) as i32; - conn.execute( - "UPDATE ai_providers SET name = ?1, provider_type = ?2, api_key = ?3, base_url = ?4, default_model = ?5, models = ?6, model_configs = ?7, is_default = ?8, config = ?9, updated_at = ?10, enabled = ?11, weight = ?12 WHERE id = ?13", - params![ - rec.name, rec.provider_type, rec.api_key, rec.base_url, - rec.default_model, rec.models, model_configs_json, is_default, rec.config, rec.updated_at, - enabled_i, weight_i, rec.id - ], - ) - } -); - -impl_repo!( - /// AI 对话历史表 CRUD - AiConversationRepo, - AiConversationRecord, - "ai_conversations", - from_row => |row| ai_conversation_from_row(row), - insert => |conn, rec| { - conn.execute( - "INSERT INTO ai_conversations (id, title, messages, provider_id, model, models, archived, pinned, prompt_tokens, completion_tokens, created_at, updated_at) - VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12)", - params![ - rec.id, rec.title, rec.messages, rec.provider_id, rec.model, rec.models, rec.archived, - if rec.pinned { 1i32 } else { 0i32 }, - rec.prompt_tokens, rec.completion_tokens, - rec.created_at, rec.updated_at - ], - ) - }, - update => |conn, rec| { - conn.execute( - "UPDATE ai_conversations SET title = ?1, messages = ?2, provider_id = ?3, model = ?4, models = ?5, archived = ?6, pinned = ?7, prompt_tokens = ?8, completion_tokens = ?9, updated_at = ?10 WHERE id = ?11", - params![ - rec.title, rec.messages, rec.provider_id, rec.model, rec.models, rec.archived, - if rec.pinned { 1i32 } else { 0i32 }, - rec.prompt_tokens, rec.completion_tokens, - rec.updated_at, rec.id - ], - ) - } -); - -impl_repo!( - /// AI 工具执行审计表 CRUD - AiToolExecutionRepo, - AiToolExecutionRecord, - "ai_tool_executions", - from_row => |row| ai_tool_execution_from_row(row), - insert => |conn, rec| { - conn.execute( - "INSERT INTO ai_tool_executions (id, conversation_id, tool_call_id, tool_name, arguments, result, status, risk_level, requested_at, executed_at, decided_by) - VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11)", - params![ - rec.id, rec.conversation_id, rec.tool_call_id, rec.tool_name, - rec.arguments, rec.result, rec.status, rec.risk_level, - rec.requested_at, rec.executed_at, rec.decided_by - ], - ) - }, - update => |conn, rec| { - conn.execute( - "UPDATE ai_tool_executions SET conversation_id = ?1, tool_call_id = ?2, tool_name = ?3, arguments = ?4, result = ?5, status = ?6, risk_level = ?7, requested_at = ?8, executed_at = ?9, decided_by = ?10 WHERE id = ?11", - params![ - rec.conversation_id, rec.tool_call_id, rec.tool_name, - rec.arguments, rec.result, rec.status, rec.risk_level, - rec.requested_at, rec.executed_at, rec.decided_by, rec.id - ], - ) - } -); - -// ai_tool_executions 无 created_at 列（用 requested_at/executed_at 计时）， -// 通用 query 宏硬编码 ORDER BY created_at 会触发 "no such column" → 调用方 unwrap_or_default 吞错。 -// 故为此表提供专用查询，绕过通用 query。详见 ai.rs audit_finalize。 -impl AiToolExecutionRepo { - /// 按 tool_call_id 查最新一条审计记录（审批回填定位用）。 - pub async fn find_by_tool_call_id( - &self, - tool_call_id: &str, - ) -> Result> { - let conn = self.conn.clone(); - let tid = tool_call_id.to_owned(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare( - "SELECT * FROM ai_tool_executions WHERE tool_call_id = ?1 ORDER BY requested_at DESC LIMIT 1", - ) - .map_err(storage_err)?; - let row = stmt - .query_row(params![tid], |row| ai_tool_execution_from_row(row)) - .optional() - .map_err(storage_err)?; - Ok(row) - }) - .await - .map_err(storage_err)? - } - - /// 列出所有 status=pending 的审计行(启动重建 pending_approvals 用) - /// - /// 专用 SELECT(非 query 宏——后者硬编码 ORDER BY created_at,而本表无该列)。 - pub async fn list_pending(&self) -> Result> { - let conn = self.conn.clone(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare("SELECT * FROM ai_tool_executions WHERE status = 'pending' ORDER BY requested_at ASC") - .map_err(storage_err)?; - let rows = stmt - .query_map([], |row| ai_tool_execution_from_row(row)) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push(r.map_err(storage_err)?); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } - - /// 审批历史面板分页查询:按 requested_at 倒序(最新在前),limit 默认 50。 - /// - /// 与 list_pending 同理走专用 SELECT,绕过通用 query 宏(后者硬编码 - /// ORDER BY created_at,本表无该列)。limit/offset 上限钳制(limit ≤ 200), - /// 防前端恶意/失误传超大值。 - pub async fn list_recent( - &self, - limit: u32, - offset: u32, - ) -> Result> { - let conn = self.conn.clone(); - // 钳制 limit 防滥用(默认 50,最大 200) - let safe_limit = limit.min(200) as i64; - let safe_offset = offset as i64; - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare( - "SELECT * FROM ai_tool_executions ORDER BY requested_at DESC LIMIT ?1 OFFSET ?2", - ) - .map_err(storage_err)?; - let rows = stmt - .query_map(params![safe_limit, safe_offset], |row| { - ai_tool_execution_from_row(row) - }) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push(r.map_err(storage_err)?); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } -} - -impl_repo!( - /// 知识库表 CRUD - KnowledgeRepo, - KnowledgeRecord, - "knowledges", - from_row => |row| knowledge_from_row(row), - insert => |conn, rec| { - let verified = if rec.verified { 1i32 } else { 0i32 }; - conn.execute( - "INSERT INTO knowledges (id, kind, title, content, tags, status, confidence, reuse_count, verified, source_project, source_ref, reasoning, created_at, updated_at) - VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12, ?13, ?14)", - params![ - rec.id, rec.kind, rec.title, rec.content, rec.tags, rec.status, rec.confidence, - rec.reuse_count, verified, rec.source_project, rec.source_ref, rec.reasoning, - rec.created_at, rec.updated_at - ], - ) - }, - update => |conn, rec| { - let verified = if rec.verified { 1i32 } else { 0i32 }; - conn.execute( - "UPDATE knowledges SET kind = ?1, title = ?2, content = ?3, tags = ?4, status = ?5, confidence = ?6, reuse_count = ?7, verified = ?8, source_project = ?9, source_ref = ?10, reasoning = ?11, updated_at = ?12 WHERE id = ?13", - params![ - rec.kind, rec.title, rec.content, rec.tags, rec.status, rec.confidence, - rec.reuse_count, verified, rec.source_project, rec.source_ref, rec.reasoning, - rec.updated_at, rec.id - ], - ) - } -); - -// KnowledgeRepo 的整体更新已由 impl_repo! 宏统一生成的 update_full 提供。 - -impl KnowledgeRepo { - /// 检索知识: title/content LIKE 匹配,可选 kind 过滤,按 reuse_count 降序,top-N - /// - /// 克制检索: top-N≤3(由调用方 limit 控制),精确匹配优先(语义模糊后做)。 - pub async fn search(&self, query: &str, kind: Option<&str>, limit: usize) -> Result> { - let conn = self.conn.clone(); - let pattern = format!("%{}%", query); - let kind = kind.map(|s| s.to_owned()); - let limit_i = limit as i64; - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut results = Vec::new(); - if let Some(k) = &kind { - let mut stmt = guard - .prepare(&format!("SELECT {KNOWLEDGE_COLS} FROM knowledges WHERE status = 'published' AND (title LIKE ?1 OR content LIKE ?2) AND kind = ?3 ORDER BY reuse_count DESC LIMIT ?4")) - .map_err(storage_err)?; - let rows = stmt - .query_map(params![pattern, pattern, k, limit_i], |row| knowledge_from_row(row)) - .map_err(storage_err)?; - for r in rows { - results.push(r.map_err(storage_err)?); - } - } else { - let mut stmt = guard - .prepare(&format!("SELECT {KNOWLEDGE_COLS} FROM knowledges WHERE status = 'published' AND (title LIKE ?1 OR content LIKE ?2) ORDER BY reuse_count DESC LIMIT ?3")) - .map_err(storage_err)?; - let rows = stmt - .query_map(params![pattern, pattern, limit_i], |row| knowledge_from_row(row)) - .map_err(storage_err)?; - for r in rows { - results.push(r.map_err(storage_err)?); - } - } - Ok(results) - }) - .await - .map_err(storage_err)? - } - - /// 按状态列出(审核收件箱用): 按 confidence 语义排序(high>medium>low),次按 created_at - /// - /// 用 CASE WHEN 替代纯 TEXT 排序(字典序 high>low>medium 非预期语义)。 - pub async fn list_by_status(&self, status: &str) -> Result> { - let conn = self.conn.clone(); - let status = status.to_owned(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare( - "SELECT id,kind,title,content,tags,status,confidence,reuse_count,verified,\ - source_project,source_ref,reasoning,created_at,updated_at \ - FROM knowledges WHERE status = ?1 - ORDER BY CASE confidence - WHEN 'high' THEN 3 - WHEN 'medium' THEN 2 - WHEN 'low' THEN 1 - ELSE 0 - END DESC, created_at DESC", - ) - .map_err(storage_err)?; - let rows = stmt - .query_map(params![status], |row| knowledge_from_row(row)) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push(r.map_err(storage_err)?); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } - - /// 复用计数 +1(SQL 行级原子操作,并发安全) - pub async fn increment_reuse_count(&self, id: &str) -> Result { - let conn = self.conn.clone(); - let id = id.to_owned(); - let now = now_millis_str(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let affected = guard - .execute( - "UPDATE knowledges SET reuse_count = reuse_count + 1, updated_at = ?1 WHERE id = ?2", - params![now, id], - ) - .map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } - - /// 写入向量嵌入(BLOB = Vec 小端字节序列化) - /// - /// embedding 列不进 KnowledgeRecord(IPC 不需要传向量给前端),专用方法读写。 - pub async fn set_embedding(&self, id: &str, embedding: &[f32]) -> Result { - let conn = self.conn.clone(); - let id = id.to_owned(); - let blob = f32s_to_blob(embedding); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let affected = guard - .execute( - "UPDATE knowledges SET embedding = ?1 WHERE id = ?2", - params![blob, id], - ) - .map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } - - /// 向量检索: 加载全部 published 且有 embedding 的记录,纯 Rust 余弦相似度取 top-N - /// - /// 返回 (记录, 相似度分数)。数据量 <50k 时暴力遍历 <50ms,够用; - /// 更大规模再升 sqlite-vec HNSW(结果不变,只提速)。 - /// - /// 列限定: 显式列出所需列(与 search/list_by_status/top_used 一致),避免 SELECT * - /// 拉到未知新增列;embedding 单独取(不入 KnowledgeRecord)。 - /// - /// TODO(性能,低优先): 调用方(hybrid_search→merge_hybrid_results→build_knowledge_context) - /// 实际只消费 id/kind/title/content/reuse_count;reasoning(AI 生成大文本)、tags、 - /// source_project/source_ref 等元字段未被使用却仍随每行读出。真正省 IO 需返回精简结构 - /// (如 KnowledgeVectorHit { id, kind, title, content, reuse_count })替换返回类型, - /// 但这会改变 search_vector 签名与 merge_hybrid_results 调用契约——当前保守不动, - /// 待向量检索量级或 reasoning 文本体积成为瓶颈再单独立项。SELECT 列化本身不省字段, - /// 仅消除 SELECT * 的隐式依赖与未知列风险。 - pub async fn search_vector(&self, query_vec: &[f32], limit: usize) -> Result> { - let conn = self.conn.clone(); - let query_vec = query_vec.to_vec(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - // 显式列: 14 个 KnowledgeRecord 字段 + embedding(余弦计算用,不入 KnowledgeRecord) - let mut stmt = guard - .prepare(&format!( - "SELECT {KNOWLEDGE_COLS_WITH_EMBEDDING} FROM knowledges WHERE status = 'published' AND embedding IS NOT NULL" - )) - .map_err(storage_err)?; - let rows = stmt - .query_map([], |row| { - let rec = knowledge_from_row(row)?; - let blob: Vec = row.get("embedding")?; - Ok((rec, blob)) - }) - .map_err(storage_err)?; - - let mut scored: Vec<(KnowledgeRecord, f32)> = Vec::new(); - for r in rows { - let (rec, blob) = r.map_err(storage_err)?; - let emb = blob_to_f32s(&blob); - // 维度不匹配(换过 embedding 模型的旧向量)跳过 - if emb.len() != query_vec.len() { - continue; - } - let score = cosine_similarity(&query_vec, &emb); - scored.push((rec, score)); - } - scored.sort_by(|a, b| b.1.partial_cmp(&a.1).unwrap_or(std::cmp::Ordering::Equal)); - scored.truncate(limit); - Ok(scored) - }) - .await - .map_err(storage_err)? - } - - /// 列出非归档知识(全部 status != 'archived'),按 confidence 语义排序 - pub async fn list_non_archived(&self) -> Result> { - let conn = self.conn.clone(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare( - "SELECT id,kind,title,content,tags,status,confidence,reuse_count,verified,\ - source_project,source_ref,reasoning,created_at,updated_at \ - FROM knowledges WHERE status != 'archived' - ORDER BY CASE confidence - WHEN 'high' THEN 3 - WHEN 'medium' THEN 2 - WHEN 'low' THEN 1 - ELSE 0 - END DESC, created_at DESC", - ) - .map_err(storage_err)?; - let rows = stmt - .query_map([], |row| knowledge_from_row(row)) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push(r.map_err(storage_err)?); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } - - /// 热门知识(已发布,按复用次数降序) - pub async fn top_used(&self, limit: usize) -> Result> { - let conn = self.conn.clone(); - let limit_i = limit as i64; - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare("SELECT id,kind,title,content,tags,status,confidence,reuse_count,verified,source_project,source_ref,reasoning,created_at,updated_at FROM knowledges WHERE status = 'published' ORDER BY reuse_count DESC LIMIT ?1") - .map_err(storage_err)?; - let rows = stmt - .query_map(params![limit_i], |row| knowledge_from_row(row)) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push(r.map_err(storage_err)?); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } -} - -impl_repo!( - /// 知识生命线事件表 CRUD(追加型审计表) - KnowledgeEventsRepo, - KnowledgeEventRecord, - "knowledge_events", - from_row => |row| knowledge_event_from_row(row), - insert => |conn, rec| { - conn.execute( - "INSERT INTO knowledge_events (id, knowledge_id, event_type, source_ref, context_json, timestamp) - VALUES (?1, ?2, ?3, ?4, ?5, ?6)", - params![rec.id, rec.knowledge_id, rec.event_type, rec.source_ref, rec.context_json, rec.timestamp], - ) - }, - update => |conn, rec| { - conn.execute( - "UPDATE knowledge_events SET knowledge_id = ?1, event_type = ?2, source_ref = ?3, context_json = ?4, timestamp = ?5 WHERE id = ?6", - params![rec.knowledge_id, rec.event_type, rec.source_ref, rec.context_json, rec.timestamp, rec.id], - ) - } -); - -impl KnowledgeEventsRepo { - /// 按知识 ID 查询全部事件(时间正序,构建生命线视图用) - pub async fn list_by_knowledge(&self, knowledge_id: &str) -> Result> { - let conn = self.conn.clone(); - let knowledge_id = knowledge_id.to_owned(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare("SELECT id,knowledge_id,event_type,source_ref,context_json,timestamp FROM knowledge_events WHERE knowledge_id = ?1 ORDER BY timestamp ASC") - .map_err(storage_err)?; - let rows = stmt - .query_map(params![knowledge_id], |row| knowledge_event_from_row(row)) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push(r.map_err(storage_err)?); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } - - /// 按知识 ID + event_type 查询最近 N 条(如引用记录翻页: type=referenced) - pub async fn list_by_knowledge_type( - &self, - knowledge_id: &str, - event_type: &str, - limit: usize, - ) -> Result> { - let conn = self.conn.clone(); - let knowledge_id = knowledge_id.to_owned(); - let event_type = event_type.to_owned(); - let limit_i = limit as i64; - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare("SELECT id,knowledge_id,event_type,source_ref,context_json,timestamp FROM knowledge_events WHERE knowledge_id = ?1 AND event_type = ?2 ORDER BY timestamp DESC LIMIT ?3") - .map_err(storage_err)?; - let rows = stmt - .query_map(params![knowledge_id, event_type, limit_i], |row| knowledge_event_from_row(row)) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push(r.map_err(storage_err)?); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } - - /// 跨知识列最近 N 条事件(全表 timestamp DESC,top-N)。 - /// - /// **专用兜底方法**:本表时间列名是 `timestamp` 而非 `created_at`,但 - /// `impl_repo!` 宏生成的 `query()` / `list_all()` 硬编码 `ORDER BY created_at` - /// (见宏内 `ORDER BY created_at DESC` 字面量),误调 `state.knowledge_events.query(...)` - /// 会触发 SQLite "no such column: created_at"。本方法走专用 SELECT 绕过宏硬编码, - /// 供需要跨知识按时间倒序浏览事件的调用方使用(对标 AiToolExecutionRepo::list_recent - /// 对 ai_tool_executions 表的同款兜底处理——那张表同样无 created_at 列)。 - /// limit 上限钳制 200,防前端恶意/失误传超大值。 - pub async fn list_recent(&self, limit: u32) -> Result> { - let conn = self.conn.clone(); - // 钳制 limit 防滥用(最大 200) - let safe_limit = limit.min(200) as i64; - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let mut stmt = guard - .prepare( - "SELECT id,knowledge_id,event_type,source_ref,context_json,timestamp \ - FROM knowledge_events ORDER BY timestamp DESC LIMIT ?1", - ) - .map_err(storage_err)?; - let rows = stmt - .query_map(params![safe_limit], |row| knowledge_event_from_row(row)) - .map_err(storage_err)?; - let mut results = Vec::new(); - for r in rows { - results.push(r.map_err(storage_err)?); - } - Ok(results) - }) - .await - .map_err(storage_err)? - } -} - -// AiConversationRepo 的整体更新已由 impl_repo! 宏统一生成的 update_full 提供。 - -impl AiConversationRepo { - /// 清空对话消息内容(保留 conversation 记录本身,只清 messages JSON + 清零 token 计数) - /// - /// "清空对话"语义:对话壳保留(侧栏仍可见,可继续在该对话内聊),仅清空历史消息。 - /// messages 是 ai_conversations 表内的 JSON 列而非独立行,故"删 messages"= 置空该列。 - pub async fn clear_messages(&self, id: &str) -> Result { - let conn = self.conn.clone(); - let id = id.to_owned(); - let now = now_millis_str(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let affected = guard - .execute( - "UPDATE ai_conversations SET messages = '[]', prompt_tokens = 0, completion_tokens = 0, updated_at = ?1 WHERE id = ?2", - params![now, id], - ) - .map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } - - /// 设置归档标记(仅改 archived,不动 updated_at) - /// - /// 区别于 update_field(后者强制 SET updated_at=now,会把归档/取消归档误判为内容更新, - /// 导致侧栏相对时间跳变为"刚刚")。归档是纯元数据标记,应保持时间不变。 - pub async fn set_archived(&self, id: &str, archived: bool) -> Result { - let conn = self.conn.clone(); - let id = id.to_owned(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let affected = guard - .execute( - "UPDATE ai_conversations SET archived = ?1 WHERE id = ?2", - params![if archived { 1 } else { 0 }, id], - ) - .map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } - - /// 设置置顶标记(仅改 pinned,不动 updated_at) — UX-17 - /// - /// 同 set_archived:置顶是纯元数据标记,不应改变相对时间。前端排序读 pinned DESC, updated_at DESC。 - pub async fn set_pinned(&self, id: &str, pinned: bool) -> Result { - let conn = self.conn.clone(); - let id = id.to_owned(); - tokio::task::spawn_blocking(move || { - let guard = conn.blocking_lock(); - let affected = guard - .execute( - "UPDATE ai_conversations SET pinned = ?1 WHERE id = ?2", - params![if pinned { 1 } else { 0 }, id], - ) - .map_err(storage_err)?; - Ok(affected > 0) - }) - .await - .map_err(storage_err)? - } -} - -// ============================================================ -// 单元测试 — 向量纯函数 + KnowledgeRepo 内存 DB -// ============================================================ - -#[cfg(test)] -mod tests { - use super::*; - use crate::db::Database; - use crate::models::KnowledgeRecord; - - // ---------- COLS 漂移防护(CR-260615-03) ---------- - - /// model_configs DB roundtrip + 老库空兼容(F-01 阶段1) - #[tokio::test] - async fn ai_provider_model_configs_roundtrip_and_old_db_compat() { - let db = Database::open_in_memory().await.expect("open_in_memory"); - let repo = AiProviderRepo::new(&db); - use crate::models::AiProviderRecord; - use df_ai_core::model::{Capability, IntelligenceTier, Modality, ModelConfig}; - - // 新格式:带多模型 + 多维度配置 - let configs = vec![ - ModelConfig::with_defaults("glm-4-flash"), - ModelConfig { - model_id: "glm-4v".into(), - modalities: vec![Modality::Text, Modality::Vision], - capabilities: vec![Capability::ToolUse], - intelligence: IntelligenceTier::Plus, - weight: 70, - context_window: 128_000, - ..ModelConfig::with_defaults("glm-4v") - }, - ]; - let rec = AiProviderRecord { - id: "p1".into(), - name: "测试".into(), - provider_type: "openai_compat".into(), - api_key: String::new(), - base_url: "https://x".into(), - default_model: "glm-4-flash".into(), - models: None, - model_configs: configs.clone(), - is_default: false, - config: None, - created_at: "0".into(), - updated_at: "0".into(), - enabled: true, - weight: 50, - }; - repo.insert(rec).await.expect("insert"); - - let got = repo.get_by_id("p1").await.expect("get").expect("row exists"); - assert_eq!(got.model_configs.len(), 2); - assert_eq!(got.model_configs[0].model_id, "glm-4-flash"); - assert_eq!(got.model_configs[1].model_id, "glm-4v"); - assert_eq!(got.model_configs[1].intelligence, IntelligenceTier::Plus); - assert_eq!(got.model_configs[1].context_window, 128_000); - - // 老库空兼容:直接写 model_configs=NULL 的行(模拟 V18 之前的老库行) - // 然后 from_row 应得空 Vec - { - let conn = db.conn(); - let g = conn.lock().await; - g.execute( - "INSERT OR REPLACE INTO ai_providers \ - (id,name,provider_type,api_key,base_url,default_model,models,model_configs,is_default,config,created_at,updated_at) \ - VALUES ('old','','openai_compat','','','','{}',NULL,0,NULL,'0','0')", - [], - ) - .expect("raw insert old row"); - } - let old = repo.get_by_id("old").await.expect("get").expect("old row"); - assert!(old.model_configs.is_empty(), "NULL 列应得空 Vec"); - - // 老格式字符串数组 JSON(向后兼容 deserialize_model_configs) - { - let conn = db.conn(); - let g = conn.lock().await; - g.execute( - "INSERT OR REPLACE INTO ai_providers \ - (id,name,provider_type,api_key,base_url,default_model,models,model_configs,is_default,config,created_at,updated_at) \ - VALUES ('legacy','','openai_compat','','','','{}','[\"glm-4-flash\",\"glm-4v\"]',0,NULL,'0','0')", - [], - ) - .expect("raw insert legacy row"); - } - let legacy = repo.get_by_id("legacy").await.expect("get").expect("legacy row"); - assert_eq!(legacy.model_configs.len(), 2, "老字符串数组应转 2 个默认 ModelConfig"); - assert_eq!(legacy.model_configs[0].model_id, "glm-4-flash"); - } - - /// KNOWLEDGE_COLS 列数须等于 KNOWLEDGE_COL_COUNT(任一处漂移:加列漏改 / 串错位 → 立即失败)。 - /// `knowledge_from_row` 按 name 取列,SELECT 漏列会在运行时被 rusqlite 报错;此断言提前到测试期捕获。 - #[test] - fn test_knowledge_cols_matches_record() { - let count = KNOWLEDGE_COLS.split(',').count(); - assert_eq!( - count, KNOWLEDGE_COL_COUNT, - "KNOWLEDGE_COLS({count}列) ≠ KNOWLEDGE_COL_COUNT({KNOWLEDGE_COL_COUNT}); \ - 修改一处须同步另一处" - ); - // search_vector 多一列 embedding - let count_with_emb = KNOWLEDGE_COLS_WITH_EMBEDDING.split(',').count(); - assert_eq!( - count_with_emb, - KNOWLEDGE_COL_COUNT + 1, - "KNOWLEDGE_COLS_WITH_EMBEDDING({count_with_emb}列) ≠ KNOWLEDGE_COL_COUNT+1({}); \ - embedding 列应单独追加", - KNOWLEDGE_COL_COUNT + 1 - ); - // 每个列名须能被 split 出来(防末尾多逗号 / 空段) - for col in KNOWLEDGE_COLS.split(',') { - assert!(!col.is_empty(), "KNOWLEDGE_COLS 含空列名段"); - } - } - - // ---------- 向量纯函数 ---------- - - #[test] - fn f32s_blob_roundtrip_nonempty() { - let v = vec![0.0, 1.5, -2.25, 3.14159, -0.0001]; - let blob = f32s_to_blob(&v); - assert_eq!(blob.len(), v.len() * 4); - let back = blob_to_f32s(&blob); - assert_eq!(back, v); - } - - #[test] - fn f32s_blob_roundtrip_empty() { - let v: Vec = vec![]; - let blob = f32s_to_blob(&v); - assert!(blob.is_empty()); - assert!(blob_to_f32s(&blob).is_empty()); - } - - #[test] - fn blob_to_f32s_drops_trailing_partial_bytes() { - // 1 个完整 f32 + 3 残字节 → chunks_exact 截断尾部 - let blob = f32s_to_blob(&[42.0]); - let with_garbage: Vec = blob.into_iter().chain([0xff, 0xff, 0xff]).collect(); - assert_eq!(blob_to_f32s(&with_garbage), vec![42.0]); - } - - #[test] - fn cosine_identical_vectors_near_one() { - let a = vec![1.0, 2.0, 3.0]; - let sim = cosine_similarity(&a, &a); - assert!((sim - 1.0).abs() < 1e-5, "identical ≈ 1.0, got {}", sim); - } - - #[test] - fn cosine_orthogonal_vectors_near_zero() { - let a = vec![1.0, 0.0]; - let b = vec![0.0, 1.0]; - let sim = cosine_similarity(&a, &b); - assert!(sim.abs() < 1e-5, "orthogonal ≈ 0.0, got {}", sim); - } - - #[test] - fn cosine_opposite_vectors_near_neg_one() { - let a = vec![1.0, 2.0, 3.0]; - let b = vec![-1.0, -2.0, -3.0]; - let sim = cosine_similarity(&a, &b); - assert!((sim + 1.0).abs() < 1e-5, "opposite ≈ -1.0, got {}", sim); - } - - #[test] - fn cosine_dimension_mismatch_zips_to_shortest() { - // 实现用 zip(a,b),维度不匹配按较短的那个对齐,不 panic - let a = vec![1.0, 0.0, 0.0]; // 3 维 - let b = vec![1.0, 0.0]; // 2 维 → zip 取前 2 个 - let sim = cosine_similarity(&a, &b); - assert!((sim - 1.0).abs() < 1e-5, "zip 截断后应 ≈ 1.0, got {}", sim); - } - - #[test] - fn cosine_zero_vector_does_not_panic() { - // 分母有 +1e-8 保护,零向量返回有限值而非 NaN/inf - let z = vec![0.0, 0.0, 0.0]; - let sim = cosine_similarity(&z, &z); - assert!(sim.is_finite(), "零向量相似度应有限, got {}", sim); - } - - // ---------- KnowledgeRepo 内存 DB ---------- - - /// 构造一条 KnowledgeRecord fixture - fn krec( - id: &str, - title: &str, - content: &str, - kind: &str, - status: &str, - confidence: Option<&str>, - reuse: i32, - ) -> KnowledgeRecord { - KnowledgeRecord { - id: id.to_string(), - kind: kind.to_string(), - title: title.to_string(), - content: content.to_string(), - tags: Some("[]".to_string()), - status: status.to_string(), - confidence: confidence.map(|s| s.to_string()), - reuse_count: reuse, - verified: false, - source_project: Some("proj-1".to_string()), - source_ref: Some("conv:c1".to_string()), - reasoning: None, - created_at: "1700000000000".to_string(), - updated_at: "1700000000000".to_string(), - } - } - - async fn setup_repo() -> KnowledgeRepo { - let db = Database::open_in_memory().await.expect("open_in_memory"); - KnowledgeRepo::new(&db) - } - - #[tokio::test] - async fn search_matches_title_and_filters_non_published() { - let repo = setup_repo().await; - // 一条命中(title 含关键词)、一条 published 不命中、一条 status 非 published 但命中 - repo.insert(krec("k1", "Rust 异步并发模型", "tokio 运行时", "lesson", "published", Some("high"), 5)) - .await - .unwrap(); - repo.insert(krec("k2", "无关标题", "无关内容", "lesson", "published", None, 0)) - .await - .unwrap(); - repo.insert(krec("k3", "Rust 异步进阶", "...", "lesson", "candidate", Some("low"), 9)) - .await - .unwrap(); - - let hits = repo.search("异步", None, 10).await.unwrap(); - // 只有 k1 是 published 且命中;k3 命中但非 published 被过滤 - assert_eq!(hits.len(), 1); - assert_eq!(hits[0].id, "k1"); - } - - #[tokio::test] - async fn search_matches_content_branch() { - let repo = setup_repo().await; - // title 不含关键词,content 含 → 命中 content LIKE 分支 - repo.insert(krec("k1", "标题", "深入理解 Rust 所有权与借用", "lesson", "published", None, 1)) - .await - .unwrap(); - - let hits = repo.search("所有权", None, 10).await.unwrap(); - assert_eq!(hits.len(), 1); - assert_eq!(hits[0].id, "k1"); - } - - #[tokio::test] - async fn search_no_keyword_match_returns_empty() { - let repo = setup_repo().await; - repo.insert(krec("k1", "Rust", "tokio", "lesson", "published", None, 1)) - .await - .unwrap(); - - let hits = repo.search("不存在的关键词xyz", None, 10).await.unwrap(); - assert!(hits.is_empty()); - } - - #[tokio::test] - async fn search_kind_filter_narrows_results() { - let repo = setup_repo().await; - repo.insert(krec("k1", "Rust 规范", "...", "lesson", "published", None, 1)) - .await - .unwrap(); - repo.insert(krec("k2", "Rust 决策", "...", "decision", "published", None, 1)) - .await - .unwrap(); - - // 不过滤 kind → 两条都命中 - let all = repo.search("Rust", None, 10).await.unwrap(); - assert_eq!(all.len(), 2); - // 只取 lesson → 仅 k1 - let only_lesson = repo.search("Rust", Some("lesson"), 10).await.unwrap(); - assert_eq!(only_lesson.len(), 1); - assert_eq!(only_lesson[0].id, "k1"); - } - - #[tokio::test] - async fn search_orders_by_reuse_count_desc() { - let repo = setup_repo().await; - repo.insert(krec("low", "Rust A", "...", "lesson", "published", None, 1)) - .await - .unwrap(); - repo.insert(krec("high", "Rust B", "...", "lesson", "published", None, 50)) - .await - .unwrap(); - repo.insert(krec("mid", "Rust C", "...", "lesson", "published", None, 10)) - .await - .unwrap(); - - let hits = repo.search("Rust", None, 10).await.unwrap(); - let ids: Vec<_> = hits.iter().map(|h| h.id.as_str()).collect(); - assert_eq!(ids, vec!["high", "mid", "low"]); - } - - #[tokio::test] - async fn search_truncates_to_limit() { - let repo = setup_repo().await; - for i in 0..5 { - repo.insert(krec(&format!("k{i}"), &format!("Rust-{i}"), "...", "lesson", "published", None, i)) - .await - .unwrap(); - } - let hits = repo.search("Rust", None, 2).await.unwrap(); - assert_eq!(hits.len(), 2); - // reuse_count 最高的两条(k4=4, k3=3) - assert_eq!(hits[0].id, "k4"); - assert_eq!(hits[1].id, "k3"); - } - - #[tokio::test] - async fn list_by_status_orders_by_confidence_semantics() { - // confidence 字典序 high>low>medium 非预期,实现用 CASE 强制 high>medium>low - let repo = setup_repo().await; - repo.insert(krec("low", "t", "c", "lesson", "pending_review", Some("low"), 0)) - .await - .unwrap(); - repo.insert(krec("high", "t", "c", "lesson", "pending_review", Some("high"), 0)) - .await - .unwrap(); - repo.insert(krec("medium", "t", "c", "lesson", "pending_review", Some("medium"), 0)) - .await - .unwrap(); - // created_at 相同,纯靠 confidence 排序 - - let list = repo.list_by_status("pending_review").await.unwrap(); - let confidences: Vec<_> = list.iter().map(|r| r.confidence.as_deref().unwrap_or("")).collect(); - assert_eq!(confidences, vec!["high", "medium", "low"]); - } - - #[tokio::test] - async fn list_by_status_filters_other_status() { - let repo = setup_repo().await; - repo.insert(krec("a", "t", "c", "lesson", "pending_review", Some("high"), 0)) - .await - .unwrap(); - repo.insert(krec("b", "t", "c", "lesson", "published", Some("high"), 0)) - .await - .unwrap(); - - let list = repo.list_by_status("pending_review").await.unwrap(); - assert_eq!(list.len(), 1); - assert_eq!(list[0].id, "a"); - } - - #[tokio::test] - async fn increment_reuse_count_is_atomic_plus_one() { - let repo = setup_repo().await; - repo.insert(krec("k1", "Rust", "...", "lesson", "published", None, 0)) - .await - .unwrap(); - - // 连续 +1 两次 - assert!(repo.increment_reuse_count("k1").await.unwrap()); - assert!(repo.increment_reuse_count("k1").await.unwrap()); - - let rec = repo.get_by_id("k1").await.unwrap().expect("记录存在"); - assert_eq!(rec.reuse_count, 2); - - // 不存在的 id → false - assert!(!repo.increment_reuse_count("nope").await.unwrap()); - } - - #[tokio::test] - async fn search_vector_orders_by_cosine_and_truncates() { - let repo = setup_repo().await; - // 三条 published 记录,embedding 维度均为 2 - repo.insert(krec("exact", "e", "c", "lesson", "published", None, 0)) - .await - .unwrap(); - repo.insert(krec("orth", "e", "c", "lesson", "published", None, 0)) - .await - .unwrap(); - repo.insert(krec("neg", "e", "c", "lesson", "published", None, 0)) - .await - .unwrap(); - // 一条非 published(应被过滤) - repo.insert(krec("hidden", "e", "c", "lesson", "candidate", None, 0)) - .await - .unwrap(); - - repo.set_embedding("exact", &[1.0, 0.0]).await.unwrap(); // 与 query 完全同向 - repo.set_embedding("orth", &[0.0, 1.0]).await.unwrap(); // 正交 ≈ 0 - repo.set_embedding("neg", &[-1.0, 0.0]).await.unwrap(); // 反向 ≈ -1 - repo.set_embedding("hidden", &[1.0, 0.0]).await.unwrap(); // 非 published 过滤掉 - - let results = repo.search_vector(&[1.0, 0.0], 10).await.unwrap(); - // hidden 被 status 过滤,剩 3 条 - assert_eq!(results.len(), 3); - // 按相似度降序: exact(≈1) > orth(≈0) > neg(≈-1) - assert_eq!(results[0].0.id, "exact"); - assert_eq!(results[1].0.id, "orth"); - assert_eq!(results[2].0.id, "neg"); - assert!((results[0].1 - 1.0).abs() < 1e-5); - assert!(results[1].1.abs() < 1e-5); - assert!((results[2].1 + 1.0).abs() < 1e-5); - } - - #[tokio::test] - async fn search_vector_skips_dimension_mismatch() { - let repo = setup_repo().await; - repo.insert(krec("dim2", "e", "c", "lesson", "published", None, 0)) - .await - .unwrap(); - repo.insert(krec("dim3", "e", "c", "lesson", "published", None, 0)) - .await - .unwrap(); - - repo.set_embedding("dim2", &[1.0, 0.0]).await.unwrap(); - repo.set_embedding("dim3", &[1.0, 0.0, 0.0]).await.unwrap(); // 维度不匹配 - - // query 是 2 维,dim3 被跳过 - let results = repo.search_vector(&[1.0, 0.0], 10).await.unwrap(); - assert_eq!(results.len(), 1); - assert_eq!(results[0].0.id, "dim2"); - } - - #[tokio::test] - async fn search_vector_truncates_to_limit() { - let repo = setup_repo().await; - for i in 0..4 { - repo.insert(krec(&format!("k{i}"), "e", "c", "lesson", "published", None, 0)) - .await - .unwrap(); - // 全部与 query 同向,相似度相同 → 仅验证 truncate 生效 - repo.set_embedding(&format!("k{i}"), &[1.0, 0.0]).await.unwrap(); - } - let results = repo.search_vector(&[1.0, 0.0], 2).await.unwrap(); - assert_eq!(results.len(), 2); - } - - #[tokio::test] - async fn search_vector_ignores_records_without_embedding() { - let repo = setup_repo().await; - // published 但未写 embedding - repo.insert(krec("noemb", "e", "c", "lesson", "published", None, 0)) - .await - .unwrap(); - let results = repo.search_vector(&[1.0, 0.0], 10).await.unwrap(); - assert!(results.is_empty()); - } -} diff --git a/crates/df-storage/src/crud/conversation_repo.rs b/crates/df-storage/src/crud/conversation_repo.rs new file mode 100644 index 0000000..27b5b52 --- /dev/null +++ b/crates/df-storage/src/crud/conversation_repo.rs @@ -0,0 +1,453 @@ +//! AI 对话域 Repo:AiProviderRepo / AiConversationRepo / AiToolExecutionRepo + +use std::sync::Arc; + +use rusqlite::{params, Connection, OptionalExtension, Row}; +use tokio::sync::Mutex; + +use df_types::error::Result; + +use crate::db::Database; +use crate::models::{AiConversationRecord, AiProviderRecord, AiToolExecutionRecord}; + +use super::impl_repo; +use super::{now_millis_str, storage_err, validate_column_name}; + +// ============================================================ +// from_row 辅助函数 +// ============================================================ + +fn ai_provider_from_row(row: &Row<'_>) -> std::result::Result { + // model_configs:DB TEXT 列存 JSON 字符串。读 Option 兼容老库 NULL, + // 再经 deserialize_model_configs 解析(老字符串数组/新对象数组/空 → Vec)。 + // 解析失败不致命:降级为空 Vec(防单行坏数据拖垮 list_all)。 + let model_configs: Vec = { + let raw: Option = row.get("model_configs").ok(); + match raw { + None => Vec::new(), + Some(s) => serde_json::from_str::(&format!( + r#"{{"v":{}}}"#, + if s.trim().is_empty() { + "null".to_string() + } else if s.trim_start().starts_with('[') || s.trim_start().starts_with('{') { + s + } else { + // 非 JSON 字面文本(理论不会出现)→ 包装为 JSON 字符串让 deserialize 兜底 + serde_json::to_string(&s).unwrap_or_else(|_| "null".into()) + } + )) + .map(|w| w.v) + .unwrap_or_default(), + } + }; + Ok(AiProviderRecord { + id: row.get("id")?, + name: row.get("name")?, + provider_type: row.get("provider_type")?, + api_key: row.get("api_key")?, + base_url: row.get("base_url")?, + default_model: row.get("default_model")?, + models: row.get("models")?, + model_configs, + is_default: row.get::<_, i32>("is_default")? != 0, + config: row.get("config")?, + created_at: row.get("created_at")?, + updated_at: row.get("updated_at")?, + // F-260614-04: enabled/weight 列老库经 v19 迁移补建,DEFAULT 1 / DEFAULT 50。 + // from_row 按 i32 取列值兼容(SQLite 无真 BOOLEAN),0→false/非0→true。 + enabled: row.get::<_, i32>("enabled").unwrap_or(1) != 0, + // weight 读侧 clamp [0,100]:与 insert/update_full 落库的 `.min(100)` 对齐, + // 防老库(clamp 落地前写入的)或外部直改 DB 产生的越界值污染路由权重语义。 + weight: row.get::<_, i32>("weight") + .unwrap_or(50) + .clamp(0, 100) as u32, + }) +} + +/// from_row 内部辅助:复用 deserialize_model_configs 解析 DB TEXT 列 JSON。 +/// 包一层 { "v": <原始值> } 把任意 JSON 值送进 deserialize_model_configs。 +#[derive(serde::Deserialize)] +struct ModelConfigsWrap { + #[serde(default, deserialize_with = "df_ai_core::model::deserialize_model_configs")] + v: Vec, +} + +fn ai_conversation_from_row(row: &Row<'_>) -> std::result::Result { + Ok(AiConversationRecord { + id: row.get("id")?, + title: row.get("title")?, + messages: row.get("messages")?, + provider_id: row.get("provider_id")?, + model: row.get("model")?, + models: row.get("models")?, + archived: row.get::<_, i32>("archived")? != 0, + pinned: row.get::<_, i32>("pinned")? != 0, + prompt_tokens: row.get("prompt_tokens")?, + completion_tokens: row.get("completion_tokens")?, + created_at: row.get("created_at")?, + updated_at: row.get("updated_at")?, + }) +} + +fn ai_tool_execution_from_row(row: &Row<'_>) -> std::result::Result { + Ok(AiToolExecutionRecord { + id: row.get("id")?, + conversation_id: row.get("conversation_id")?, + tool_call_id: row.get("tool_call_id")?, + tool_name: row.get("tool_name")?, + arguments: row.get("arguments")?, + result: row.get("result")?, + status: row.get("status")?, + risk_level: row.get("risk_level")?, + requested_at: row.get("requested_at")?, + executed_at: row.get("executed_at")?, + decided_by: row.get("decided_by")?, + }) +} + +// ============================================================ +// Repo 实现 +// ============================================================ + +impl_repo!( + /// AI 提供商配置表 CRUD + AiProviderRepo, + AiProviderRecord, + "ai_providers", + from_row => |row| ai_provider_from_row(row), + insert => |conn, rec| { + let is_default = if rec.is_default { 1i32 } else { 0i32 }; + // model_configs:Vec → JSON 字符串落 TEXT 列 + let model_configs_json = serde_json::to_string(&rec.model_configs).unwrap_or_else(|_| "[]".into()); + // F-260614-04: enabled/weight 落库(SQLite 无 BOOLEAN,i32 承载)。 + let enabled_i = if rec.enabled { 1i32 } else { 0i32 }; + let weight_i = rec.weight.min(100) as i32; + conn.execute( + "INSERT OR REPLACE INTO ai_providers (id, name, provider_type, api_key, base_url, default_model, models, model_configs, is_default, config, created_at, updated_at, enabled, weight) + VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12, ?13, ?14)", + params![ + rec.id, rec.name, rec.provider_type, rec.api_key, rec.base_url, + rec.default_model, rec.models, model_configs_json, is_default, rec.config, rec.created_at, rec.updated_at, + enabled_i, weight_i + ], + ) + }, + update => |conn, rec| { + let is_default = if rec.is_default { 1i32 } else { 0i32 }; + let model_configs_json = serde_json::to_string(&rec.model_configs).unwrap_or_else(|_| "[]".into()); + let enabled_i = if rec.enabled { 1i32 } else { 0i32 }; + let weight_i = rec.weight.min(100) as i32; + conn.execute( + "UPDATE ai_providers SET name = ?1, provider_type = ?2, api_key = ?3, base_url = ?4, default_model = ?5, models = ?6, model_configs = ?7, is_default = ?8, config = ?9, updated_at = ?10, enabled = ?11, weight = ?12 WHERE id = ?13", + params![ + rec.name, rec.provider_type, rec.api_key, rec.base_url, + rec.default_model, rec.models, model_configs_json, is_default, rec.config, rec.updated_at, + enabled_i, weight_i, rec.id + ], + ) + } +); + +impl_repo!( + /// AI 对话历史表 CRUD + AiConversationRepo, + AiConversationRecord, + "ai_conversations", + from_row => |row| ai_conversation_from_row(row), + insert => |conn, rec| { + conn.execute( + "INSERT INTO ai_conversations (id, title, messages, provider_id, model, models, archived, pinned, prompt_tokens, completion_tokens, created_at, updated_at) + VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12)", + params![ + rec.id, rec.title, rec.messages, rec.provider_id, rec.model, rec.models, rec.archived, + if rec.pinned { 1i32 } else { 0i32 }, + rec.prompt_tokens, rec.completion_tokens, + rec.created_at, rec.updated_at + ], + ) + }, + update => |conn, rec| { + conn.execute( + "UPDATE ai_conversations SET title = ?1, messages = ?2, provider_id = ?3, model = ?4, models = ?5, archived = ?6, pinned = ?7, prompt_tokens = ?8, completion_tokens = ?9, updated_at = ?10 WHERE id = ?11", + params![ + rec.title, rec.messages, rec.provider_id, rec.model, rec.models, rec.archived, + if rec.pinned { 1i32 } else { 0i32 }, + rec.prompt_tokens, rec.completion_tokens, + rec.updated_at, rec.id + ], + ) + } +); + +impl_repo!( + /// AI 工具执行审计表 CRUD + AiToolExecutionRepo, + AiToolExecutionRecord, + "ai_tool_executions", + from_row => |row| ai_tool_execution_from_row(row), + insert => |conn, rec| { + conn.execute( + "INSERT INTO ai_tool_executions (id, conversation_id, tool_call_id, tool_name, arguments, result, status, risk_level, requested_at, executed_at, decided_by) + VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11)", + params![ + rec.id, rec.conversation_id, rec.tool_call_id, rec.tool_name, + rec.arguments, rec.result, rec.status, rec.risk_level, + rec.requested_at, rec.executed_at, rec.decided_by + ], + ) + }, + update => |conn, rec| { + conn.execute( + "UPDATE ai_tool_executions SET conversation_id = ?1, tool_call_id = ?2, tool_name = ?3, arguments = ?4, result = ?5, status = ?6, risk_level = ?7, requested_at = ?8, executed_at = ?9, decided_by = ?10 WHERE id = ?11", + params![ + rec.conversation_id, rec.tool_call_id, rec.tool_name, + rec.arguments, rec.result, rec.status, rec.risk_level, + rec.requested_at, rec.executed_at, rec.decided_by, rec.id + ], + ) + } +); + +// ai_tool_executions 无 created_at 列(用 requested_at/executed_at 计时), +// 通用 query 宏硬编码 ORDER BY created_at 会触发 "no such column" → 调用方 unwrap_or_default 吞错。 +// 故为此表提供专用查询,绕过通用 query。详见 ai.rs audit_finalize。 +impl AiToolExecutionRepo { + /// 按 tool_call_id 查最新一条审计记录(审批回填定位用)。 + pub async fn find_by_tool_call_id( + &self, + tool_call_id: &str, + ) -> Result> { + let conn = self.conn.clone(); + let tid = tool_call_id.to_owned(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare( + "SELECT * FROM ai_tool_executions WHERE tool_call_id = ?1 ORDER BY requested_at DESC LIMIT 1", + ) + .map_err(storage_err)?; + let row = stmt + .query_row(params![tid], |row| ai_tool_execution_from_row(row)) + .optional() + .map_err(storage_err)?; + Ok(row) + }) + .await + .map_err(storage_err)? + } + + /// 列出所有 status=pending 的审计行(启动重建 pending_approvals 用) + /// + /// 专用 SELECT(非 query 宏——后者硬编码 ORDER BY created_at,而本表无该列)。 + pub async fn list_pending(&self) -> Result> { + let conn = self.conn.clone(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare("SELECT * FROM ai_tool_executions WHERE status = 'pending' ORDER BY requested_at ASC") + .map_err(storage_err)?; + let rows = stmt + .query_map([], |row| ai_tool_execution_from_row(row)) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push(r.map_err(storage_err)?); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } + + /// 审批历史面板分页查询:按 requested_at 倒序(最新在前),limit 默认 50。 + /// + /// 与 list_pending 同理走专用 SELECT,绕过通用 query 宏(后者硬编码 + /// ORDER BY created_at,本表无该列)。limit/offset 上限钳制(limit ≤ 200), + /// 防前端恶意/失误传超大值。 + pub async fn list_recent( + &self, + limit: u32, + offset: u32, + ) -> Result> { + let conn = self.conn.clone(); + // 钳制 limit 防滥用(默认 50,最大 200) + let safe_limit = limit.min(200) as i64; + let safe_offset = offset as i64; + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare( + "SELECT * FROM ai_tool_executions ORDER BY requested_at DESC LIMIT ?1 OFFSET ?2", + ) + .map_err(storage_err)?; + let rows = stmt + .query_map(params![safe_limit, safe_offset], |row| { + ai_tool_execution_from_row(row) + }) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push(r.map_err(storage_err)?); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } +} + +// AiConversationRepo 的整体更新已由 impl_repo! 宏统一生成的 update_full 提供。 + +impl AiConversationRepo { + /// 清空对话消息内容(保留 conversation 记录本身,只清 messages JSON + 清零 token 计数) + /// + /// "清空对话"语义:对话壳保留(侧栏仍可见,可继续在该对话内聊),仅清空历史消息。 + /// messages 是 ai_conversations 表内的 JSON 列而非独立行,故"删 messages"= 置空该列。 + pub async fn clear_messages(&self, id: &str) -> Result { + let conn = self.conn.clone(); + let id = id.to_owned(); + let now = now_millis_str(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let affected = guard + .execute( + "UPDATE ai_conversations SET messages = '[]', prompt_tokens = 0, completion_tokens = 0, updated_at = ?1 WHERE id = ?2", + params![now, id], + ) + .map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } + + /// 设置归档标记(仅改 archived,不动 updated_at) + /// + /// 区别于 update_field(后者强制 SET updated_at=now,会把归档/取消归档误判为内容更新, + /// 导致侧栏相对时间跳变为"刚刚")。归档是纯元数据标记,应保持时间不变。 + pub async fn set_archived(&self, id: &str, archived: bool) -> Result { + let conn = self.conn.clone(); + let id = id.to_owned(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let affected = guard + .execute( + "UPDATE ai_conversations SET archived = ?1 WHERE id = ?2", + params![if archived { 1 } else { 0 }, id], + ) + .map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } + + /// 设置置顶标记(仅改 pinned,不动 updated_at) — UX-17 + /// + /// 同 set_archived:置顶是纯元数据标记,不应改变相对时间。前端排序读 pinned DESC, updated_at DESC。 + pub async fn set_pinned(&self, id: &str, pinned: bool) -> Result { + let conn = self.conn.clone(); + let id = id.to_owned(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let affected = guard + .execute( + "UPDATE ai_conversations SET pinned = ?1 WHERE id = ?2", + params![if pinned { 1 } else { 0 }, id], + ) + .map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } +} + +// ============================================================ +// 单元测试 — AiProviderRepo model_configs DB roundtrip + 老库兼容 +// ============================================================ + +#[cfg(test)] +mod tests { + use super::*; + use crate::db::Database; + use crate::models::AiProviderRecord; + use df_ai_core::model::{Capability, IntelligenceTier, Modality, ModelConfig}; + + /// model_configs DB roundtrip + 老库空兼容(F-01 阶段1) + #[tokio::test] + async fn ai_provider_model_configs_roundtrip_and_old_db_compat() { + let db = Database::open_in_memory().await.expect("open_in_memory"); + let repo = AiProviderRepo::new(&db); + + // 新格式:带多模型 + 多维度配置 + let configs = vec![ + ModelConfig::with_defaults("glm-4-flash"), + ModelConfig { + model_id: "glm-4v".into(), + modalities: vec![Modality::Text, Modality::Vision], + capabilities: vec![Capability::ToolUse], + intelligence: IntelligenceTier::Plus, + weight: 70, + context_window: 128_000, + ..ModelConfig::with_defaults("glm-4v") + }, + ]; + let rec = AiProviderRecord { + id: "p1".into(), + name: "测试".into(), + provider_type: "openai_compat".into(), + api_key: String::new(), + base_url: "https://x".into(), + default_model: "glm-4-flash".into(), + models: None, + model_configs: configs.clone(), + is_default: false, + config: None, + created_at: "0".into(), + updated_at: "0".into(), + enabled: true, + weight: 50, + }; + repo.insert(rec).await.expect("insert"); + + let got = repo.get_by_id("p1").await.expect("get").expect("row exists"); + assert_eq!(got.model_configs.len(), 2); + assert_eq!(got.model_configs[0].model_id, "glm-4-flash"); + assert_eq!(got.model_configs[1].model_id, "glm-4v"); + assert_eq!(got.model_configs[1].intelligence, IntelligenceTier::Plus); + assert_eq!(got.model_configs[1].context_window, 128_000); + + // 老库空兼容:直接写 model_configs=NULL 的行(模拟 V18 之前的老库行) + // 然后 from_row 应得空 Vec + { + let conn = db.conn(); + let g = conn.lock().await; + g.execute( + "INSERT OR REPLACE INTO ai_providers \ + (id,name,provider_type,api_key,base_url,default_model,models,model_configs,is_default,config,created_at,updated_at) \ + VALUES ('old','','openai_compat','','','','{}',NULL,0,NULL,'0','0')", + [], + ) + .expect("raw insert old row"); + } + let old = repo.get_by_id("old").await.expect("get").expect("old row"); + assert!(old.model_configs.is_empty(), "NULL 列应得空 Vec"); + + // 老格式字符串数组 JSON(向后兼容 deserialize_model_configs) + { + let conn = db.conn(); + let g = conn.lock().await; + g.execute( + "INSERT OR REPLACE INTO ai_providers \ + (id,name,provider_type,api_key,base_url,default_model,models,model_configs,is_default,config,created_at,updated_at) \ + VALUES ('legacy','','openai_compat','','','','{}','[\"glm-4-flash\",\"glm-4v\"]',0,NULL,'0','0')", + [], + ) + .expect("raw insert legacy row"); + } + let legacy = repo.get_by_id("legacy").await.expect("get").expect("legacy row"); + assert_eq!(legacy.model_configs.len(), 2, "老字符串数组应转 2 个默认 ModelConfig"); + assert_eq!(legacy.model_configs[0].model_id, "glm-4-flash"); + } +} diff --git a/crates/df-storage/src/crud/idea_repo.rs b/crates/df-storage/src/crud/idea_repo.rs new file mode 100644 index 0000000..e5fb6a1 --- /dev/null +++ b/crates/df-storage/src/crud/idea_repo.rs @@ -0,0 +1,871 @@ +//! 想法/知识域 Repo:IdeaRepo / KnowledgeRepo / KnowledgeEventsRepo + 向量工具(embedding BLOB 序列化 + 余弦相似度) + +use std::sync::Arc; + +use rusqlite::{params, Connection, OptionalExtension, Row}; +use tokio::sync::Mutex; + +use df_types::error::Result; + +use crate::db::Database; +use crate::models::{IdeaRecord, KnowledgeEventRecord, KnowledgeRecord}; + +use super::impl_repo; +use super::{now_millis_str, storage_err, validate_column_name}; + +// ============================================================ +// 知识库 SELECT 列清单(防 COLS 漂移) +// ============================================================ + +/// `knowledges` 表对应 `KnowledgeRecord` 14 个字段的列名(顺序与结构体一致)。 +/// +/// 多处 `search`/`search_vector` 内联 COLS 串的 DRY 收口(CR-260615-03):集中一处定义, +/// 配合下方 `KNOWLEDGE_COL_COUNT` 断言,任一处加列漏改会被测试 `test_knowledge_cols_matches_record` +/// 立即捕获(`knowledge_from_row` 按 name 取列,SELECT 漏列会运行时 rusqlite 报错,故提前断言)。 +const KNOWLEDGE_COLS: &str = "id,kind,title,content,tags,status,confidence,reuse_count,verified,source_project,source_ref,reasoning,created_at,updated_at"; +/// `KnowledgeRecord` 字段数(与上面列清单的逗号分隔项数一致,被测试断言)。 +/// 仅测试期消费(列漂移断言);保留为非 `cfg(test)` 以便测试外的阅读者一眼看到字段数。 +#[cfg_attr(not(test), allow(dead_code))] +const KNOWLEDGE_COL_COUNT: usize = 14; +/// `search_vector` 用的列清单:KNOWLEDGE_COLS + embedding(余弦计算用,不入 KnowledgeRecord)。 +const KNOWLEDGE_COLS_WITH_EMBEDDING: &str = concat!( + "id,kind,title,content,tags,status,confidence,reuse_count,verified,", + "source_project,source_ref,reasoning,created_at,updated_at,embedding" +); + +// ============================================================ +// 向量工具 — embedding BLOB 序列化 + 余弦相似度 +// ============================================================ + +/// Vec → 小端字节 BLOB +fn f32s_to_blob(v: &[f32]) -> Vec { + v.iter().flat_map(|f| f.to_le_bytes()).collect() +} + +/// BLOB → Vec(长度非 4 倍数时截断尾部残字节) +fn blob_to_f32s(blob: &[u8]) -> Vec { + blob.chunks_exact(4) + .map(|c| f32::from_le_bytes([c[0], c[1], c[2], c[3]])) + .collect() +} + +/// 余弦相似度(确定性数学,与任何实现结果一致) +fn cosine_similarity(a: &[f32], b: &[f32]) -> f32 { + let dot: f32 = a.iter().zip(b).map(|(x, y)| x * y).sum(); + let norm_a: f32 = a.iter().map(|x| x * x).sum::().sqrt(); + let norm_b: f32 = b.iter().map(|x| x * x).sum::().sqrt(); + dot / (norm_a * norm_b + 1e-8) +} + +// ============================================================ +// from_row 辅助函数 +// ============================================================ + +fn idea_from_row(row: &Row<'_>) -> std::result::Result { + Ok(IdeaRecord { + id: row.get("id")?, + title: row.get("title")?, + description: row.get("description")?, + status: row.get("status")?, + priority: row.get("priority")?, + score: row.get("score")?, + tags: row.get("tags")?, + source: row.get("source")?, + promoted_to: row.get("promoted_to")?, + ai_analysis: row.get("ai_analysis")?, + scores: row.get("scores")?, + created_at: row.get("created_at")?, + updated_at: row.get("updated_at")?, + }) +} + +fn knowledge_from_row(row: &Row<'_>) -> std::result::Result { + Ok(KnowledgeRecord { + id: row.get("id")?, + kind: row.get("kind")?, + title: row.get("title")?, + content: row.get("content")?, + tags: row.get("tags")?, + status: row.get("status")?, + confidence: row.get("confidence")?, + reuse_count: row.get("reuse_count")?, + verified: row.get::<_, i32>("verified")? != 0, + source_project: row.get("source_project")?, + source_ref: row.get("source_ref")?, + reasoning: row.get("reasoning")?, + created_at: row.get("created_at")?, + updated_at: row.get("updated_at")?, + }) +} + +fn knowledge_event_from_row(row: &Row<'_>) -> std::result::Result { + Ok(KnowledgeEventRecord { + id: row.get("id")?, + knowledge_id: row.get("knowledge_id")?, + event_type: row.get("event_type")?, + source_ref: row.get("source_ref")?, + context_json: row.get("context_json")?, + timestamp: row.get("timestamp")?, + }) +} + +// ============================================================ +// Repo 实现 +// ============================================================ + +impl_repo!( + /// 想法表 CRUD + IdeaRepo, + IdeaRecord, + "ideas", + from_row => |row| idea_from_row(row), + insert => |conn, rec| { + conn.execute( + "INSERT INTO ideas (id, title, description, status, priority, score, tags, source, promoted_to, ai_analysis, scores, created_at, updated_at) + VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12, ?13)", + params![ + rec.id, rec.title, rec.description, rec.status, rec.priority, + rec.score, rec.tags, rec.source, rec.promoted_to, rec.ai_analysis, + rec.scores, rec.created_at, rec.updated_at + ], + ) + }, + update => |conn, rec| { + conn.execute( + "UPDATE ideas SET title = ?1, description = ?2, status = ?3, priority = ?4, score = ?5, tags = ?6, source = ?7, promoted_to = ?8, ai_analysis = ?9, scores = ?10, updated_at = ?11 WHERE id = ?12", + params![ + rec.title, rec.description, rec.status, rec.priority, + rec.score, rec.tags, rec.source, rec.promoted_to, rec.ai_analysis, + rec.scores, rec.updated_at, rec.id + ], + ) + } +); + +impl_repo!( + /// 知识库表 CRUD + KnowledgeRepo, + KnowledgeRecord, + "knowledges", + from_row => |row| knowledge_from_row(row), + insert => |conn, rec| { + let verified = if rec.verified { 1i32 } else { 0i32 }; + conn.execute( + "INSERT INTO knowledges (id, kind, title, content, tags, status, confidence, reuse_count, verified, source_project, source_ref, reasoning, created_at, updated_at) + VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12, ?13, ?14)", + params![ + rec.id, rec.kind, rec.title, rec.content, rec.tags, rec.status, rec.confidence, + rec.reuse_count, verified, rec.source_project, rec.source_ref, rec.reasoning, + rec.created_at, rec.updated_at + ], + ) + }, + update => |conn, rec| { + let verified = if rec.verified { 1i32 } else { 0i32 }; + conn.execute( + "UPDATE knowledges SET kind = ?1, title = ?2, content = ?3, tags = ?4, status = ?5, confidence = ?6, reuse_count = ?7, verified = ?8, source_project = ?9, source_ref = ?10, reasoning = ?11, updated_at = ?12 WHERE id = ?13", + params![ + rec.kind, rec.title, rec.content, rec.tags, rec.status, rec.confidence, + rec.reuse_count, verified, rec.source_project, rec.source_ref, rec.reasoning, + rec.updated_at, rec.id + ], + ) + } +); + +// KnowledgeRepo 的整体更新已由 impl_repo! 宏统一生成的 update_full 提供。 + +impl KnowledgeRepo { + /// 检索知识: title/content LIKE 匹配,可选 kind 过滤,按 reuse_count 降序,top-N + /// + /// 克制检索: top-N≤3(由调用方 limit 控制),精确匹配优先(语义模糊后做)。 + pub async fn search(&self, query: &str, kind: Option<&str>, limit: usize) -> Result> { + let conn = self.conn.clone(); + let pattern = format!("%{}%", query); + let kind = kind.map(|s| s.to_owned()); + let limit_i = limit as i64; + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut results = Vec::new(); + if let Some(k) = &kind { + let mut stmt = guard + .prepare(&format!("SELECT {KNOWLEDGE_COLS} FROM knowledges WHERE status = 'published' AND (title LIKE ?1 OR content LIKE ?2) AND kind = ?3 ORDER BY reuse_count DESC LIMIT ?4")) + .map_err(storage_err)?; + let rows = stmt + .query_map(params![pattern, pattern, k, limit_i], |row| knowledge_from_row(row)) + .map_err(storage_err)?; + for r in rows { + results.push(r.map_err(storage_err)?); + } + } else { + let mut stmt = guard + .prepare(&format!("SELECT {KNOWLEDGE_COLS} FROM knowledges WHERE status = 'published' AND (title LIKE ?1 OR content LIKE ?2) ORDER BY reuse_count DESC LIMIT ?3")) + .map_err(storage_err)?; + let rows = stmt + .query_map(params![pattern, pattern, limit_i], |row| knowledge_from_row(row)) + .map_err(storage_err)?; + for r in rows { + results.push(r.map_err(storage_err)?); + } + } + Ok(results) + }) + .await + .map_err(storage_err)? + } + + /// 按状态列出(审核收件箱用): 按 confidence 语义排序(high>medium>low),次按 created_at + /// + /// 用 CASE WHEN 替代纯 TEXT 排序(字典序 high>low>medium 非预期语义)。 + pub async fn list_by_status(&self, status: &str) -> Result> { + let conn = self.conn.clone(); + let status = status.to_owned(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare( + "SELECT id,kind,title,content,tags,status,confidence,reuse_count,verified,\ + source_project,source_ref,reasoning,created_at,updated_at \ + FROM knowledges WHERE status = ?1 + ORDER BY CASE confidence + WHEN 'high' THEN 3 + WHEN 'medium' THEN 2 + WHEN 'low' THEN 1 + ELSE 0 + END DESC, created_at DESC", + ) + .map_err(storage_err)?; + let rows = stmt + .query_map(params![status], |row| knowledge_from_row(row)) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push(r.map_err(storage_err)?); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } + + /// 复用计数 +1(SQL 行级原子操作,并发安全) + pub async fn increment_reuse_count(&self, id: &str) -> Result { + let conn = self.conn.clone(); + let id = id.to_owned(); + let now = now_millis_str(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let affected = guard + .execute( + "UPDATE knowledges SET reuse_count = reuse_count + 1, updated_at = ?1 WHERE id = ?2", + params![now, id], + ) + .map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } + + /// 写入向量嵌入(BLOB = Vec 小端字节序列化) + /// + /// embedding 列不进 KnowledgeRecord(IPC 不需要传向量给前端),专用方法读写。 + pub async fn set_embedding(&self, id: &str, embedding: &[f32]) -> Result { + let conn = self.conn.clone(); + let id = id.to_owned(); + let blob = f32s_to_blob(embedding); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let affected = guard + .execute( + "UPDATE knowledges SET embedding = ?1 WHERE id = ?2", + params![blob, id], + ) + .map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } + + /// 向量检索: 加载全部 published 且有 embedding 的记录,纯 Rust 余弦相似度取 top-N + /// + /// 返回 (记录, 相似度分数)。数据量 <50k 时暴力遍历 <50ms,够用; + /// 更大规模再升 sqlite-vec HNSW(结果不变,只提速)。 + /// + /// 列限定: 显式列出所需列(与 search/list_by_status/top_used 一致),避免 SELECT * + /// 拉到未知新增列;embedding 单独取(不入 KnowledgeRecord)。 + /// + /// TODO(性能,低优先): 调用方(hybrid_search→merge_hybrid_results→build_knowledge_context) + /// 实际只消费 id/kind/title/content/reuse_count;reasoning(AI 生成大文本)、tags、 + /// source_project/source_ref 等元字段未被使用却仍随每行读出。真正省 IO 需返回精简结构 + /// (如 KnowledgeVectorHit { id, kind, title, content, reuse_count })替换返回类型, + /// 但这会改变 search_vector 签名与 merge_hybrid_results 调用契约——当前保守不动, + /// 待向量检索量级或 reasoning 文本体积成为瓶颈再单独立项。SELECT 列化本身不省字段, + /// 仅消除 SELECT * 的隐式依赖与未知列风险。 + pub async fn search_vector(&self, query_vec: &[f32], limit: usize) -> Result> { + let conn = self.conn.clone(); + let query_vec = query_vec.to_vec(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + // 显式列: 14 个 KnowledgeRecord 字段 + embedding(余弦计算用,不入 KnowledgeRecord) + let mut stmt = guard + .prepare(&format!( + "SELECT {KNOWLEDGE_COLS_WITH_EMBEDDING} FROM knowledges WHERE status = 'published' AND embedding IS NOT NULL" + )) + .map_err(storage_err)?; + let rows = stmt + .query_map([], |row| { + let rec = knowledge_from_row(row)?; + let blob: Vec = row.get("embedding")?; + Ok((rec, blob)) + }) + .map_err(storage_err)?; + + let mut scored: Vec<(KnowledgeRecord, f32)> = Vec::new(); + for r in rows { + let (rec, blob) = r.map_err(storage_err)?; + let emb = blob_to_f32s(&blob); + // 维度不匹配(换过 embedding 模型的旧向量)跳过 + if emb.len() != query_vec.len() { + continue; + } + let score = cosine_similarity(&query_vec, &emb); + scored.push((rec, score)); + } + scored.sort_by(|a, b| b.1.partial_cmp(&a.1).unwrap_or(std::cmp::Ordering::Equal)); + scored.truncate(limit); + Ok(scored) + }) + .await + .map_err(storage_err)? + } + + /// 列出非归档知识(全部 status != 'archived'),按 confidence 语义排序 + pub async fn list_non_archived(&self) -> Result> { + let conn = self.conn.clone(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare( + "SELECT id,kind,title,content,tags,status,confidence,reuse_count,verified,\ + source_project,source_ref,reasoning,created_at,updated_at \ + FROM knowledges WHERE status != 'archived' + ORDER BY CASE confidence + WHEN 'high' THEN 3 + WHEN 'medium' THEN 2 + WHEN 'low' THEN 1 + ELSE 0 + END DESC, created_at DESC", + ) + .map_err(storage_err)?; + let rows = stmt + .query_map([], |row| knowledge_from_row(row)) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push(r.map_err(storage_err)?); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } + + /// 热门知识(已发布,按复用次数降序) + pub async fn top_used(&self, limit: usize) -> Result> { + let conn = self.conn.clone(); + let limit_i = limit as i64; + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare("SELECT id,kind,title,content,tags,status,confidence,reuse_count,verified,source_project,source_ref,reasoning,created_at,updated_at FROM knowledges WHERE status = 'published' ORDER BY reuse_count DESC LIMIT ?1") + .map_err(storage_err)?; + let rows = stmt + .query_map(params![limit_i], |row| knowledge_from_row(row)) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push(r.map_err(storage_err)?); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } +} + +impl_repo!( + /// 知识生命线事件表 CRUD(追加型审计表) + KnowledgeEventsRepo, + KnowledgeEventRecord, + "knowledge_events", + from_row => |row| knowledge_event_from_row(row), + insert => |conn, rec| { + conn.execute( + "INSERT INTO knowledge_events (id, knowledge_id, event_type, source_ref, context_json, timestamp) + VALUES (?1, ?2, ?3, ?4, ?5, ?6)", + params![rec.id, rec.knowledge_id, rec.event_type, rec.source_ref, rec.context_json, rec.timestamp], + ) + }, + update => |conn, rec| { + conn.execute( + "UPDATE knowledge_events SET knowledge_id = ?1, event_type = ?2, source_ref = ?3, context_json = ?4, timestamp = ?5 WHERE id = ?6", + params![rec.knowledge_id, rec.event_type, rec.source_ref, rec.context_json, rec.timestamp, rec.id], + ) + } +); + +impl KnowledgeEventsRepo { + /// 按知识 ID 查询全部事件(时间正序,构建生命线视图用) + pub async fn list_by_knowledge(&self, knowledge_id: &str) -> Result> { + let conn = self.conn.clone(); + let knowledge_id = knowledge_id.to_owned(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare("SELECT id,knowledge_id,event_type,source_ref,context_json,timestamp FROM knowledge_events WHERE knowledge_id = ?1 ORDER BY timestamp ASC") + .map_err(storage_err)?; + let rows = stmt + .query_map(params![knowledge_id], |row| knowledge_event_from_row(row)) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push(r.map_err(storage_err)?); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } + + /// 按知识 ID + event_type 查询最近 N 条(如引用记录翻页: type=referenced) + pub async fn list_by_knowledge_type( + &self, + knowledge_id: &str, + event_type: &str, + limit: usize, + ) -> Result> { + let conn = self.conn.clone(); + let knowledge_id = knowledge_id.to_owned(); + let event_type = event_type.to_owned(); + let limit_i = limit as i64; + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare("SELECT id,knowledge_id,event_type,source_ref,context_json,timestamp FROM knowledge_events WHERE knowledge_id = ?1 AND event_type = ?2 ORDER BY timestamp DESC LIMIT ?3") + .map_err(storage_err)?; + let rows = stmt + .query_map(params![knowledge_id, event_type, limit_i], |row| knowledge_event_from_row(row)) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push(r.map_err(storage_err)?); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } + + /// 跨知识列最近 N 条事件(全表 timestamp DESC,top-N)。 + /// + /// **专用兜底方法**:本表时间列名是 `timestamp` 而非 `created_at`,但 + /// `impl_repo!` 宏生成的 `query()` / `list_all()` 硬编码 `ORDER BY created_at` + /// (见宏内 `ORDER BY created_at DESC` 字面量),误调 `state.knowledge_events.query(...)` + /// 会触发 SQLite "no such column: created_at"。本方法走专用 SELECT 绕过宏硬编码, + /// 供需要跨知识按时间倒序浏览事件的调用方使用(对标 AiToolExecutionRepo::list_recent + /// 对 ai_tool_executions 表的同款兜底处理——那张表同样无 created_at 列)。 + /// limit 上限钳制 200,防前端恶意/失误传超大值。 + pub async fn list_recent(&self, limit: u32) -> Result> { + let conn = self.conn.clone(); + // 钳制 limit 防滥用(最大 200) + let safe_limit = limit.min(200) as i64; + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare( + "SELECT id,knowledge_id,event_type,source_ref,context_json,timestamp \ + FROM knowledge_events ORDER BY timestamp DESC LIMIT ?1", + ) + .map_err(storage_err)?; + let rows = stmt + .query_map(params![safe_limit], |row| knowledge_event_from_row(row)) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push(r.map_err(storage_err)?); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } +} + +// ============================================================ +// 单元测试 — 向量纯函数 + KnowledgeRepo 内存 DB + COLS 漂移防护 +// ============================================================ + +#[cfg(test)] +mod tests { + use super::*; + use crate::db::Database; + + // ---------- COLS 漂移防护(CR-260615-03) ---------- + + /// KNOWLEDGE_COLS 列数须等于 KNOWLEDGE_COL_COUNT(任一处漂移:加列漏改 / 串错位 → 立即失败)。 + /// `knowledge_from_row` 按 name 取列,SELECT 漏列会在运行时被 rusqlite 报错;此断言提前到测试期捕获。 + #[test] + fn test_knowledge_cols_matches_record() { + let count = KNOWLEDGE_COLS.split(',').count(); + assert_eq!( + count, KNOWLEDGE_COL_COUNT, + "KNOWLEDGE_COLS({count}列) ≠ KNOWLEDGE_COL_COUNT({KNOWLEDGE_COL_COUNT}); \ + 修改一处须同步另一处" + ); + // search_vector 多一列 embedding + let count_with_emb = KNOWLEDGE_COLS_WITH_EMBEDDING.split(',').count(); + assert_eq!( + count_with_emb, + KNOWLEDGE_COL_COUNT + 1, + "KNOWLEDGE_COLS_WITH_EMBEDDING({count_with_emb}列) ≠ KNOWLEDGE_COL_COUNT+1({}); \ + embedding 列应单独追加", + KNOWLEDGE_COL_COUNT + 1 + ); + // 每个列名须能被 split 出来(防末尾多逗号 / 空段) + for col in KNOWLEDGE_COLS.split(',') { + assert!(!col.is_empty(), "KNOWLEDGE_COLS 含空列名段"); + } + } + + // ---------- 向量纯函数 ---------- + + #[test] + fn f32s_blob_roundtrip_nonempty() { + let v = vec![0.0, 1.5, -2.25, 3.14159, -0.0001]; + let blob = f32s_to_blob(&v); + assert_eq!(blob.len(), v.len() * 4); + let back = blob_to_f32s(&blob); + assert_eq!(back, v); + } + + #[test] + fn f32s_blob_roundtrip_empty() { + let v: Vec = vec![]; + let blob = f32s_to_blob(&v); + assert!(blob.is_empty()); + assert!(blob_to_f32s(&blob).is_empty()); + } + + #[test] + fn blob_to_f32s_drops_trailing_partial_bytes() { + // 1 个完整 f32 + 3 残字节 → chunks_exact 截断尾部 + let blob = f32s_to_blob(&[42.0]); + let with_garbage: Vec = blob.into_iter().chain([0xff, 0xff, 0xff]).collect(); + assert_eq!(blob_to_f32s(&with_garbage), vec![42.0]); + } + + #[test] + fn cosine_identical_vectors_near_one() { + let a = vec![1.0, 2.0, 3.0]; + let sim = cosine_similarity(&a, &a); + assert!((sim - 1.0).abs() < 1e-5, "identical ≈ 1.0, got {}", sim); + } + + #[test] + fn cosine_orthogonal_vectors_near_zero() { + let a = vec![1.0, 0.0]; + let b = vec![0.0, 1.0]; + let sim = cosine_similarity(&a, &b); + assert!(sim.abs() < 1e-5, "orthogonal ≈ 0.0, got {}", sim); + } + + #[test] + fn cosine_opposite_vectors_near_neg_one() { + let a = vec![1.0, 2.0, 3.0]; + let b = vec![-1.0, -2.0, -3.0]; + let sim = cosine_similarity(&a, &b); + assert!((sim + 1.0).abs() < 1e-5, "opposite ≈ -1.0, got {}", sim); + } + + #[test] + fn cosine_dimension_mismatch_zips_to_shortest() { + // 实现用 zip(a,b),维度不匹配按较短的那个对齐,不 panic + let a = vec![1.0, 0.0, 0.0]; // 3 维 + let b = vec![1.0, 0.0]; // 2 维 → zip 取前 2 个 + let sim = cosine_similarity(&a, &b); + assert!((sim - 1.0).abs() < 1e-5, "zip 截断后应 ≈ 1.0, got {}", sim); + } + + #[test] + fn cosine_zero_vector_does_not_panic() { + // 分母有 +1e-8 保护,零向量返回有限值而非 NaN/inf + let z = vec![0.0, 0.0, 0.0]; + let sim = cosine_similarity(&z, &z); + assert!(sim.is_finite(), "零向量相似度应有限, got {}", sim); + } + + // ---------- KnowledgeRepo 内存 DB ---------- + + /// 构造一条 KnowledgeRecord fixture + fn krec( + id: &str, + title: &str, + content: &str, + kind: &str, + status: &str, + confidence: Option<&str>, + reuse: i32, + ) -> KnowledgeRecord { + KnowledgeRecord { + id: id.to_string(), + kind: kind.to_string(), + title: title.to_string(), + content: content.to_string(), + tags: Some("[]".to_string()), + status: status.to_string(), + confidence: confidence.map(|s| s.to_string()), + reuse_count: reuse, + verified: false, + source_project: Some("proj-1".to_string()), + source_ref: Some("conv:c1".to_string()), + reasoning: None, + created_at: "1700000000000".to_string(), + updated_at: "1700000000000".to_string(), + } + } + + async fn setup_repo() -> KnowledgeRepo { + let db = Database::open_in_memory().await.expect("open_in_memory"); + KnowledgeRepo::new(&db) + } + + #[tokio::test] + async fn search_matches_title_and_filters_non_published() { + let repo = setup_repo().await; + // 一条命中(title 含关键词)、一条 published 不命中、一条 status 非 published 但命中 + repo.insert(krec("k1", "Rust 异步并发模型", "tokio 运行时", "lesson", "published", Some("high"), 5)) + .await + .unwrap(); + repo.insert(krec("k2", "无关标题", "无关内容", "lesson", "published", None, 0)) + .await + .unwrap(); + repo.insert(krec("k3", "Rust 异步进阶", "...", "lesson", "candidate", Some("low"), 9)) + .await + .unwrap(); + + let hits = repo.search("异步", None, 10).await.unwrap(); + // 只有 k1 是 published 且命中;k3 命中但非 published 被过滤 + assert_eq!(hits.len(), 1); + assert_eq!(hits[0].id, "k1"); + } + + #[tokio::test] + async fn search_matches_content_branch() { + let repo = setup_repo().await; + // title 不含关键词,content 含 → 命中 content LIKE 分支 + repo.insert(krec("k1", "标题", "深入理解 Rust 所有权与借用", "lesson", "published", None, 1)) + .await + .unwrap(); + + let hits = repo.search("所有权", None, 10).await.unwrap(); + assert_eq!(hits.len(), 1); + assert_eq!(hits[0].id, "k1"); + } + + #[tokio::test] + async fn search_no_keyword_match_returns_empty() { + let repo = setup_repo().await; + repo.insert(krec("k1", "Rust", "tokio", "lesson", "published", None, 1)) + .await + .unwrap(); + + let hits = repo.search("不存在的关键词xyz", None, 10).await.unwrap(); + assert!(hits.is_empty()); + } + + #[tokio::test] + async fn search_kind_filter_narrows_results() { + let repo = setup_repo().await; + repo.insert(krec("k1", "Rust 规范", "...", "lesson", "published", None, 1)) + .await + .unwrap(); + repo.insert(krec("k2", "Rust 决策", "...", "decision", "published", None, 1)) + .await + .unwrap(); + + // 不过滤 kind → 两条都命中 + let all = repo.search("Rust", None, 10).await.unwrap(); + assert_eq!(all.len(), 2); + // 只取 lesson → 仅 k1 + let only_lesson = repo.search("Rust", Some("lesson"), 10).await.unwrap(); + assert_eq!(only_lesson.len(), 1); + assert_eq!(only_lesson[0].id, "k1"); + } + + #[tokio::test] + async fn search_orders_by_reuse_count_desc() { + let repo = setup_repo().await; + repo.insert(krec("low", "Rust A", "...", "lesson", "published", None, 1)) + .await + .unwrap(); + repo.insert(krec("high", "Rust B", "...", "lesson", "published", None, 50)) + .await + .unwrap(); + repo.insert(krec("mid", "Rust C", "...", "lesson", "published", None, 10)) + .await + .unwrap(); + + let hits = repo.search("Rust", None, 10).await.unwrap(); + let ids: Vec<_> = hits.iter().map(|h| h.id.as_str()).collect(); + assert_eq!(ids, vec!["high", "mid", "low"]); + } + + #[tokio::test] + async fn search_truncates_to_limit() { + let repo = setup_repo().await; + for i in 0..5 { + repo.insert(krec(&format!("k{i}"), &format!("Rust-{i}"), "...", "lesson", "published", None, i)) + .await + .unwrap(); + } + let hits = repo.search("Rust", None, 2).await.unwrap(); + assert_eq!(hits.len(), 2); + // reuse_count 最高的两条(k4=4, k3=3) + assert_eq!(hits[0].id, "k4"); + assert_eq!(hits[1].id, "k3"); + } + + #[tokio::test] + async fn list_by_status_orders_by_confidence_semantics() { + // confidence 字典序 high>low>medium 非预期,实现用 CASE 强制 high>medium>low + let repo = setup_repo().await; + repo.insert(krec("low", "t", "c", "lesson", "pending_review", Some("low"), 0)) + .await + .unwrap(); + repo.insert(krec("high", "t", "c", "lesson", "pending_review", Some("high"), 0)) + .await + .unwrap(); + repo.insert(krec("medium", "t", "c", "lesson", "pending_review", Some("medium"), 0)) + .await + .unwrap(); + // created_at 相同,纯靠 confidence 排序 + + let list = repo.list_by_status("pending_review").await.unwrap(); + let confidences: Vec<_> = list.iter().map(|r| r.confidence.as_deref().unwrap_or("")).collect(); + assert_eq!(confidences, vec!["high", "medium", "low"]); + } + + #[tokio::test] + async fn list_by_status_filters_other_status() { + let repo = setup_repo().await; + repo.insert(krec("a", "t", "c", "lesson", "pending_review", Some("high"), 0)) + .await + .unwrap(); + repo.insert(krec("b", "t", "c", "lesson", "published", Some("high"), 0)) + .await + .unwrap(); + + let list = repo.list_by_status("pending_review").await.unwrap(); + assert_eq!(list.len(), 1); + assert_eq!(list[0].id, "a"); + } + + #[tokio::test] + async fn increment_reuse_count_is_atomic_plus_one() { + let repo = setup_repo().await; + repo.insert(krec("k1", "Rust", "...", "lesson", "published", None, 0)) + .await + .unwrap(); + + // 连续 +1 两次 + assert!(repo.increment_reuse_count("k1").await.unwrap()); + assert!(repo.increment_reuse_count("k1").await.unwrap()); + + let rec = repo.get_by_id("k1").await.unwrap().expect("记录存在"); + assert_eq!(rec.reuse_count, 2); + + // 不存在的 id → false + assert!(!repo.increment_reuse_count("nope").await.unwrap()); + } + + #[tokio::test] + async fn search_vector_orders_by_cosine_and_truncates() { + let repo = setup_repo().await; + // 三条 published 记录,embedding 维度均为 2 + repo.insert(krec("exact", "e", "c", "lesson", "published", None, 0)) + .await + .unwrap(); + repo.insert(krec("orth", "e", "c", "lesson", "published", None, 0)) + .await + .unwrap(); + repo.insert(krec("neg", "e", "c", "lesson", "published", None, 0)) + .await + .unwrap(); + // 一条非 published(应被过滤) + repo.insert(krec("hidden", "e", "c", "lesson", "candidate", None, 0)) + .await + .unwrap(); + + repo.set_embedding("exact", &[1.0, 0.0]).await.unwrap(); // 与 query 完全同向 + repo.set_embedding("orth", &[0.0, 1.0]).await.unwrap(); // 正交 ≈ 0 + repo.set_embedding("neg", &[-1.0, 0.0]).await.unwrap(); // 反向 ≈ -1 + repo.set_embedding("hidden", &[1.0, 0.0]).await.unwrap(); // 非 published 过滤掉 + + let results = repo.search_vector(&[1.0, 0.0], 10).await.unwrap(); + // hidden 被 status 过滤,剩 3 条 + assert_eq!(results.len(), 3); + // 按相似度降序: exact(≈1) > orth(≈0) > neg(≈-1) + assert_eq!(results[0].0.id, "exact"); + assert_eq!(results[1].0.id, "orth"); + assert_eq!(results[2].0.id, "neg"); + assert!((results[0].1 - 1.0).abs() < 1e-5); + assert!(results[1].1.abs() < 1e-5); + assert!((results[2].1 + 1.0).abs() < 1e-5); + } + + #[tokio::test] + async fn search_vector_skips_dimension_mismatch() { + let repo = setup_repo().await; + repo.insert(krec("dim2", "e", "c", "lesson", "published", None, 0)) + .await + .unwrap(); + repo.insert(krec("dim3", "e", "c", "lesson", "published", None, 0)) + .await + .unwrap(); + + repo.set_embedding("dim2", &[1.0, 0.0]).await.unwrap(); + repo.set_embedding("dim3", &[1.0, 0.0, 0.0]).await.unwrap(); // 维度不匹配 + + // query 是 2 维,dim3 被跳过 + let results = repo.search_vector(&[1.0, 0.0], 10).await.unwrap(); + assert_eq!(results.len(), 1); + assert_eq!(results[0].0.id, "dim2"); + } + + #[tokio::test] + async fn search_vector_truncates_to_limit() { + let repo = setup_repo().await; + for i in 0..4 { + repo.insert(krec(&format!("k{i}"), "e", "c", "lesson", "published", None, 0)) + .await + .unwrap(); + // 全部与 query 同向,相似度相同 → 仅验证 truncate 生效 + repo.set_embedding(&format!("k{i}"), &[1.0, 0.0]).await.unwrap(); + } + let results = repo.search_vector(&[1.0, 0.0], 2).await.unwrap(); + assert_eq!(results.len(), 2); + } + + #[tokio::test] + async fn search_vector_ignores_records_without_embedding() { + let repo = setup_repo().await; + // published 但未写 embedding + repo.insert(krec("noemb", "e", "c", "lesson", "published", None, 0)) + .await + .unwrap(); + let results = repo.search_vector(&[1.0, 0.0], 10).await.unwrap(); + assert!(results.is_empty()); + } +} diff --git a/crates/df-storage/src/crud/mod.rs b/crates/df-storage/src/crud/mod.rs new file mode 100644 index 0000000..c624092 --- /dev/null +++ b/crates/df-storage/src/crud/mod.rs @@ -0,0 +1,277 @@ +//! Repository 模式的 CRUD 操作 +//! +//! 为 7 张表各提供一个 Repo 结构体,统一实现 insert / get_by_id / list_all / query / update_field / delete。 +//! +//! 按表域拆分(SMELL-P1-9,对齐 SMELL-P0-2 拆分思路降 2212 行单文件复杂度): +//! - [`mod@settings`]:SettingsRepo + 列白名单(allowed_columns_for/validate_column_name/is_allowed_column) +//! - [`mod@project_repo`]:ProjectRepo/BranchRepo/ReleaseRepo/WorkflowRepo/NodeExecutionRepo +//! - [`mod@task_repo`]:TaskRepo(含 advance_status_atomic 状态机收口) +//! - [`mod@conversation_repo`]:AiProviderRepo/AiConversationRepo/AiToolExecutionRepo +//! - [`mod@idea_repo`]:IdeaRepo/KnowledgeRepo/KnowledgeEventsRepo + 向量工具 +//! +//! re-export(`pub use ...::*`)保持 `df_storage::crud::XxxRepo` / +//! `df_storage::crud::is_allowed_column` 路径不变,**调用方零改动**。 + +mod conversation_repo; +mod idea_repo; +mod project_repo; +mod settings; +mod task_repo; + +pub use conversation_repo::*; +pub use idea_repo::*; +pub use project_repo::*; +pub use settings::*; +pub use task_repo::*; + +// ============================================================ +// 辅助宏 — 消除多个 Repo 的重复样板 +// ============================================================ + +/// 一次性为一张表生成完整的 Repo 结构体及其 CRUD 实现。 +/// +/// 用法: `impl_repo!(RepoName, ModelType, "table_name", from_row_body, insert_body, update_body);` +/// +/// 约束:`$record` 须实现 `Clone`(`update_full` 内部 clone 后丢给 spawn_blocking)。 +/// +/// 宏通过 `#[macro_use]`(见本文件末 `mod` 声明上方的文本注入)对各子模块可见, +/// 子模块 `impl_repo!(...)` 直接调用。`from_row`/`insert`/`update` 体内引用的 helper +/// (storage_err/now_millis_str/validate_column_name 及各 from_row)须在调用处可见。 +macro_rules! impl_repo { + ( + $(#[$meta:meta])* + $name:ident, $record:ident, $table:literal, + from_row => |$row:ident| $from_body:expr, + insert => |$conn:ident, $rec:ident| $insert_body:expr, + update => |$u_conn:ident, $u_rec:ident| $update_body:expr + ) => { + $(#[$meta])* + pub struct $name { + conn: Arc>, + } + + impl $name { + pub fn new(db: &Database) -> Self { + Self { conn: db.conn() } + } + + pub async fn insert(&self, record: $record) -> Result { + let conn = self.conn.clone(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let id = record.id.clone(); + let $conn = &*guard; + let $rec = &record; + $insert_body.map_err(storage_err)?; + Ok(id) + }) + .await + .map_err(storage_err)? + } + + pub async fn get_by_id(&self, id: &str) -> Result> { + let conn = self.conn.clone(); + let id = id.to_owned(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare(&format!("SELECT * FROM {} WHERE id = ?1", $table)) + .map_err(storage_err)?; + let row = stmt + .query_row(params![id], |$row| $from_body) + .optional() + .map_err(storage_err)?; + Ok(row) + }) + .await + .map_err(storage_err)? + } + + pub async fn list_all(&self) -> Result> { + let conn = self.conn.clone(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare(&format!("SELECT * FROM {} ORDER BY created_at DESC", $table)) + .map_err(storage_err)?; + let rows = stmt + .query_map([], |$row| $from_body) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push( + r.map_err(storage_err)?, + ); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } + + pub async fn query(&self, field: &str, value: &str) -> Result> { + // 白名单校验,防止 SQL 注入 + validate_column_name(field, $table)?; + let conn = self.conn.clone(); + let sql = format!("SELECT * FROM {} WHERE {} = ?1 ORDER BY created_at DESC", $table, field); + let value = value.to_owned(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare(&sql) + .map_err(storage_err)?; + let rows = stmt + .query_map(params![value], |$row| $from_body) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push( + r.map_err(storage_err)?, + ); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } + + pub async fn update_field(&self, id: &str, field: &str, value: &str) -> Result { + validate_column_name(field, $table)?; + let conn = self.conn.clone(); + let sql = format!("UPDATE {} SET {} = ?1, updated_at = ?2 WHERE id = ?3", $table, field); + let id = id.to_owned(); + let value = value.to_owned(); + let now = now_millis_str(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let affected = guard + .execute(&sql, params![value, now, id]) + .map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } + + /// 整体更新记录(全可变字段,保留 id 与 created_at),单次原子写 + pub async fn update_full(&self, record: &$record) -> Result { + let conn = self.conn.clone(); + let rec = record.clone(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let $u_conn = &*guard; + let $u_rec = &rec; + let affected = + $update_body.map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } + + pub async fn delete(&self, id: &str) -> Result { + let conn = self.conn.clone(); + let id = id.to_owned(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let affected = guard + .execute(&format!("DELETE FROM {} WHERE id = ?1", $table), params![id]) + .map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } + } + }; +} + +// 让 impl_repo! 宏对下方声明的子模块可见(textual scope 注入)。 +// 必须放在宏定义之后、`mod` 声明之前——顺序错了子模块拿不到宏。 +pub(crate) use impl_repo; + +// ============================================================ +// 公共工具(各子模块 use super::* 复用) +// ============================================================ + +/// rusqlite 错误 → df-types `Error::Storage` 统一包装(R-PD-10 DRY, +/// 替代散落的 `.map_err(storage_err)`)。 +pub(crate) fn storage_err(e: E) -> df_types::error::Error { + df_types::error::Error::Storage(e.to_string()) +} + +/// 规范化路径用于比较:canonicalize 解析绝对规范路径(失败降级), +/// 统一正斜杠 + 小写。与 `df_project::scan::normalize_path` **同算法镜像**(df-storage +/// 不依赖 df-project,故独立实现;改动须同步)。防 `C:\a\b` vs `C:/a/b/` 绕过。 +pub(crate) fn normalize_stored_path(p: &str) -> String { + match std::path::Path::new(p).canonicalize() { + Ok(abs) => abs.to_string_lossy().replace('\\', "/").to_lowercase(), + Err(_) => p + .trim_end_matches(['\\', '/']) + .replace('\\', "/") + .to_lowercase(), + } +} + +// ============================================================ +// 时间工具 +// ============================================================ + +pub(crate) fn now_millis_str() -> String { + // 转发 df_types::now_millis(DRY:时间获取统一入口在 df-types,避免本 crate 与 commands 层各写一份 SystemTime 调用) + df_types::now_millis().to_string() +} + +// ============================================================ +// 基线测试 — 锁 Repo 数/表名防回归(SMELL-P1-9 对齐 SMELL-P0-2) +// ============================================================ +// +// 拆分后,各 Repo 散落 5 个子模块。此基线测试确保: +// ① 所有 Repo 类型经 mod.rs `pub use` re-export 后在 `df_storage::crud::` 路径下仍可统一访问 +// (调用方 `df_storage::crud::XxxRepo::new(&db)` 路径不破,降回归)。 +// ② 各 Repo::new 在内存 DB 上构造不 panic + 列白名单覆盖每张表 +// (拆分时误删 Repo / 漏 re-export / 表名漂移会被此测试立即捕获)。 + +#[cfg(test)] +mod baseline_tests { + use super::*; + use crate::db::Database; + + /// 每张表都登记了列白名单(拆分后白名单仍在 settings.rs 单一来源,无表遗漏)。 + /// 新增表须同步登记白名单,否则此测试失败(防遗漏)。 + #[test] + fn all_known_tables_have_column_whitelist() { + let tables = [ + "ideas", "projects", "tasks", "releases", "branches", "workflow_executions", + "node_executions", "ai_providers", "ai_conversations", "ai_tool_executions", + "knowledges", "knowledge_events", + ]; + for t in tables { + assert!( + allowed_columns_for(t).is_some(), + "表 {t} 应登记列白名单(防 SQL 注入 + 按表隔离);拆分后白名单单一来源在 settings.rs" + ); + } + } + + /// re-export 完整性:所有 Repo 类型经 `df_storage::crud::XxxRepo` 路径可访问 + new 不 panic。 + /// 若拆分时漏 `pub use` 某子模块,对应 Repo 名在此处编译失败(立即暴露)。 + #[tokio::test] + async fn all_repos_constructible_in_memory() { + let db = Database::open_in_memory().await.expect("open_in_memory"); + // 各子模块 Repo:new 仅取 conn 句柄,不触 DB IO,构造成功即证明类型可见 + Database 路径通。 + let _ = SettingsRepo::new(&db); + let _ = IdeaRepo::new(&db); + let _ = ProjectRepo::new(&db); + let _ = TaskRepo::new(&db); + let _ = BranchRepo::new(&db); + let _ = ReleaseRepo::new(&db); + let _ = WorkflowRepo::new(&db); + let _ = NodeExecutionRepo::new(&db); + let _ = AiProviderRepo::new(&db); + let _ = AiConversationRepo::new(&db); + let _ = AiToolExecutionRepo::new(&db); + let _ = KnowledgeRepo::new(&db); + let _ = KnowledgeEventsRepo::new(&db); + } +} diff --git a/crates/df-storage/src/crud/project_repo.rs b/crates/df-storage/src/crud/project_repo.rs new file mode 100644 index 0000000..d224008 --- /dev/null +++ b/crates/df-storage/src/crud/project_repo.rs @@ -0,0 +1,360 @@ +//! 项目域 Repo:ProjectRepo / BranchRepo / ReleaseRepo / WorkflowRepo / NodeExecutionRepo + +use std::sync::Arc; + +use rusqlite::{params, Connection, OptionalExtension, Row}; +use tokio::sync::Mutex; + +use df_types::error::Result; + +use crate::db::Database; +use crate::models::{ + BranchRecord, NodeExecutionRecord, ProjectRecord, ReleaseRecord, WorkflowRecord, +}; + +use super::impl_repo; +use super::{normalize_stored_path, now_millis_str, storage_err, validate_column_name}; + +// ============================================================ +// from_row 辅助函数 +// ============================================================ + +fn project_from_row(row: &Row<'_>) -> std::result::Result { + Ok(ProjectRecord { + id: row.get("id")?, + name: row.get("name")?, + description: row.get("description")?, + status: row.get("status")?, + idea_id: row.get("idea_id")?, + path: row.get("path")?, + stack: row.get("stack")?, + created_at: row.get("created_at")?, + updated_at: row.get("updated_at")?, + }) +} + +fn branch_from_row(row: &Row<'_>) -> std::result::Result { + Ok(BranchRecord { + id: row.get("id")?, + project_id: row.get("project_id")?, + task_id: row.get("task_id")?, + name: row.get("name")?, + base: row.get("base")?, + status: row.get("status")?, + created_at: row.get("created_at")?, + updated_at: row.get("updated_at")?, + merged_at: row.get("merged_at")?, + }) +} + +fn release_from_row(row: &Row<'_>) -> std::result::Result { + Ok(ReleaseRecord { + id: row.get("id")?, + project_id: row.get("project_id")?, + version: row.get("version")?, + status: row.get("status")?, + task_ids: row.get("task_ids")?, + changelog: row.get("changelog")?, + created_at: row.get("created_at")?, + released_at: row.get("released_at")?, + }) +} + +fn workflow_from_row(row: &Row<'_>) -> std::result::Result { + Ok(WorkflowRecord { + id: row.get("id")?, + name: row.get("name")?, + dag_json: row.get("dag_json")?, + status: row.get("status")?, + triggered_by: row.get("triggered_by")?, + project_id: row.get("project_id")?, + task_id: row.get("task_id")?, + created_at: row.get("created_at")?, + completed_at: row.get("completed_at")?, + }) +} + +fn node_execution_from_row( + row: &Row<'_>, +) -> std::result::Result { + Ok(NodeExecutionRecord { + id: row.get("id")?, + workflow_id: row.get("workflow_id")?, + node_id: row.get("node_id")?, + node_type: row.get("node_type")?, + status: row.get("status")?, + input_json: row.get("input_json")?, + output_json: row.get("output_json")?, + error_message: row.get("error_message")?, + started_at: row.get("started_at")?, + completed_at: row.get("completed_at")?, + }) +} + +// ============================================================ +// Repo 实现 +// ============================================================ + +impl_repo!( + /// 项目表 CRUD + ProjectRepo, + ProjectRecord, + "projects", + from_row => |row| project_from_row(row), + insert => |conn, rec| { + conn.execute( + "INSERT INTO projects (id, name, description, status, idea_id, path, stack, created_at, updated_at) + VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9)", + params![ + rec.id, rec.name, rec.description, rec.status, rec.idea_id, + rec.path, rec.stack, rec.created_at, rec.updated_at + ], + ) + }, + update => |conn, rec| { + conn.execute( + "UPDATE projects SET name = ?1, description = ?2, status = ?3, idea_id = ?4, path = ?5, stack = ?6, updated_at = ?7 WHERE id = ?8", + params![ + rec.name, rec.description, rec.status, rec.idea_id, + rec.path, rec.stack, rec.updated_at, rec.id + ], + ) + } +); + +impl ProjectRepo { + /// 列出未删除项目(deleted_at IS NULL) + pub async fn list_active(&self) -> Result> { + let conn = self.conn.clone(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare("SELECT id, name, description, status, idea_id, path, stack, created_at, updated_at FROM projects WHERE deleted_at IS NULL ORDER BY created_at DESC") + .map_err(storage_err)?; + let rows = stmt + .query_map([], |row| project_from_row(row)) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push(r.map_err(storage_err)?); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } + + /// 软删:标记 deleted_at(进回收站,可恢复)。仅作用于未删项目,返回是否命中。 + pub async fn soft_delete(&self, id: &str) -> Result { + let conn = self.conn.clone(); + let id = id.to_owned(); + let now = now_millis_str(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let affected = guard + .execute( + "UPDATE projects SET deleted_at = ?1, updated_at = ?1 WHERE id = ?2 AND deleted_at IS NULL", + params![now, id], + ) + .map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } + + /// 恢复:清 deleted_at(从回收站还原) + pub async fn restore(&self, id: &str) -> Result { + let conn = self.conn.clone(); + let id = id.to_owned(); + let now = now_millis_str(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let affected = guard + .execute( + "UPDATE projects SET deleted_at = NULL, updated_at = ?1 WHERE id = ?2 AND deleted_at IS NOT NULL", + params![now, id], + ) + .map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } + + /// 查找已绑定该规范化路径的项目(排除 exclude_id 自身)。无冲突返回 None。 + /// + /// DRY(R-PD-11):统一 project.rs::find_binding_conflict 与 tool_registry.rs::bind_dir_to_project + /// 的「列项目→排除自身→按规范化路径比较」逻辑。本 crate 不依赖 df-project,故 norm_path 须由 + /// 调用方先经 `df_project::scan::normalize_path`(内含 canonicalize,防 `C:\a\b` vs `C:/a/b/` 绕过)。 + pub async fn find_path_conflict( + &self, + norm_path: &str, + exclude_id: Option<&str>, + ) -> Result> { + let projects = self.list_active().await?; + Ok(projects.into_iter().find(|p| { + let excluded = exclude_id.is_some_and(|eid| p.id == eid); + !excluded && p.path.as_ref().is_some_and(|pp| { + // 路径规范化由调用方完成目标侧;这里为已存项目路径补规范化 + // (历史路径写法可能不规范,统一规范化比较避免误判/漏判) + normalize_stored_path(pp) == norm_path + }) + })) + } + + /// 列出回收站(deleted_at IS NOT NULL),按更新时间(≈删除时间)降序 + pub async fn list_deleted(&self) -> Result> { + let conn = self.conn.clone(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare("SELECT id, name, description, status, idea_id, path, stack, created_at, updated_at FROM projects WHERE deleted_at IS NOT NULL ORDER BY updated_at DESC") + .map_err(storage_err)?; + let rows = stmt + .query_map([], |row| project_from_row(row)) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push(r.map_err(storage_err)?); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } + + /// 彻底删除:事务级联删 branches→releases→tasks→projects(不可恢复) + /// + /// SQLite 已开 PRAGMA foreign_keys=ON 但表无 ON DELETE CASCADE,ALTER 改不了 FK 约束, + /// 故应用层级联:子表先于父表删,单事务保证一致性。 + pub async fn purge_with_descendants(&self, id: &str) -> Result { + let conn = self.conn.clone(); + let id = id.to_owned(); + tokio::task::spawn_blocking(move || { + let mut guard = conn.blocking_lock(); + let tx = guard.transaction().map_err(storage_err)?; + tx.execute("DELETE FROM branches WHERE project_id = ?1", params![id]) + .map_err(storage_err)?; + tx.execute("DELETE FROM releases WHERE project_id = ?1", params![id]) + .map_err(storage_err)?; + tx.execute("DELETE FROM tasks WHERE project_id = ?1", params![id]) + .map_err(storage_err)?; + let affected = tx + .execute("DELETE FROM projects WHERE id = ?1", params![id]) + .map_err(storage_err)?; + tx.commit().map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } +} + +impl_repo!( + /// 分支表 CRUD + BranchRepo, + BranchRecord, + "branches", + from_row => |row| branch_from_row(row), + insert => |conn, rec| { + conn.execute( + "INSERT INTO branches (id, project_id, task_id, name, base, status, created_at, updated_at, merged_at) + VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9)", + params![ + rec.id, rec.project_id, rec.task_id, rec.name, rec.base, + rec.status, rec.created_at, rec.updated_at, rec.merged_at + ], + ) + }, + update => |conn, rec| { + conn.execute( + "UPDATE branches SET project_id = ?1, task_id = ?2, name = ?3, base = ?4, status = ?5, updated_at = ?6, merged_at = ?7 WHERE id = ?8", + params![ + rec.project_id, rec.task_id, rec.name, rec.base, + rec.status, rec.updated_at, rec.merged_at, rec.id + ], + ) + } +); + +impl_repo!( + /// 发布表 CRUD + ReleaseRepo, + ReleaseRecord, + "releases", + from_row => |row| release_from_row(row), + insert => |conn, rec| { + conn.execute( + "INSERT INTO releases (id, project_id, version, status, task_ids, changelog, created_at, released_at) + VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8)", + params![ + rec.id, rec.project_id, rec.version, rec.status, rec.task_ids, + rec.changelog, rec.created_at, rec.released_at + ], + ) + }, + update => |conn, rec| { + conn.execute( + "UPDATE releases SET project_id = ?1, version = ?2, status = ?3, task_ids = ?4, changelog = ?5, released_at = ?6 WHERE id = ?7", + params![ + rec.project_id, rec.version, rec.status, rec.task_ids, + rec.changelog, rec.released_at, rec.id + ], + ) + } +); + +impl_repo!( + /// 工作流执行表 CRUD + WorkflowRepo, + WorkflowRecord, + "workflow_executions", + from_row => |row| workflow_from_row(row), + insert => |conn, rec| { + conn.execute( + "INSERT INTO workflow_executions (id, name, dag_json, status, triggered_by, project_id, task_id, created_at, completed_at) + VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9)", + params![ + rec.id, rec.name, rec.dag_json, rec.status, rec.triggered_by, + rec.project_id, rec.task_id, rec.created_at, rec.completed_at + ], + ) + }, + update => |conn, rec| { + conn.execute( + "UPDATE workflow_executions SET name = ?1, dag_json = ?2, status = ?3, triggered_by = ?4, project_id = ?5, task_id = ?6, completed_at = ?7 WHERE id = ?8", + params![ + rec.name, rec.dag_json, rec.status, rec.triggered_by, + rec.project_id, rec.task_id, rec.completed_at, rec.id + ], + ) + } +); + +impl_repo!( + /// 节点执行表 CRUD + NodeExecutionRepo, + NodeExecutionRecord, + "node_executions", + from_row => |row| node_execution_from_row(row), + insert => |conn, rec| { + conn.execute( + "INSERT INTO node_executions (id, workflow_id, node_id, node_type, status, input_json, output_json, error_message, started_at, completed_at) + VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10)", + params![ + rec.id, rec.workflow_id, rec.node_id, rec.node_type, rec.status, + rec.input_json, rec.output_json, rec.error_message, rec.started_at, rec.completed_at + ], + ) + }, + update => |conn, rec| { + conn.execute( + "UPDATE node_executions SET workflow_id = ?1, node_id = ?2, node_type = ?3, status = ?4, input_json = ?5, output_json = ?6, error_message = ?7, started_at = ?8, completed_at = ?9 WHERE id = ?10", + params![ + rec.workflow_id, rec.node_id, rec.node_type, rec.status, + rec.input_json, rec.output_json, rec.error_message, rec.started_at, rec.completed_at, rec.id + ], + ) + } +); diff --git a/crates/df-storage/src/crud/settings.rs b/crates/df-storage/src/crud/settings.rs new file mode 100644 index 0000000..2740bd5 --- /dev/null +++ b/crates/df-storage/src/crud/settings.rs @@ -0,0 +1,206 @@ +//! 通用键值设置 — app_settings 表(KV,不适合 impl_repo! 宏)+ 列白名单(防 SQL 注入) + +use std::sync::Arc; + +use rusqlite::{params, Connection, OptionalExtension}; +use tokio::sync::Mutex; + +use df_types::error::{Error, Result}; + +use crate::db::Database; + +use super::{now_millis_str, storage_err}; + +// ============================================================ +// 通用键值设置 — app_settings 表 +// ============================================================ + +/// 通用应用设置 KV Repo(表 `app_settings`) +/// +/// 前端 localStorage 迁移目标:存主题/语言/AI 偏好/连接配置等,`value` 为 JSON 字符串。 +/// KV 无固定 schema 记录结构,故不走 `impl_repo!` 宏,手写更直白。 +pub struct SettingsRepo { + conn: Arc>, +} + +impl SettingsRepo { + pub fn new(db: &Database) -> Self { + Self { conn: db.conn() } + } + + /// 取单个 key 的值(JSON 字符串),不存在返回 None + pub async fn get(&self, key: &str) -> Result> { + let conn = self.conn.clone(); + let key = key.to_owned(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let v = guard + .query_row( + "SELECT value FROM app_settings WHERE key = ?1", + params![key], + |row| row.get::<_, String>(0), + ) + .optional() + .map_err(storage_err)?; + Ok(v) + }) + .await + .map_err(storage_err)? + } + + /// 写 key/value(`INSERT OR REPLACE`),刷新 updated_at + pub async fn set(&self, key: &str, value: &str) -> Result { + let conn = self.conn.clone(); + let key = key.to_owned(); + let value = value.to_owned(); + let now = now_millis_str(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let affected = guard + .execute( + "INSERT OR REPLACE INTO app_settings (key, value, updated_at) VALUES (?1, ?2, ?3)", + params![key, value, now], + ) + .map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } + + /// 取全部 key/value + pub async fn get_all(&self) -> Result> { + let conn = self.conn.clone(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare("SELECT key, value FROM app_settings") + .map_err(storage_err)?; + let rows = stmt + .query_map([], |row| Ok((row.get::<_, String>(0)?, row.get::<_, String>(1)?))) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push(r.map_err(storage_err)?); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } + + /// 删除 key,返回是否实际删除 + pub async fn delete(&self, key: &str) -> Result { + let conn = self.conn.clone(); + let key = key.to_owned(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let affected = guard + .execute("DELETE FROM app_settings WHERE key = ?1", params![key]) + .map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } +} + +// ============================================================ +// 列名白名单 — 防止 SQL 注入 +// ============================================================ + +/// 按表返回允许用于 query / update_field 的列名。 +/// +/// 白名单两重作用:① 防注入(列名参数化前校验);② 按表隔离(update_task 误传 +/// projects 的 "name" 会在校验阶段拒绝,而非靠 SQLite "no such column" 兜底报错)。 +/// 专用更新路径的列不列入:knowledges.embedding(set_embedding)、projects.deleted_at +/// (soft_delete/restore)、tasks.deleted_at(soft_delete/restore)。未登记的表返回 None +/// → 放行(仅靠参数化防注入,向后兼容)。 +pub fn allowed_columns_for(table: &str) -> Option<&'static [&'static str]> { + Some(match table { + "ideas" => &[ + "id", "title", "description", "status", "priority", "score", "tags", "source", + "promoted_to", "ai_analysis", "scores", "created_at", "updated_at", + ], + "projects" => &[ + "id", "name", "description", "status", "idea_id", "path", "stack", "created_at", + "updated_at", + ], + "tasks" => &[ + // id/created_at 不列入:主键与创建时间不可通过通用 update_field 改写 + // (防篡改主键/伪造创建时间/跨项目移动)。 + // status 不列入(D-260616-04 status 收口):所有 status 改动走 + // advance_status_atomic(CAS SQL `WHERE id AND status=expected`,唯一 status 写入路径), + // 不经通用 update_field。防 update_task IPC / AI 工具旁路改 status 绕过状态机 + // (can_transition 校验)与 review_rounds 累加。后人勿把 status 补回白名单。 + // review_rounds 不列入:它仅 advance_status_atomic 退回转换 + // (in_review→in_progress / testing→in_review)时原子 +1(收口:仅该专用路径可改, + // update_task/update_field 白名单均不含)。后人勿把 review_rounds 补进白名单, + // 否则破坏「review_rounds 唯一写入路径」收口、引入旁路写导致计数错乱。 + "project_id", "title", "description", "priority", "branch_name", + "assignee", "workflow_def_id", "base_branch", + // output_json:ai_execute 写产出 / ai_self_review 读产出自审 / human_review 展示对象 + // (决策 a:task 中心,产出跟 task 走)。非状态机收口字段,合法可写。 + "output_json", + "updated_at", + // TODO(B-260616-16): project_id 跨表存在性校验待 commands/task.rs 层补。 + // 通用 CRUD 层(db repo)只懂表/列语义,不持有跨表业务约束(查 projects 表存在性)。 + // project_id 当前可在白名单内改写,合法目标存在性由上层命令层校验。 + ], + "releases" => &[ + "id", "project_id", "version", "status", "task_ids", "changelog", "created_at", + "released_at", + ], + "branches" => &[ + "id", "project_id", "task_id", "name", "base", "status", "created_at", "updated_at", + "merged_at", + ], + "workflow_executions" => &[ + "id", "name", "dag_json", "status", "triggered_by", "project_id", "task_id", + "created_at", "completed_at", + ], + "node_executions" => &[ + "id", "workflow_id", "node_id", "node_type", "status", "input_json", "output_json", + "error_message", "started_at", "completed_at", + ], + "ai_providers" => &[ + "id", "name", "provider_type", "api_key", "base_url", "default_model", "models", + "is_default", "config", "created_at", "updated_at", + ], + "ai_conversations" => &[ + "id", "title", "messages", "provider_id", "model", "models", "archived", + "prompt_tokens", "completion_tokens", "created_at", "updated_at", + ], + "ai_tool_executions" => &[ + "id", "conversation_id", "tool_call_id", "tool_name", "arguments", "result", + "status", "risk_level", "requested_at", "executed_at", "decided_by", + ], + "knowledges" => &[ + "id", "kind", "title", "content", "tags", "status", "confidence", "reuse_count", + "verified", "source_project", "source_ref", "reasoning", "created_at", "updated_at", + ], + "knowledge_events" => &[ + "id", "knowledge_id", "event_type", "source_ref", "context_json", "timestamp", + ], + _ => return None, + }) +} + +pub(crate) fn validate_column_name(field: &str, table: &str) -> Result<()> { + match allowed_columns_for(table) { + Some(cols) if cols.contains(&field) => Ok(()), + Some(_) => Err(Error::Storage(format!("表 {} 不允许的字段名: {}", table, field))), + // 未登记表保守拒绝(FR-S6: 原放行 Ok,若未来未登记表走通用查询路径,列名直进字符串拼接即 SQL 注入;与 is_allowed_column 的 None=>false 对齐) + None => Err(Error::Storage(format!("表 {} 未登记列白名单,拒绝防注入", table))), + } +} + +/// 是否允许更新某表的某列(按表隔离白名单)。未登记表返回 false(保守拒绝,防误传)。 +/// +/// 供 AI 工具层等外部调用方复用 CRUD 白名单,避免双份字段列表不同步。 +pub fn is_allowed_column(table: &str, field: &str) -> bool { + match allowed_columns_for(table) { + Some(cols) => cols.contains(&field), + None => false, + } +} diff --git a/crates/df-storage/src/crud/task_repo.rs b/crates/df-storage/src/crud/task_repo.rs new file mode 100644 index 0000000..a146639 --- /dev/null +++ b/crates/df-storage/src/crud/task_repo.rs @@ -0,0 +1,218 @@ +//! 任务域 Repo:TaskRepo(含 advance_status_atomic 状态机收口) + +use std::sync::Arc; + +use rusqlite::{params, Connection, OptionalExtension, Row}; +use tokio::sync::Mutex; + +use df_types::error::Result; + +use crate::db::Database; +use crate::models::TaskRecord; + +use super::impl_repo; +use super::{now_millis_str, storage_err, validate_column_name}; + +// ============================================================ +// from_row 辅助函数 +// ============================================================ + +fn task_from_row(row: &Row<'_>) -> std::result::Result { + Ok(TaskRecord { + id: row.get("id")?, + project_id: row.get("project_id")?, + title: row.get("title")?, + description: row.get("description")?, + status: row.get("status")?, + priority: row.get("priority")?, + branch_name: row.get("branch_name")?, + assignee: row.get("assignee")?, + workflow_def_id: row.get("workflow_def_id")?, + base_branch: row.get("base_branch")?, + review_rounds: row.get("review_rounds")?, + output_json: row.get("output_json")?, + created_at: row.get("created_at")?, + updated_at: row.get("updated_at")?, + }) +} + +// ============================================================ +// Repo 实现 +// ============================================================ + +impl_repo!( + /// 任务表 CRUD + TaskRepo, + TaskRecord, + "tasks", + from_row => |row| task_from_row(row), + insert => |conn, rec| { + conn.execute( + "INSERT INTO tasks (id, project_id, title, description, status, priority, branch_name, assignee, workflow_def_id, base_branch, review_rounds, output_json, created_at, updated_at) + VALUES (?1, ?2, ?3, ?4, ?5, ?6, ?7, ?8, ?9, ?10, ?11, ?12, ?13, ?14)", + params![ + rec.id, rec.project_id, rec.title, rec.description, rec.status, rec.priority, + rec.branch_name, rec.assignee, rec.workflow_def_id, rec.base_branch, + rec.review_rounds, rec.output_json, rec.created_at, rec.updated_at + ], + ) + }, + update => |conn, rec| { + conn.execute( + "UPDATE tasks SET project_id = ?1, title = ?2, description = ?3, status = ?4, priority = ?5, branch_name = ?6, assignee = ?7, workflow_def_id = ?8, base_branch = ?9, review_rounds = ?10, output_json = ?11, updated_at = ?12 WHERE id = ?13", + params![ + rec.project_id, rec.title, rec.description, rec.status, rec.priority, + rec.branch_name, rec.assignee, rec.workflow_def_id, rec.base_branch, + rec.review_rounds, rec.output_json, rec.updated_at, rec.id + ], + ) + } +); + +impl TaskRepo { + /// 列出未删除任务(deleted_at IS NULL)— 对标 ProjectRepo::list_active + /// + /// 显式列出 14 个 TaskRecord 列名(同 ProjectRepo::list_active 写法), + /// 不 SELECT deleted_at:TaskRecord 不带该字段,取了 from_row 会因未知列报错。 + pub async fn list_active(&self) -> Result> { + let conn = self.conn.clone(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare("SELECT id, project_id, title, description, status, priority, branch_name, assignee, workflow_def_id, base_branch, review_rounds, output_json, created_at, updated_at FROM tasks WHERE deleted_at IS NULL ORDER BY created_at DESC") + .map_err(storage_err)?; + let rows = stmt + .query_map([], |row| task_from_row(row)) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push(r.map_err(storage_err)?); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } + + /// 软删:标记 deleted_at(进回收站,可恢复)。仅作用于未删任务,返回是否命中。 + /// 对标 ProjectRepo::soft_delete。 + pub async fn soft_delete(&self, id: &str) -> Result { + let conn = self.conn.clone(); + let id = id.to_owned(); + let now = now_millis_str(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let affected = guard + .execute( + "UPDATE tasks SET deleted_at = ?1, updated_at = ?1 WHERE id = ?2 AND deleted_at IS NULL", + params![now, id], + ) + .map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } + + /// 恢复:清 deleted_at(从回收站还原)。仅作用于已删任务,返回是否命中。 + /// 对标 ProjectRepo::restore。 + pub async fn restore(&self, id: &str) -> Result { + let conn = self.conn.clone(); + let id = id.to_owned(); + let now = now_millis_str(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let affected = guard + .execute( + "UPDATE tasks SET deleted_at = NULL, updated_at = ?1 WHERE id = ?2 AND deleted_at IS NOT NULL", + params![now, id], + ) + .map_err(storage_err)?; + Ok(affected > 0) + }) + .await + .map_err(storage_err)? + } + + /// 原子推进任务状态(任务推进链 F-260616-02 唯一 status 写入路径) + /// + /// 下沉 SQL `WHERE id=? AND status=?expected` 做 CAS(Compare-And-Swap)防 TOCTOU: + /// 并发推进/旁路修改若已改 status,affected_rows==0,本方法返回 None,调用方 + /// (task_advance_node)据此报「状态已变,推进中止」。`review_rounds` 不进通用 + /// update_field 白名单(收口:仅本方法可改 status 与 review_rounds)。 + /// + /// - `expected`:调用方读取的当前 status(状态机校验时的 from),CAS 前置。 + /// - `new_status`:目标 status(状态机 can_transition 已校验合法)。 + /// - `bump_rounds`:退回转换(in_review→in_progress / testing→in_review)传 true, + /// 一并 `review_rounds = review_rounds + 1`(同 UPDATE 原子,避免读改写竞争)。 + /// 前向推进 / 进出 blocked / 进 cancelled 传 false,不动 review_rounds。 + /// + /// 返回:成功推进返回更新后的 TaskRecord;affected==0(状态已变/任务不存在)返回 None。 + pub async fn advance_status_atomic( + &self, + id: &str, + expected: &str, + new_status: &str, + bump_rounds: bool, + ) -> Result> { + let conn = self.conn.clone(); + let id = id.to_owned(); + let expected = expected.to_owned(); + let new_status = new_status.to_owned(); + let now = now_millis_str(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + // CAS:WHERE id AND status=expected 锁定当前态;affected==0 即并发已改动。 + // deleted_at IS NULL:回收站任务(soft_delete 设了 deleted_at)CAS 必败→affected=0, + // 返回 None,杜绝回收站任务被推进(D-02 软删语义收口,一处关闭)。 + let sql = if bump_rounds { + "UPDATE tasks SET status = ?1, review_rounds = review_rounds + 1, updated_at = ?2 \ + WHERE id = ?3 AND status = ?4 AND deleted_at IS NULL" + } else { + "UPDATE tasks SET status = ?1, updated_at = ?2 \ + WHERE id = ?3 AND status = ?4 AND deleted_at IS NULL" + }; + let affected = guard + .execute(sql, params![new_status, now, id, expected]) + .map_err(storage_err)?; + if affected == 0 { + return Ok(None); + } + // 回读更新后的记录(含新 status / 累加后的 review_rounds / 新 updated_at)。 + let mut stmt = guard + .prepare("SELECT id, project_id, title, description, status, priority, branch_name, assignee, workflow_def_id, base_branch, review_rounds, output_json, created_at, updated_at FROM tasks WHERE id = ?1") + .map_err(storage_err)?; + let row = stmt + .query_row(params![id], |row| task_from_row(row)) + .optional() + .map_err(storage_err)?; + Ok(row) + }) + .await + .map_err(storage_err)? + } + + /// 列出回收站(deleted_at IS NOT NULL),按更新时间(≈删除时间)降序。对标 ProjectRepo::list_deleted。 + /// + /// 注:任务表无专用 list_active_by_project 方法,按项目列活跃任务由 commands/task.rs + /// 的 list_tasks 用 list_active 后内存过滤 project_id 实现(任务量小,无需 SQL 下推)。 + pub async fn list_deleted(&self) -> Result> { + let conn = self.conn.clone(); + tokio::task::spawn_blocking(move || { + let guard = conn.blocking_lock(); + let mut stmt = guard + .prepare("SELECT id, project_id, title, description, status, priority, branch_name, assignee, workflow_def_id, base_branch, review_rounds, output_json, created_at, updated_at FROM tasks WHERE deleted_at IS NOT NULL ORDER BY updated_at DESC") + .map_err(storage_err)?; + let rows = stmt + .query_map([], |row| task_from_row(row)) + .map_err(storage_err)?; + let mut results = Vec::new(); + for r in rows { + results.push(r.map_err(storage_err)?); + } + Ok(results) + }) + .await + .map_err(storage_err)? + } +} diff --git a/docs/todo.md b/docs/todo.md index 329cf7c..b05ebe3 100644 --- a/docs/todo.md +++ b/docs/todo.md @@ -176,7 +176,7 @@ - [ ] **SMELL-P1-6** — **[P1🟡]** String 类型滥用替 newtype/enum —— execution_id/status/tool_type 等 5+ 处改强类型 - [x] ✅(workflow weckqp9mv no-action·enum 负收益:archived/pinned 是 bool DB 列直映射,动作式 enum 在 crud 边界每调必 bool↔enum 转换只搬噪声;2 crud+2 IPC 调用者紧邻定义已自解释;真正歧义在前端 store.archiveConversation(id,true) 但正解是拆 archive/unarchive 两 TS 方法(纯前端重构)非推 enum 到 JS 边界;IPC 引 enum 触 Tauri 序列化层 prod runtime 才报错·主代 grep 印证) **SMELL-P1-7** — **[P1🟡]** bool 参数陷阱 —— `set_archived(id, bool)` / `set_pinned(id, bool)` 改 enum - [x] ✅(workflow wowdnw4ba·6篇文档加「实施状态(2026-06-18核对)」标注段·grep file:line佐证·不改设计正文·主代 grep 6篇标注段全在:B-03:11/AiNode:11/类型对齐:7/F-01:11/F-05:10/secret:10) **SMELL-P1-8** — **[P1🟡]** 架构文档过时 6 篇更新 —— B-03/AiNode自审/前后端类型对齐/F-01/F-05/secret下沉 标注实际实施状态 -- [ ] **SMELL-P1-9** — **[P1🟡]** crud.rs 2212 行按表拆分 —— project_repo / task_repo / conversation_repo / idea_repo 独立文件 +- [x] ✅(agent crud-split 实施 + 主代兜底核验·crud.rs 2212行→crud/ 6文件(mod/settings/project_repo/task_repo/conversation_repo/idea_repo)·re-export `pub use *_repo::*` 零调用方改动·宏 pub(crate) use + 子模块 use super::impl_repo·基线测试 all_known_tables_have_column_whitelist(12表)+ all_repos_constructible_in_memory(13Repo)·主代 cargo check --workspace EXIT 0 + df-storage 35+11 passed + devflow 96 passed) **SMELL-P1-9** — **[P1🟡]** crud.rs 2212 行按表拆分 —— project_repo / task_repo / conversation_repo / idea_repo 独立文件 **⚪ P2 可选优化 (5项)**: diff --git a/docs/待决策.md b/docs/待决策.md index 651b227..aae32af 100644 --- a/docs/待决策.md +++ b/docs/待决策.md @@ -151,7 +151,7 @@ - **e-1(主代已定✅)**:旧 loop save_conversation 保持原路径(save 接 conv_id 零改动)。低风险技术细节,主代裁决采纳。 - **推荐**:**✅ b-1 采纳(messages per-conv,技术必然)+ c-1 保持 3 + Settings 配置 + e-1 原路径**。拍板后启动阶段2 批1(PerConvState 数据结构 + 访问器)。 - **关联**:todo F-260616-09 / 设计文档 / memory aichat-arch-extensibility -- **状态**:🟡 待用户拍板(b-1/c-1) +- **状态**:✅ **主代自主裁决采纳(2026-06-19)** —— 用户授权「自主推进,能多角度确定的方案不等审批」。b-1 多角度论证充分(messages per-conv 是决策 e 技术必然必要条件,无替代);c-1 合理默认(global=3,Settings 配置后续);e-1 零改动技术细节。启动阶段2 批1。用户醒后可追认/推翻。 --- diff --git a/docs/待审查.md b/docs/待审查.md index 7fc9da5..cf25d85 100644 --- a/docs/待审查.md +++ b/docs/待审查.md @@ -694,6 +694,15 @@ **待修项回流 todo**: **无** 🔴/🟡 项 +### CR-260619-03 SMELL-P1-9 crud.rs按表拆分 + F-09批1 PerConvState(agent crud-split + f09-batch1 实施·主代统一兜底核验) — 🟡 待审 + +- **范围**(2 独立任务攒批): + - **SMELL-P1-9 crud.rs 拆分**(agent crud-split):`crud.rs` 2212 行 → `crud/` 6 文件(mod.rs 宏+工具+re-export / settings / project_repo / task_repo / conversation_repo / idea_repo)。12 Repo + from_row 按表域归属;re-export `pub use {conversation,idea,project,settings,task}_repo::*` 零调用方改动;宏 `pub(crate) use impl_repo` + 子模块 `use super::impl_repo`;基线测试 all_known_tables_have_column_whitelist(12 表)+ all_repos_constructible_in_memory(13 Repo new 不 panic)。 + - **F-09 批1 PerConvState**(agent f09-batch1,主代已自主裁决采纳决策 b-1):`commands/ai/mod.rs` 新增 PerConvState struct(9 字段,初值逐字对齐 AiSession::new)+ AiSession.per_conv HashMap 字段(共存期顶层字段仍是真相源)+ conv()(惰性创建)/conv_read()(只读)访问器 + 3 单测。**纯新增无行为变化**,批2 迁移承接。 +- **验证(主代统一兜底,稳定态·两 agent 都完成)**:cargo check --workspace EXIT 0 + df-storage 35 passed+11 integration + devflow 96 passed(原93+F-09批1新3)。 +- **审查要点(供审查 agent)**:① SMELL re-export 完整性(全仓 `df_storage::crud::XxxRepo` 路径不变,cargo --workspace 0 error 印证);② 宏可见性(pub(crate) use + 子模块 use super + helper import);③ 12 Repo/from_row 归属表正确(idea/project/task/conversation/settings 域);④ 基线测试锁回归(12 表白名单 + 13 Repo 构造);⑤ F-09 批1 PerConvState 字段初值对齐 AiSession::new + conv/conv_read 逻辑 + #[allow(dead_code)] 共存期标注合理(批2 迁移后移除);⑥ F-09 批1 纯新增无行为变化(顶层字段未迁移)。 +- **关联**:todo SMELL-P1-9 销账 / F-09 阶段2 批1(批2-8 待)。 + --- ## 已审归档 diff --git a/src-tauri/src/commands/ai/mod.rs b/src-tauri/src/commands/ai/mod.rs index 93fd4d5..18a311e 100644 --- a/src-tauri/src/commands/ai/mod.rs +++ b/src-tauri/src/commands/ai/mod.rs @@ -347,6 +347,13 @@ pub struct AiSession { /// 二次确认（AE-05）。换会话（new/switch）必须清空重审（见 ai_conversation_create / /// ai_conversation_switch 清字段）。 pub session_trust: HashSet, + /// F-260616-09 B 阶段批1 多会话并发架构：会话级状态按 conversation_id 切分。 + /// + /// **批1 状态(纯新增无行为变化)**:此 HashMap 初始空,批1 不迁移任何调用方 —— + /// 现有所有路径继续读写顶层单例字段(messages/generating/stop_flag 等),完全无感。 + /// 批2 承接迁移:调用方改走 `session.conv(conv_id).*`,迁移完成后顶层单例字段废弃删除。 + #[allow(dead_code)] // F-09 B 批1 共存期:批2 迁移承接(0 写入方) + pub per_conv: HashMap, } impl AiSession { @@ -364,6 +371,7 @@ impl AiSession { iteration_used: 0, model_override: None, session_trust: HashSet::new(), + per_conv: HashMap::new(), } } @@ -401,6 +409,194 @@ impl AiSession { SessionState::Idle } } + + // ============================================================ + // F-260616-09 B 阶段批1 PerConvState 访问器(设计 §2.1.2) + // + // 批1 仅定义访问器,**不迁移调用方**(批2 承接)。当前 0 调用方,行为完全不变。 + // 批2 迁移:所有 `session.messages` / `session.generating` 等读写改走 + // `session.conv(conv_id).*`(写)/ `session.conv_read(conv_id)`(读)。 + // ============================================================ + + /// 取或惰性创建指定会话的可变状态(设计 §2.1.2)。 + /// + /// 不存在则 `entry().or_insert_with(PerConvState::new)` 创建一份默认状态, + /// 初值对齐 [`AiSession::new`] / [`PerConvState::new`]。 + /// 已存在则返回同一实例引用(同 conv_id 复用)。 + /// + /// **批1 状态**:0 调用方(批2 迁移承接)。行为不变。 + #[allow(dead_code)] // F-09 B 批1 共存期:批2 迁移承接(0 调用方) + pub fn conv(&mut self, conv_id: &str) -> &mut PerConvState { + self.per_conv + .entry(conv_id.to_string()) + .or_insert_with(PerConvState::new) + } + + /// 只读取指定会话状态(不创建)。 + /// + /// 未创建的 conv_id 返 `None`,已创建返 `Some(&PerConvState)`。用于读取路径(如 + /// 列举/查询)避免误触发惰性创建。 + /// + /// **批1 状态**:0 调用方(批2 迁移承接)。行为不变。 + #[allow(dead_code)] // F-09 B 批1 共存期:批2 迁移承接(0 调用方) + pub fn conv_read(&self, conv_id: &str) -> Option<&PerConvState> { + self.per_conv.get(conv_id) + } +} + +#[cfg(test)] +mod tests_f09_per_conv { + use super::*; + + /// 验证 PerConvState::new() 各字段初值正确(对齐 AiSession::new 基线)。 + #[test] + fn test_per_conv_state_new() { + let s = PerConvState::new(); + assert!(!s.generating, "generating 初值应为 false"); + assert_eq!(s.iteration_used, 0, "iteration_used 初值应为 0"); + assert!(s.agent_language.is_none(), "agent_language 初值应为 None"); + assert!(s.model_override.is_none(), "model_override 初值应为 None"); + assert!(s.session_trust.is_empty(), "session_trust 初值应为空 HashSet"); + assert!(s.created_at.is_none(), "created_at 初值应为 None"); + // stop_flag 初值 false(与 AiSession::new 对齐) + assert!( + !s.stop_flag.load(std::sync::atomic::Ordering::SeqCst), + "stop_flag 初值应为 false" + ); + // notify 是 Arc,无可见状态,仅验证不 panic(构造即视为对齐) + // messages 是 ContextManager,内部状态;仅验证能取到引用(初值对齐由 ContextManager::new 保证) + let _messages_ref: &ContextManager = &s.messages; + } + + /// 验证 conv() 首次创建 + 二次复用同一实例。 + #[test] + fn test_conv_lazy_create() { + let mut session = AiSession::new(); + assert!(session.per_conv.is_empty(), "新建 session 的 per_conv 应为空"); + + // 首次 conv() 创建 + let ptr_first = { + let s = session.conv("conv-a"); + assert!(!s.generating); + assert_eq!(s.iteration_used, 0); + s as *const PerConvState + }; + assert_eq!(session.per_conv.len(), 1, "首次 conv() 后 per_conv 应有 1 项"); + + // 二次 conv(同 conv_id) 返回同一实例(指针相等) + let ptr_second = session.conv("conv-a") as *const PerConvState; + assert_eq!( + ptr_first, ptr_second, + "同 conv_id 二次 conv() 应返回同一实例" + ); + assert_eq!(session.per_conv.len(), 1, "二次 conv(同 id) 不应新增项"); + + // 不同 conv_id 创建独立实例 + let _s_b = session.conv("conv-b"); + assert_eq!(session.per_conv.len(), 2, "不同 conv_id 应新建独立实例"); + } + + /// 验证 conv_read() 未创建的 conv_id 返 None(只读,不惰性创建)。 + #[test] + fn test_conv_read_none() { + let session = AiSession::new(); + // 未创建的 conv_id + assert!( + session.conv_read("never-created").is_none(), + "未创建的 conv_id conv_read() 应返 None" + ); + // conv_read 不触发创建(per_conv 仍空) + assert!( + session.per_conv.is_empty(), + "conv_read 不应触发惰性创建" + ); + + // 已创建的 conv_id 返 Some(只读访问) + let mut session = session; + let _ = session.conv("exists"); + assert!( + session.conv_read("exists").is_some(), + "已创建的 conv_id conv_read() 应返 Some" + ); + } +} + +// ============================================================ +// F-260616-09 B 阶段 多会话并发架构 — PerConvState(批1 纯新增) +// +// 设计文档: docs/02-架构设计/F-09-多会话并发架构设计-2026-06-19.md §2.1.2 +// +// 批1 范围(纯新增无行为变化): +// - 仅新增 PerConvState struct + new() + AiSession.per_conv 字段 + conv()/conv_read() 访问器 +// - **不迁移**现有单例字段(messages/generating/stop_flag/notify/iteration_used/ +// agent_language/model_override/session_trust 保留原样),不迁移现有调用方(批2 承接) +// - per_conv HashMap 初始空,旧调用路径继续读写 AiSession 顶层单例字段,完全无感 +// +// 共存期:批1 后顶层单例字段(唯一真相源)与 per_conv(全空,无写入)并存,行为不变。 +// 批2 迁移:把所有调用方从 session.messages / session.generating 等改为 session.conv(conv_id).*, +// 迁移完成后顶层单例字段废弃删除(批3)。 +// ============================================================ + +/// 单会话级状态(F-260616-09 B 阶段批1 纯新增,设计 §2.1.2)。 +/// +/// 持有当前会话(按 conversation_id 切分)私有的可变状态。批1 仅定义 + 惰性创建访问器, +/// **不迁移** AiSession 顶层单例字段(共存期顶层字段仍是唯一真相源)。 +/// +/// 字段初值逐字对齐 [`AiSession::new`](#method.new),保证批2 迁移调用方时行为不变。 +#[allow(dead_code)] // F-09 B 批1 共存期:0 调用方(批2 迁移承接),字段待迁移后消费 +pub struct PerConvState { + /// 对话历史(ContextManager:会话级消息真相源,裁剪仅影响发送视图) + pub messages: ContextManager, + /// 是否正在生成 + pub generating: bool, + /// 停止信号(会话级):ai_chat_stop 置位,agentic loop / stream_llm 检测后尽快退出 + pub stop_flag: Arc, + /// 即时停止唤醒(会话级):阻塞在 stream.next() 时 notify_one() 立即唤醒跳出 select! + pub notify: Arc, + /// agentic loop 累计已跑 iteration 计数(语义同 AiSession.iteration_used) + pub iteration_used: usize, + /// 当前 agent 循环的语言设置(用于审批后恢复循环) + pub agent_language: Option, + /// 用户指定模型 override(主对话专用,语义同 AiSession.model_override) + pub model_override: Option, + /// 会话级信任(Session Trust):随会话销毁,不落库 + pub session_trust: HashSet, + /// 会话创建时间(懒创建:首条消息落库前仅存内存,upsert 时用作 created_at) + pub created_at: Option, +} + +impl PerConvState { + /// 新建默认会话级状态。 + /// + /// 各字段初值**逐字对齐 [`AiSession::new`]** —— 批2 迁移调用方时行为不变。 + /// - messages: ContextManager::new(ContextConfig::default()) + /// - generating: false + /// - stop_flag: Arc::new(AtomicBool::new(false)) + /// - notify: Arc::new(tokio::sync::Notify::new()) + /// - iteration_used: 0 + /// - agent_language: None + /// - model_override: None + /// - session_trust: HashSet::new() + /// - created_at: None(批1 新增字段,AiSession 现有 active_conv_created_at 同语义) + pub fn new() -> Self { + Self { + messages: ContextManager::new(ContextConfig::default()), + generating: false, + stop_flag: Arc::new(AtomicBool::new(false)), + notify: Arc::new(tokio::sync::Notify::new()), + iteration_used: 0, + agent_language: None, + model_override: None, + session_trust: HashSet::new(), + created_at: None, + } + } +} + +impl Default for PerConvState { + fn default() -> Self { + Self::new() + } } /// 待审批的工具调用