dsh-vscode-mode 0.1.1

package/src/store.ts

@@ -0,0 +1,244 @@
+/**
+ * dsh-vscode-mode host — 存储层（sidecar 读写合并 + 归档持久化 + stale 检测）。
+ * 迁移自原 src/index.ts 的 fs/ctx IO 部分，语义一字不改。
+ * 作者 ddj 2026-08-20
+ */
+import type { ArchiveBatch, ArchiveData, DiffRecord, SidecarData } from './shared/types.js'
+import { archiveEntryFor, groupByBatch, normalizeRecord } from './model.js'
+
+export const SIDECAR = '.dsh-edit-review.json'
+export const SIDECAR_ARCHIVE = '.dsh-edit-review-archive.json'
+export const READ_CAP = 8 * 1024 * 1024
+
+/** DSH 上下文与会话类型较宽松：本地无 dsh 类型声明，显式 any + 文档约束。 */
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+export type Ctx = any
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+export type Session = any
+
+/** 会话沙箱策略（policyOf）：可能不存在，调用方需容忍 undefined。 */
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+export function policyOf(ctx: Ctx, session: Session | undefined): any {
+  const svc = ctx.get('sandboxPolicy')
+  if (!svc) return undefined
+  return session ? svc.resolve({ session }) : svc.resolve()
+}
+
+/** 按会话 cwd 解析目标路径。 */
+export async function resolveTarget(ctx: Ctx, session: Session | undefined, path: string): Promise<string> {
+  const fs = ctx.get('fs')
+  const cwd = session?.header?.cwd
+  return fs.resolve(path, cwd ? { cwd } : {})
+}
+
+/** 读取工作区主 sidecar 文本（缺失/失败 → null）。 */
+async function readSidecarText(ctx: Ctx, cwd: string | null): Promise<string | null> {
+  const fs = ctx.get('fs')
+  if (!fs || !cwd) return null
+  try {
+    return await fs.readText(await fs.resolve(SIDECAR, { cwd }))
+  } catch (error) {
+    return null
+  }
+}
+
+/**
+ * 解析主 sidecar：v2 直接返回；v1 迁移为 v2 结构（按会话 cwd 分桶）。
+ * @author ddj 2026年08月20号
+ * @param text sidecar 文本
+ * @returns v2 数据或 null
+ */
+export function parseSidecar(text: string | null): SidecarData | null {
+  if (!text) return null
+  try {
+    const data = JSON.parse(text)
+    if (data && typeof data === 'object' && data.version === 2 && data.workspaces && typeof data.workspaces === 'object') {
+      return data as SidecarData
+    }
+    if (data && data.version === 1 && data.sessions && typeof data.sessions === 'object') {
+      const workspaces: SidecarData['workspaces'] = {}
+      for (const bucket of Object.values(data.sessions)) {
+        const b = bucket as { cwd?: unknown; records?: unknown; at?: unknown }
+        if (typeof b.cwd === 'string' && b.records) {
+          workspaces[b.cwd] = { at: typeof b.at === 'number' ? b.at : Date.now(), records: b.records as Record<string, DiffRecord> }
+        }
+      }
+      return { version: 2, updatedAt: typeof data.updatedAt === 'string' ? data.updatedAt : new Date().toISOString(), workspaces }
+    }
+    return null
+  } catch (error) {
+    return null
+  }
+}
+
+/** 加载某工作区的记录桶（内存 Map；缺失 → 空）。 */
+export async function loadBucket(ctx: Ctx, cwd: string): Promise<Map<string, DiffRecord>> {
+  const text = await readSidecarText(ctx, cwd)
+  const data = parseSidecar(text)
+  const map = new Map<string, DiffRecord>()
+  if (data && data.workspaces[cwd] && typeof data.workspaces[cwd].records === 'object') {
+    for (const rec of Object.values(data.workspaces[cwd].records)) {
+      const n = normalizeRecord(rec)
+      if (n) map.set(n.callId, n)
+    }
+  }
+  return map
+}
+
+/** 保存某工作区的记录桶（写前合并，不覆盖其他工作区）。 */
+export async function saveBucket(ctx: Ctx, cwd: string, recsMap: Map<string, DiffRecord>, session?: Session): Promise<void> {
+  const fs = ctx.get('fs')
+  if (!fs || !cwd) return
+  try {
+    const target = await fs.resolve(SIDECAR, { cwd })
+    const existing = parseSidecar(await readSidecarText(ctx, cwd)) ?? { version: 2, updatedAt: '', workspaces: {} } as SidecarData
+    existing.workspaces[cwd] = { at: Date.now(), records: Object.fromEntries(recsMap) }
+    existing.updatedAt = new Date().toISOString()
+    await fs.writeText(target, JSON.stringify(existing), void 0, void 0, policyOf(ctx, session))
+  } catch (error) {
+    console.error('edrv saveBucket failed', error)
+  }
+}
+
+/**
+ * 读取记录目标文件内容（按 path 缓存一次；缺失/过大/失败 → null/'' 占位）。
+ * @author ddj 2026年08月20号
+ * @param cache path → 内容（null=缺失/读取失败，''=过大跳过）
+ */
+export async function readForCache(ctx: Ctx, session: Session, cache: Map<string, string | null>, path: string): Promise<string | null> {
+  if (cache.has(path)) return cache.get(path) ?? null
+  const fs = ctx.get('fs')
+  if (!fs) { cache.set(path, null); return null }
+  try {
+    const target = await resolveTarget(ctx, session, path)
+    const info = await fs.stat(target)
+    if (!info || info.type !== 'file') { cache.set(path, null); return null }
+    if ((info.size ?? 0) > READ_CAP) { cache.set(path, ''); return '' }
+    const content = await fs.readText(target)
+    cache.set(path, content)
+    return content
+  } catch (error) {
+    cache.set(path, null)
+    return null
+  }
+}
+
+/**
+ * 判断记录是否已无任何可操作差异（stale）：待决策 hunk 的新文本在磁盘均找不到，
+ * 或新建文件已不存在，或全为空差异。满足则应由 host 自动归档。
+ * @author ddj 2026年08月20号
+ */
+export async function recordIsStale(ctx: Ctx, session: Session, cache: Map<string, string | null>, rec: DiffRecord): Promise<boolean> {
+  if (rec.superseded === true) return true
+  if (!Array.isArray(rec.hunks) || !rec.hunks.length) return true
+  const perHunk = Array.isArray(rec.decisions.perHunk) ? rec.decisions.perHunk : []
+  let hasLocatablePending = false
+  for (let i = 0; i < rec.hunks.length; i++) {
+    const st = perHunk.length ? perHunk[i] : rec.decisions.call
+    if (st === 'accepted' || st === 'rejected') continue
+    if (rec.create === true) {
+      const content = await readForCache(ctx, session, cache, rec.path)
+      if (content !== null) { hasLocatablePending = true; break }
+      continue
+    }
+    const precise = rec.toolName === 'edit' && rec.callHunk ? rec.callHunk : rec.hunks[i]
+    const oldText = precise?.oldText ?? null
+    const newText = precise?.newText ?? null
+    if (oldText !== null && oldText === newText) continue
+    if (newText === null || newText === undefined) continue
+    const content = await readForCache(ctx, session, cache, rec.path)
+    if (content !== null && content.indexOf(newText) >= 0) { hasLocatablePending = true; break }
+  }
+  return !hasLocatablePending
+}
+
+/**
+ * edrv.list 全量轮询时自动清理 stale 记录：标记 superseded 并归档，
+ * 防止"幽灵差异"长期留在审查列表且无法操作。
+ * @author ddj 2026年08月20号
+ * @returns 自动归档条数
+ */
+export async function autoArchiveStale(ctx: Ctx, session: Session, cwd: string, bucket: Map<string, DiffRecord>): Promise<number> {
+  const fs = ctx.get('fs')
+  if (!fs) return 0
+  const cache = new Map<string, string | null>()
+  const stale: DiffRecord[] = []
+  for (const rec of bucket.values()) {
+    if (rec.archived || rec.superseded === true) continue
+    if (await recordIsStale(ctx, session, cache, rec)) stale.push(rec)
+  }
+  if (!stale.length) return 0
+  for (const r of stale) { r.superseded = true; r.at = new Date().toISOString() }
+  await archiveRecords(ctx, session, cwd, bucket, stale, '差异无法定位（已被后续修改覆盖），自动归档')
+  return stale.length
+}
+
+/** 读取归档 sidecar 文本（缺失/失败 → null）。 */
+export async function readArchiveText(ctx: Ctx, cwd: string | null): Promise<string | null> {
+  const fs = ctx.get('fs')
+  if (!fs || !cwd) return null
+  try {
+    return await fs.readText(await fs.resolve(SIDECAR_ARCHIVE, { cwd }))
+  } catch (error) {
+    return null
+  }
+}
+
+/** 解析归档 sidecar（损坏 → 空数组）。 */
+export function parseArchive(text: string | null): ArchiveBatch[] {
+  if (!text) return []
+  try {
+    const data = JSON.parse(text)
+    if (data && typeof data === 'object' && Array.isArray(data.batches)) return data.batches as ArchiveBatch[]
+  } catch (error) { /* 忽略损坏 */ }
+  return []
+}
+
+/** 追加归档条目（按 cwd+path+batch 合并，同 callId 去重）。 */
+export async function appendArchiveEntries(ctx: Ctx, cwd: string, entries: Array<ReturnType<typeof archiveEntryFor>>, session?: Session): Promise<void> {
+  const fs = ctx.get('fs')
+  if (!fs || !cwd) return
+  try {
+    const target = await fs.resolve(SIDECAR_ARCHIVE, { cwd })
+    const existing = parseArchive(await readArchiveText(ctx, cwd))
+    for (const e of entries) {
+      const idx = existing.findIndex((x) => x.cwd === e.cwd && x.path === e.path && x.batch === e.batch)
+      if (idx >= 0) {
+        const exist = existing[idx]
+        const seen = new Set((exist.records || []).map((r) => r.callId))
+        for (const rec of e.records) {
+          if (!seen.has(String(rec.callId))) {
+            exist.records.push(rec as unknown as ArchiveBatch['records'][number])
+            seen.add(String(rec.callId))
+          }
+        }
+        exist.lastAt = e.at
+        exist.reason = e.reason
+      } else {
+        existing.push(e as unknown as ArchiveBatch)
+      }
+    }
+    const data: ArchiveData = { version: 1, updatedAt: new Date().toISOString(), batches: existing }
+    await fs.writeText(target, JSON.stringify(data), void 0, void 0, policyOf(ctx, session))
+  } catch (error) {
+    console.error('edrv appendArchiveEntries failed', error)
+  }
+}
+
+/** 归档记录：标记 archived、按批次写入归档、落盘工作区桶。 */
+export async function archiveRecords(
+  ctx: Ctx,
+  session: Session,
+  cwd: string,
+  bucket: Map<string, DiffRecord>,
+  recs: DiffRecord[],
+  reason: string,
+): Promise<void> {
+  if (!recs.length) return
+  const fresh = recs.filter((r) => !r.archived)
+  for (const r of recs) r.archived = true
+  const entries: Array<ReturnType<typeof archiveEntryFor>> = []
+  for (const list of groupByBatch(recs).values()) entries.push(archiveEntryFor(list, cwd, reason))
+  await appendArchiveEntries(ctx, cwd, entries, session)
+  if (fresh.length) await saveBucket(ctx, cwd, bucket, session)
+}

package/src/workspace.ts

@@ -0,0 +1,74 @@
+/**
+ * dsh-vscode-mode host — 工作区文件清单扫描与缓存（快速打开/搜索用）。
+ * 迁移自原 src/index.ts 的 listWorkspaceFiles，语义不改。
+ * 作者 ddj 2026-08-20
+ */
+import type { Ctx, Session } from './store.js'
+import { policyOf } from './store.js'
+
+/** 文件清单缓存 TTL 与扫描上限（避免大工作区反复全量扫描）。 */
+const FILE_INDEX_TTL = 60_000
+const SCAN_CAP = 6000
+/** cwd → { at, files }（随会话销毁清理）。 */
+const fileIndex = new Map<string, { at: number; files: string[] }>()
+
+/**
+ * 扫描工作区文件清单（快速打开/搜索用）：subprocess（Windows powershell / POSIX find），
+ * 排除常见重目录，容量上限 SCAN_CAP；结果缓存（TTL FILE_INDEX_TTL）。失败返回 null。
+ * @author ddj 2026年08月20号
+ * @param ctx DSH 上下文
+ * @param session 会话（取工作区根）
+ * @param cwd 工作区目录
+ * @returns 绝对路径清单；不可用（无 subprocess/fs）或扫描失败返回 null
+ */
+export async function listWorkspaceFiles(ctx: Ctx, session: Session, cwd: string): Promise<string[] | null> {
+  const cached = fileIndex.get(cwd)
+  if (cached && Date.now() - cached.at < FILE_INDEX_TTL) return cached.files
+  const sub = ctx.get('subprocess')
+  const fs = ctx.get('fs')
+  if (!sub || !fs) return null
+  const policy = policyOf(ctx, session)
+  let root: string
+  try {
+    const rootTarget = await fs.resolve(policy?.workspaceRoot ?? '.', {})
+    root = fs.processPath(rootTarget)
+  } catch (error) {
+    root = cwd
+  }
+  const lines: string[] = []
+  const run = async (argv: string[]): Promise<void> => {
+    const handle = sub.spawn({
+      argv,
+      stdio: { stdout: { maxBytes: 8 << 20 }, stderr: { maxBytes: 1 << 16 }, stdin: 'ignore' },
+      graceMs: 20000,
+    })
+    const outcome = await handle.done
+    const code = outcome?.exitCode ?? outcome?.code
+    if (code !== 0) throw new Error('scan exit ' + code)
+    const text = handle.collected?.stdout?.readFrom(0).text ?? ''
+    for (const ln of text.split(/\r?\n/)) {
+      const s = ln.trim()
+      if (s) lines.push(s)
+    }
+  }
+  try {
+    if (process.platform === 'win32') {
+      const cmd = 'Get-ChildItem -LiteralPath "' + String(root).replace(/"/g, '""') + '" -Recurse -File -ErrorAction SilentlyContinue | Where-Object { $_.FullName -notmatch \'(node_modules|\\.git|\\.tmp|\\.cache|dist|build|vendor|coverage|__pycache__)\' } | Select-Object -First ' + SCAN_CAP + ' | ForEach-Object { $_.FullName }'
+      await run(['powershell', '-NoProfile', '-NonInteractive', '-Command', cmd])
+    } else {
+      const cmd = 'find ' + JSON.stringify(root) + ' -type f -not -path \'*/node_modules/*\' -not -path \'*/.git/*\' -not -path \'*/.tmp/*\' -not -path \'*/.cache/*\' -not -path \'*/dist/*\' -not -path \'*/build/*\' -not -path \'*/vendor/*\' -not -path \'*/coverage/*\' 2>/dev/null | head -n ' + SCAN_CAP
+      await run(['/bin/sh', '-c', cmd])
+    }
+  } catch (error) {
+    console.error('edrv scan failed', error)
+    return null
+  }
+  const files = lines.slice(0, SCAN_CAP)
+  if (files.length) fileIndex.set(cwd, { at: Date.now(), files })
+  return files
+}
+
+/** 会话销毁时清理文件索引缓存。 */
+export function dropFileIndex(cwd: string): void {
+  fileIndex.delete(cwd)
+}