dsh-vscode-mode 0.14.2 → 0.14.3

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-vscode-mode",
-  "version": "0.14.2",
+  "version": "0.14.3",
   "description": "DSH 上的类 VSCode 编码体验：Monaco 编辑器（文件页签/QuickOpen/状态栏，可驻 DSH 0.1.5+ 官方右侧 Sidebar 与对话同屏）+ 命令栏（Ctrl+Shift+P）与可扩展指令系统 + VS Code 兼容代码片段（.code-snippets 全局/项目，IntelliSense 展开）+ Agent 编辑差异审查（整文件差异/采纳/拒绝/归档/回滚）+ SVN 集成（侧栏变更面板/日志弹窗/并排差异）+ 语言服务器（LSP）智能（跳转定义/引用/hover/大纲 + VSIX 扩展市场安装），状态持久化到工作区旁车，界面状态（页签/光标/侧栏/编辑 Tab/导航历史）镜像到 DSH home 跨重启恢复",
   "icon": "assets/icon.svg",
   "keywords": [

package/src/archiveDb.ts

@@ -0,0 +1,476 @@
+/**
+ * 工作区差异归档 V2：SQLite 事务追加与按需读取；旧 JSON 只读留存，不参与热路径。
+ * 数据库位于 DSH home 的 data/，不会被缓存清理器删除。
+ * 作者 ddj 2026-09-30
+ */
+import { mkdirSync } from 'node:fs'
+import { createHash } from 'node:crypto'
+import { dirname } from 'node:path'
+import { DatabaseSync } from 'node:sqlite'
+import type { ArchiveBatch, ArchiveEntry, ArchiveRecord, ArchiveSummary, DiffRecord, SidecarData } from './shared/types.js'
+import { archiveDbFile } from './paths.js'
+
+export type RevertState = 'started' | 'applied'
+export interface ActiveChange { upserts: DiffRecord[]; removed: string[] }
+export interface ArchivePage { entries: ArchiveEntry[]; nextCursor: string | null }
+
+/**
+ * 数据库行中 null 批次用独立字符串键，避免 SQLite UNIQUE 对 NULL 不去重。
+ * @author ddj 2026年09月30号
+ * @param batch 原始批次
+ * @returns 可用于唯一约束的字符串键
+ */
+const batchKey = (batch: number | null): string => batch === null ? 'null' : 'n' + batch
+
+/** 仅由本插件打开的工作区连接；插件卸载时 closeArchiveDbs 关闭。 */
+const connections = new Map<string, DatabaseSync>()
+
+type Row = Record<string, string | number | null>
+
+/**
+ * 打开且核验单工作区数据库；已有异常库绝不重置成空库。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区绝对路径
+ * @returns 已缓存连接
+ */
+function dbFor(cwd: string): DatabaseSync {
+  const existing = connections.get(cwd)
+  if (existing) return existing
+  const file = archiveDbFile(cwd)
+  mkdirSync(dirname(file), { recursive: true, mode: 0o700 })
+  const db = new DatabaseSync(file, { timeout: 2000 })
+  try {
+    db.exec('PRAGMA journal_mode=WAL; PRAGMA synchronous=FULL; PRAGMA busy_timeout=2000')
+    db.exec('CREATE TABLE IF NOT EXISTS metadata (name TEXT PRIMARY KEY, value TEXT NOT NULL)')
+    const version = db.prepare("SELECT value FROM metadata WHERE name='schema'").get() as Row | undefined
+    const workspace = db.prepare("SELECT value FROM metadata WHERE name='cwd'").get() as Row | undefined
+    if ((version && version.value !== '2' && version.value !== '3') || (workspace && workspace.value !== cwd)) throw new Error('归档数据库 schema/工作区不匹配')
+    if (!!version !== !!workspace) throw new Error('归档数据库元数据不完整')
+    if (!version) {
+      const other = db.prepare("SELECT name FROM sqlite_master WHERE type='table' AND name NOT LIKE 'sqlite_%' AND name<>'metadata' LIMIT 1").get()
+      if (other) throw new Error('归档数据库缺失元数据且已有数据表')
+    }
+    db.exec(`CREATE TABLE IF NOT EXISTS batches (
+        id INTEGER PRIMARY KEY, cwd TEXT NOT NULL, path TEXT NOT NULL, batch_key TEXT NOT NULL,
+        batch INTEGER, at TEXT NOT NULL, last_at TEXT, reason TEXT,
+        accepted INTEGER NOT NULL DEFAULT 0, rejected INTEGER NOT NULL DEFAULT 0,
+        pending INTEGER NOT NULL DEFAULT 0, superseded INTEGER NOT NULL DEFAULT 0,
+        n_records INTEGER NOT NULL DEFAULT 0, UNIQUE(cwd,path,batch_key)
+      );
+      CREATE TABLE IF NOT EXISTS records (
+        batch_id INTEGER NOT NULL REFERENCES batches(id), call_id TEXT NOT NULL,
+        payload TEXT NOT NULL, accepted INTEGER NOT NULL, rejected INTEGER NOT NULL,
+        pending INTEGER NOT NULL, superseded INTEGER NOT NULL,
+        PRIMARY KEY(batch_id,call_id)
+      );
+      CREATE INDEX IF NOT EXISTS archive_by_path ON batches(cwd,path,batch);
+      CREATE INDEX IF NOT EXISTS archive_by_call ON records(call_id);
+      CREATE TABLE IF NOT EXISTS revert_intents (
+        call_id TEXT NOT NULL, scope TEXT NOT NULL, hunk_index INTEGER NOT NULL,
+        state TEXT NOT NULL CHECK(state IN ('started','applied')),
+        PRIMARY KEY(call_id,scope,hunk_index)
+      )`)
+    if (!version) {
+      db.exec('BEGIN IMMEDIATE')
+      try {
+        db.prepare("INSERT INTO metadata(name,value) VALUES('schema','2')").run()
+        db.prepare("INSERT INTO metadata(name,value) VALUES('cwd',?)").run(cwd)
+        db.exec('COMMIT')
+      } catch (error) {
+        if (db.isTransaction) db.exec('ROLLBACK')
+        throw error
+      }
+    }
+    if (version?.value === '3') {
+      const active = db.prepare("SELECT name FROM sqlite_master WHERE type='table' AND name='active_records'").get()
+      const source = db.prepare("SELECT value FROM metadata WHERE name='active_source'").get()
+      if (!active || !source) throw new Error('活跃记录数据库元数据不完整')
+    }
+    connections.set(cwd, db)
+    return db
+  } catch (error) {
+    db.close()
+    throw error
+  }
+}
+
+/**
+ * 只在隔离的切换窗口导入旧 sidecar；不接受模糊空库或解析损坏的输入。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @param source 旧文件原文（null 仅表示明确不存在）
+ * @param confirm 再读源文件原文，防止迁移途中旧 Host 改写
+ * @returns 活跃记录 Map
+ */
+export async function migrateActive(cwd: string, source: string | null, confirm: () => Promise<string | null>): Promise<Map<string, DiffRecord>> {
+  const db = dbFor(cwd)
+  const version = String((db.prepare("SELECT value FROM metadata WHERE name='schema'").get() as Row).value)
+  if (version === '3') return readActive(cwd)
+  const digest = source === null ? 'absent' : createHash('sha256').update(source).digest('hex')
+  let recs: Record<string, DiffRecord> = {}
+  if (source !== null) {
+    let parsed: SidecarData
+    try { parsed = JSON.parse(source) as SidecarData } catch { throw new Error('旧活跃 sidecar 损坏，停止迁移') }
+    if (parsed?.version !== 2 || !parsed.workspaces || typeof parsed.workspaces !== 'object' || Array.isArray(parsed.workspaces)) throw new Error('旧活跃 sidecar 格式无效')
+    const workspace = parsed.workspaces[cwd]
+    if (workspace && (!workspace.records || typeof workspace.records !== 'object' || Array.isArray(workspace.records))) throw new Error('旧工作区记录格式无效')
+    recs = workspace?.records ?? {}
+    for (const [key, rec] of Object.entries(recs)) {
+      if (!rec || rec.callId !== key || typeof rec.path !== 'string' || !Array.isArray(rec.hunks)
+        || !rec.decisions || !Array.isArray(rec.decisions.perHunk) || !Number.isInteger(rec.batch)
+        || typeof rec.at !== 'string') throw new Error('旧活跃记录损坏：' + key)
+    }
+  }
+  if (await confirm() !== source) throw new Error('迁移期间旧活跃 sidecar 被修改')
+  if (activeReady(cwd)) return readActive(cwd)
+  db.exec('BEGIN IMMEDIATE')
+  try {
+    db.exec(`CREATE TABLE active_records(call_id TEXT PRIMARY KEY, payload TEXT NOT NULL,
+      path TEXT NOT NULL, archived INTEGER NOT NULL, pending INTEGER NOT NULL, at TEXT NOT NULL);
+      CREATE INDEX active_by_path ON active_records(path,archived)`)
+    const add = db.prepare('INSERT INTO active_records(call_id,payload,path,archived,pending,at) VALUES(?,?,?,?,?,?)')
+    for (const rec of Object.values(recs)) add.run(rec.callId, JSON.stringify(rec), rec.path, rec.archived ? 1 : 0, rec.decisions?.perHunk?.filter((d) => d === 'pending').length ?? 0, rec.at)
+    if (await confirm() !== source) throw new Error('迁移期间旧活跃 sidecar 被修改')
+    db.prepare("INSERT INTO metadata(name,value) VALUES('active_source',?)").run(digest)
+    db.prepare("UPDATE metadata SET value='3' WHERE name='schema'").run()
+    db.exec('COMMIT')
+    return new Map(Object.values(recs).map((rec) => [rec.callId, rec]))
+  } catch (error) {
+    if (db.isTransaction) db.exec('ROLLBACK')
+    throw error
+  }
+}
+
+/**
+ * 判断已核验的工作区是否完成活跃表迁移，不写入工作区记录。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @returns 迁移完成返回 true
+ */
+export function activeReady(cwd: string): boolean {
+  return (dbFor(cwd).prepare("SELECT value FROM metadata WHERE name='schema'").get() as Row).value === '3'
+}
+
+/**
+ * 对旧文件原文核对迁移标记，旧 Host 若继续写旧 sidecar 则拒绝启用新活跃库。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @param source 当前遗留文件原文
+ */
+export function checkActiveSource(cwd: string, source: string | null): void {
+  const db = dbFor(cwd)
+  const row = db.prepare("SELECT value FROM metadata WHERE name='active_source'").get() as Row | undefined
+  if (!row) throw new Error('活跃记录缺少迁移标记')
+  const digest = source === null ? 'absent' : createHash('sha256').update(source).digest('hex')
+  if (row.value !== digest) throw new Error('旧活跃 sidecar 在迁移后发生变化，请停止操作并人工核对')
+}
+
+/**
+ * 只读当前工作区活跃表；未迁移时禁止回退到空库。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @returns 完整有限记录桶
+ */
+export function readActive(cwd: string): Map<string, DiffRecord> {
+  const db = dbFor(cwd)
+  if ((db.prepare("SELECT value FROM metadata WHERE name='schema'").get() as Row).value !== '3') throw new Error('工作区活跃记录未迁移')
+  const rows = db.prepare('SELECT payload FROM active_records ORDER BY rowid').all() as Row[]
+  const recs = rows.map((row) => JSON.parse(String(row.payload)) as DiffRecord)
+  return new Map(recs.map((rec) => [rec.callId, rec]))
+}
+
+/**
+ * 单次 SQLite 事务提交活跃行变化及可选归档，绝不重写整个记录桶。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @param changes 明确新增、更新和移除的记录
+ * @param entries 可选本次归档批次
+ * @returns 新增归档批次数
+ */
+export function commitActive(cwd: string, changes: ActiveChange, entries: ArchiveBatch[] = []): number {
+  const db = dbFor(cwd)
+  if ((db.prepare("SELECT value FROM metadata WHERE name='schema'").get() as Row).value !== '3') throw new Error('工作区活跃记录未迁移')
+  const upsert = db.prepare(`INSERT INTO active_records(call_id,payload,path,archived,pending,at) VALUES(?,?,?,?,?,?)
+    ON CONFLICT(call_id) DO UPDATE SET payload=excluded.payload,path=excluded.path,
+      archived=excluded.archived,pending=excluded.pending,at=excluded.at`)
+  const remove = db.prepare('DELETE FROM active_records WHERE call_id=?')
+  db.exec('BEGIN IMMEDIATE')
+  try {
+    const added = entries.length ? insertArchive(db, cwd, entries) : 0
+    for (const rec of changes.upserts) {
+      const pending = rec.decisions?.perHunk?.filter((decision) => decision === 'pending').length ?? 0
+      upsert.run(rec.callId, JSON.stringify(rec), rec.path, rec.archived ? 1 : 0, pending, rec.at)
+    }
+    for (const callId of changes.removed) remove.run(callId)
+    db.exec('COMMIT')
+    return added
+  } catch (error) {
+    if (db.isTransaction) db.exec('ROLLBACK')
+    throw error
+  }
+}
+
+/**
+ * 轻量活跃摘要；旧 JSON 保留为迁移前快照而不参与统计。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @returns 活跃条数和每文件待处理计数
+ */
+export function activeSummary(cwd: string): { active: number; pendingByFile: { path: string; pending: number }[] } {
+  const db = dbFor(cwd)
+  if ((db.prepare("SELECT value FROM metadata WHERE name='schema'").get() as Row).value !== '3') throw new Error('工作区活跃记录未迁移')
+  const active = Number((db.prepare('SELECT count(*) AS n FROM active_records WHERE archived=0').get() as Row).n)
+  const rows = db.prepare('SELECT path, sum(pending) AS pending FROM active_records WHERE archived=0 AND pending>0 GROUP BY path ORDER BY pending DESC').all() as Row[]
+  return { active, pendingByFile: rows.map((row) => ({ path: String(row.path), pending: Number(row.pending) })) }
+}
+
+/**
+ * 对一批同路径记录做事务内去重追加，返回成功插入的批次数。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @param entries 已按旧语义构造的归档批次
+ * @returns 新增批次数
+ */
+export function appendArchive(cwd: string, entries: ArchiveBatch[]): number {
+  if (!entries.length) return 0
+  const db = dbFor(cwd)
+  db.exec('BEGIN IMMEDIATE')
+  try {
+    const added = insertArchive(db, cwd, entries)
+    db.exec('COMMIT')
+    return added
+  } catch (error) {
+    if (db.isTransaction) db.exec('ROLLBACK')
+    throw error
+  }
+}
+
+/**
+ * 在调用方事务内追加归档批次；归档与活跃行可共用一次提交。
+ * @author ddj 2026年09月30号
+ * @param db 当前工作区连接
+ * @param cwd 工作区
+ * @param entries 待归档条目
+ * @returns 新增批次数
+ */
+function insertArchive(db: DatabaseSync, cwd: string, entries: ArchiveBatch[]): number {
+  const get = db.prepare('SELECT id FROM batches WHERE cwd=? AND path=? AND batch_key=?')
+  const add = db.prepare('INSERT INTO batches(cwd,path,batch_key,batch,at,last_at,reason) VALUES(?,?,?,?,?,?,?)')
+  const update = db.prepare('UPDATE batches SET last_at=?, reason=? WHERE id=?')
+  const insert = db.prepare('INSERT OR IGNORE INTO records(batch_id,call_id,payload,accepted,rejected,pending,superseded) VALUES(?,?,?,?,?,?,?)')
+  const tally = db.prepare(`UPDATE batches SET n_records=n_records+1, accepted=accepted+?, rejected=rejected+?,
+    pending=pending+?, superseded=superseded+? WHERE id=?`)
+  let added = 0
+  for (const entry of entries) {
+    if (entry.cwd !== cwd) throw new Error('归档批次工作区不匹配')
+    const key = batchKey(entry.batch)
+    const found = get.get(cwd, entry.path, key) as Row | undefined
+    const id = found ? Number(found.id) : Number(add.run(cwd, entry.path, key, entry.batch, entry.at, entry.lastAt ?? null, entry.reason).lastInsertRowid)
+    if (found) update.run(entry.at, entry.reason, id)
+    else added++
+    for (const record of entry.records) {
+      const sum = record.summary
+      const result = insert.run(id, record.callId, JSON.stringify(record), sum.accepted, sum.rejected, sum.pending, sum.superseded ? 1 : 0)
+      if (result.changes) tally.run(sum.accepted, sum.rejected, sum.pending, sum.superseded ? 1 : 0, id)
+    }
+  }
+  const clear = db.prepare('DELETE FROM revert_intents WHERE call_id=?')
+  for (const entry of entries) for (const record of entry.records) clear.run(record.callId)
+  return added
+}
+
+/**
+ * 返回某工作区所有归档批次的轻量摘要，不取记录 JSON。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @returns 归档列表
+ */
+export function listArchives(cwd: string): ArchiveEntry[] {
+  const rows = dbFor(cwd).prepare(`SELECT path,batch,at,last_at,reason,n_records,accepted,rejected,pending,superseded
+    FROM batches WHERE cwd=? ORDER BY COALESCE(batch,-1) DESC, at DESC`).all(cwd) as Row[]
+  return rows.map((row) => ({
+    path: String(row.path), batch: row.batch === null ? null : Number(row.batch),
+    at: String(row.at), lastAt: String(row.last_at ?? row.at), reason: row.reason === null ? null : String(row.reason),
+    nRecords: Number(row.n_records), summary: {
+      accepted: Number(row.accepted), rejected: Number(row.rejected), pending: Number(row.pending), superseded: Number(row.superseded),
+    } as ArchiveSummary,
+  }))
+}
+
+/**
+ * 固定快照上界的 keyset 分页，新增批次不挤动后续页面。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @param cursor 上页返回的不透明游标
+ * @param limit 每页条数，缺省 50，最多 100
+ * @returns 摘要页和下一页游标
+ */
+export function pageArchives(cwd: string, cursor?: string, limit = 50): ArchivePage {
+  if (!Number.isInteger(limit) || limit < 1 || limit > 100) throw new Error('归档分页大小不合法')
+  const db = dbFor(cwd)
+  const newest = Number((db.prepare('SELECT COALESCE(MAX(id),0) AS id FROM batches WHERE cwd=?').get(cwd) as Row).id)
+  let snapshot = newest
+  let last: [number, string, number] | null = null
+  if (cursor) {
+    try {
+      const value = JSON.parse(Buffer.from(cursor, 'base64url').toString('utf8')) as { cwd: string; snapshot: number; last: [number, string, number] }
+      if (value.cwd !== cwd || !Number.isSafeInteger(value.snapshot) || value.snapshot < 0 || value.snapshot > newest
+        || !Array.isArray(value.last) || value.last.length !== 3 || !Number.isSafeInteger(value.last[0])
+        || typeof value.last[1] !== 'string' || !Number.isSafeInteger(value.last[2]) || value.last[2] > value.snapshot) throw new Error('invalid')
+      snapshot = value.snapshot
+      last = value.last
+    } catch { throw new Error('归档分页游标无效') }
+  }
+  const args: Array<string | number> = [cwd, snapshot]
+  let where = 'WHERE cwd=? AND id<=?'
+  if (last) {
+    where += ' AND (COALESCE(batch,-1),at,id) < (?,?,?)'
+    args.push(...last)
+  }
+  const rows = db.prepare(`SELECT id,path,batch,at,last_at,reason,n_records,accepted,rejected,pending,superseded
+    FROM batches ${where} ORDER BY COALESCE(batch,-1) DESC, at DESC, id DESC LIMIT ?`).all(...args, limit + 1) as Row[]
+  const entries = rows.slice(0, limit).map((row) => ({
+    path: String(row.path), batch: row.batch === null ? null : Number(row.batch), at: String(row.at),
+    lastAt: String(row.last_at ?? row.at), reason: row.reason === null ? null : String(row.reason),
+    nRecords: Number(row.n_records), summary: {
+      accepted: Number(row.accepted), rejected: Number(row.rejected), pending: Number(row.pending), superseded: Number(row.superseded),
+    },
+  }))
+  const tail = rows[limit - 1]
+  const nextCursor = rows.length > limit && tail
+    ? Buffer.from(JSON.stringify({ cwd, snapshot, last: [tail.batch === null ? -1 : Number(tail.batch), tail.at, tail.id] })).toString('base64url') : null
+  return { entries, nextCursor }
+}
+
+/**
+ * 仅查询需要展开的路径/批次（无 path 时维持 RPC 返回全库语义）。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @param path 可选目标路径
+ * @param batch 可选批次
+ * @returns 原有 ArchiveBatch 格式
+ */
+export function readArchives(cwd: string, path?: string, batch?: number | null): ArchiveBatch[] {
+  const db = dbFor(cwd)
+  let where = 'WHERE cwd=?'
+  const args: Array<string | number> = [cwd]
+  if (path) { where += ' AND path=?'; args.push(path) }
+  if (batch !== undefined) { where += ' AND batch_key=?'; args.push(batchKey(batch)) }
+  const rows = db.prepare(`SELECT id,cwd,path,batch,at,last_at,reason FROM batches ${where} ORDER BY id`).all(...args) as Row[]
+  const records = db.prepare('SELECT payload FROM records WHERE batch_id=? ORDER BY rowid')
+  return rows.map((row) => ({
+    cwd: String(row.cwd), path: String(row.path), batch: row.batch === null ? null : Number(row.batch),
+    at: String(row.at), ...(row.last_at === null ? {} : { lastAt: String(row.last_at) }), reason: String(row.reason ?? ''),
+    records: (records.all(row.id as number) as Row[]).map((record) => JSON.parse(String(record.payload)) as ArchiveRecord),
+  }))
+}
+
+/**
+ * 查询已提交的归档记录，用于数据库成功但活跃 sidecar 写入失败后的幂等补偿。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @param callId 调用标识
+ * @param path 目标文件路径
+ * @param batch 对应文件批次
+ * @returns 已归档快照或 null
+ */
+export function committedRecord(cwd: string, callId: string, path: string, batch: number | null): ArchiveRecord | null {
+  const row = dbFor(cwd).prepare(`SELECT r.payload FROM records r JOIN batches b ON b.id=r.batch_id
+    WHERE b.cwd=? AND b.path=? AND b.batch_key=? AND r.call_id=? LIMIT 1`).get(cwd, path, batchKey(batch), callId) as Row | undefined
+  return row ? JSON.parse(String(row.payload)) as ArchiveRecord : null
+}
+
+/**
+ * 查询一次 Undo 的持久化阶段；started 表示写入结果无法仅凭 sidecar 判断。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @param callId 记录标识
+ * @param scope 决策范围
+ * @param hunkIndex 块索引（整调用为 -1）
+ * @returns 阶段或 null
+ */
+export function revertState(cwd: string, callId: string, scope: string, hunkIndex: number): RevertState | null {
+  const row = dbFor(cwd).prepare('SELECT state FROM revert_intents WHERE call_id=? AND scope=? AND hunk_index=?')
+    .get(callId, scope, hunkIndex) as Row | undefined
+  return row ? String(row.state) as RevertState : null
+}
+
+/**
+ * 持久化 Undo 阶段（started 在写文件前，applied 在写文件后）。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @param callId 记录标识
+ * @param scope 决策范围
+ * @param hunkIndex 块索引
+ * @param state 阶段
+ */
+export function setRevertState(cwd: string, callId: string, scope: string, hunkIndex: number, state: RevertState): void {
+  dbFor(cwd).prepare(`INSERT INTO revert_intents(call_id,scope,hunk_index,state) VALUES(?,?,?,?)
+    ON CONFLICT(call_id,scope,hunk_index) DO UPDATE SET state=excluded.state`).run(callId, scope, hunkIndex, state)
+}
+
+/**
+ * 一次事务登记同记录的所有批量 Undo 意图，登记不完整时不写任何文件。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @param callId 记录标识
+ * @param indices 块索引
+ */
+export function startReverts(cwd: string, callId: string, indices: number[]): void {
+  const db = dbFor(cwd)
+  const insert = db.prepare(`INSERT INTO revert_intents(call_id,scope,hunk_index,state) VALUES(?,'hunk',?,'started')`)
+  db.exec('BEGIN IMMEDIATE')
+  try {
+    for (const index of indices) insert.run(callId, index)
+    db.exec('COMMIT')
+  } catch (error) {
+    if (db.isTransaction) db.exec('ROLLBACK')
+    throw error
+  }
+}
+
+/**
+ * 撤销尚未执行成功的 Undo 意图，保留成功写入后的阶段供补偿。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @param callId 记录标识
+ * @param scope 决策范围
+ * @param hunkIndex 块索引
+ */
+export function clearRevertState(cwd: string, callId: string, scope: string, hunkIndex: number): void {
+  dbFor(cwd).prepare('DELETE FROM revert_intents WHERE call_id=? AND scope=? AND hunk_index=?')
+    .run(callId, scope, hunkIndex)
+}
+
+/**
+ * 已执行但 sidecar 未更新的 Undo，从磁盘恢复决定而不重写文件。
+ * @author ddj 2026年09月30号
+ * @param cwd 工作区
+ * @param record 活跃记录
+ * @returns 是否应用了持久化决定
+ */
+export function restoreReverts(cwd: string, record: DiffRecord): boolean {
+  const rows = dbFor(cwd).prepare('SELECT scope,hunk_index,state FROM revert_intents WHERE call_id=?')
+    .all(record.callId) as Row[]
+  let changed = false
+  for (const row of rows) {
+    if (row.state !== 'applied') continue
+    if (row.scope === 'call') {
+      record.decisions.call = 'rejected'
+      record.decisions.perHunk = record.decisions.perHunk.map(() => 'rejected')
+    } else if (row.scope === 'hunk' && Number(row.hunk_index) >= 0 && Number(row.hunk_index) < record.hunks.length) {
+      record.decisions.perHunk[Number(row.hunk_index)] = 'rejected'
+    } else continue
+    changed = true
+  }
+  return changed
+}
+
+/**
+ * 关闭所有工作区连接，供 Host 卸载以及测试释放 SQLite 句柄。
+ * @author ddj 2026年09月30号
+ */
+export function closeArchiveDbs(): void {
+  for (const db of connections.values()) db.close()
+  connections.clear()
+}

package/src/capture.ts

@@ -240,12 +240,19 @@ async function buildRemoteRecord(cwd: string, exec: CaptureExec, args: Record<st
 async function persistRecord(ctx: Ctx, registry: Registry, session: Session, cwd: string, record: DiffRecord): Promise<void> {
   let bucket = registry.get(cwd)
   if (!bucket) { bucket = await loadBucket(ctx, cwd); registry.set(cwd, bucket) }
+  const prior = new Map(bucket)
   record.batch = fileMaxBatch(bucket, record.path) + 1
   bucket.set(record.callId, record)
   // 同一文件的多次 edit/write 都保留为独立记录，便于逐条审查。
   // 旧实现按文件 batch 自动归档 prior，会导致界面看起来每个文件只剩一条差异。
   prune(bucket)
-  await saveBucket(ctx, cwd, bucket, session)
+  const removed = [...prior.keys()].filter((callId) => !bucket.has(callId))
+  try { await saveBucket(ctx, cwd, bucket, session, { upserts: [record], removed }) }
+  catch (error) {
+    bucket.clear()
+    for (const [callId, rec] of prior) bucket.set(callId, rec)
+    throw error
+  }
 }
 
 /**

package/src/client/diffDockStore.ts

@@ -13,9 +13,22 @@ export type DiffDockMode = 'chat' | 'editor' | 'editor-empty'
  */
 export interface DiffDockSnapshot {
   mode: DiffDockMode
+  layout?: 'side' | 'tab'
   [key: string]: unknown
 }
 
+/**
+ * 判断编辑器差异栏应显示在指定布局，空态仅在尚有差异文件时显示。
+ * @author ddj 2026年09月29号
+ * @param snapshot 编辑器发布的差异快照
+ * @param layout 目标布局
+ * @returns 是否显示差异栏
+ */
+export function dockInLayout(snapshot: DiffDockSnapshot | null, layout: 'side' | 'tab'): boolean {
+  if (!snapshot || (snapshot.layout ?? 'tab') !== layout) return false
+  return snapshot.mode !== 'editor-empty' || Boolean(snapshot.fileTotal)
+}
+
 type DiffDockListener = () => void
 type DiffDockSource = object
 

package/src/client/monaco/diffRender.ts

@@ -175,6 +175,9 @@ export function createDiffRenderer(log) {
       const overlay = root && root.querySelector('.edrv-minus-overlay')
       if (overlay) overlay.innerHTML = ''
     }
+    // #region debug log
+    try { log(sessionId, '[DEBUG diffRender] regions=' + pendingRegions.length + ' zones=' + createdZones.length + ' ms=' + (Date.now() - renderT0)) } catch (e) { /* 诊断不可影响渲染 */ }
+    // #endregion
   }
 
   /** 卸载清理（EditorView unmount 时调用）。 */

package/src/client/state/regions.ts

@@ -29,6 +29,24 @@ export function countLinesBefore(text: string, index: number): number {
   return lineBefore(text, index)
 }
 
+/**
+ * 在预计算的换行位置中找出 offset 前的行数。
+ * @author ddj 2026年09月29号
+ * @param breaks 换行符位置（递增）
+ * @param offset 目标位置
+ * @returns 目标位置前的换行数
+ */
+function lineAt(breaks: number[], offset: number): number {
+  let left = 0
+  let right = breaks.length
+  while (left < right) {
+    const middle = (left + right) >>> 1
+    if (breaks[middle] < offset) left = middle + 1
+    else right = middle
+  }
+  return left
+}
+
 /**
  * 行级公共前缀/后缀裁剪：old/new 首尾相同的行视为未变化（上下文），只保留真正变更的中间段。
  * @author ddj 2026年08月20号
@@ -58,6 +76,8 @@ export function diffRegions(records: RecordView[], content: string | null): Regi
   if (content === null) return regions
   const normalized = normalizeForCompare(content)
   const lines = splitLines(normalized)
+  const breaks: number[] = []
+  for (let index = normalized.indexOf('\n'); index >= 0; index = normalized.indexOf('\n', index + 1)) breaks.push(index)
   for (const rec of records) {
     if (rec.create) {
       for (let i = 0; i < rec.hunks.length; i++) {
@@ -81,7 +101,7 @@ export function diffRegions(records: RecordView[], content: string | null): Regi
         regions.push({ callId: rec.callId, idx: entry.idx, stale: true, status, create: false, oldLines: entry.hunk.oldText === null ? [] : entry.hunk.oldText.split('\n'), newLines: entry.hunk.newText ? entry.hunk.newText.split('\n') : [], rec, superseded: rec.superseded === true })
         continue
       }
-      const start = countLinesBefore(normalized, location.start) + 1
+      const start = lineAt(breaks, location.start) + 1
       const oldLines = entry.hunk.oldText === null ? [] : entry.hunk.oldText.split('\n')
       const newLines = entry.hunk.newText.split('\n')
       const trimmed = trimCommonLines(oldLines, newLines)

package/src/client/styles/editor.css

@@ -248,6 +248,18 @@
   [data-edrv-diff-dock-shell] .edrv-diffbar-editor-content { width: max-content; }
   [data-edrv-diff-dock-shell] .edrv-diffbar-files .edrv-diffbar-file { max-width: 160px; }
 }
+/* 侧栏差异栏在编辑内容底部居中悬浮，按内容宽度收缩；窄栏只在工具条内横滚。@author ddj 2026年09月29号 */
+[data-edrv-view] .edrv-side-diff-dock[data-edrv-diff-dock-shell] { position: absolute; left: 50%; bottom: 28px; z-index: 25; width: max-content; max-width: calc(100% - 20px); max-height: min(300px, calc(100% - 38px)); margin: 0; transform: translateX(-50%); border-radius: 12px; background: var(--dsw-alias-bg-overlay, #fff); box-shadow: 0 6px 20px rgba(0,0,0,.18); overflow-x: hidden; overflow-y: auto; }
+[data-edrv-view] .edrv-side-diff-dock .edrv-diffbar-editor-row { padding-inline: 8px; }
+[data-edrv-view] .edrv-side-diff-dock .edrv-diffbar-editor-content { gap: 5px; width: max-content; }
+[data-edrv-view] .edrv-side-diff-dock .edrv-diffbar-nav,
+[data-edrv-view] .edrv-side-diff-dock .edrv-diffbar-files,
+[data-edrv-view] .edrv-side-diff-dock .edrv-diffbar-actions { gap: 3px; }
+[data-edrv-view] .edrv-side-diff-dock .edrv-pill { padding-inline: 5px; }
+[data-edrv-view] .edrv-side-diff-dock .edrv-diffbar-files .edrv-diffbar-file { min-width: 0; max-width: 72px; }
+[data-edrv-view] .edrv-side-diff-dock .edrv-diffbar-details-toggle { width: 26px; min-width: 26px; }
+[data-edrv-view] .edrv-side-diff-dock .edrv-diffbar-body { max-height: min(180px, 25vh); max-width: 100%; }
+[data-edrv-view] .edrv-side-diff-dock .edrv-diff-details { max-height: min(260px, 35vh); overflow-y: auto; }
 
 /* ===== 侧边栏（活动栏 + 面板区 + 文件树） ===== */
 [data-edrv-view] .edrv-editor-row { display: flex; flex: 1 1 auto; min-width: 0; min-height: 0; overflow: hidden; }

package/src/client/ui/ConversationDiffDock.ts

@@ -9,7 +9,7 @@ import { rpc } from '../rpc.js'
 import { openDiffView } from '../events.js'
 import { summarize } from '../state/records.js'
 import { nextDiffPath } from '../diffDock.js'
-import { readDiffDock, subscribeDiffDock } from '../diffDockStore.js'
+import { dockInLayout, readDiffDock, subscribeDiffDock } from '../diffDockStore.js'
 import { DiffBox } from './DiffBox.js'
 import { CommandPalette } from './CommandPalette.js'
 
@@ -65,7 +65,7 @@ export function ConversationDiffDock(props) {
   // 与 EditorView 同时挂载时由 commandPaletteStore 的单实例认领保证只渲染一份。
   const palette = React.createElement(CommandPalette, { key: 'edrv-palette', sessionId })
   if (editorSnapshot) {
-    if (editorSnapshot.mode === 'editor-empty' && !editorSnapshot.fileTotal) return palette
+    if (!dockInLayout(editorSnapshot, 'tab')) return palette
     return React.createElement(React.Fragment, null,
       renderDock(React.createElement(DiffBox, Object.assign({}, editorSnapshot, { dock: true })), sessionId),
       palette)