@a9i5k4/dsh-auto-memory 2.1.7 → 2.1.9

package/lib/client.js

@@ -1223,6 +1223,17 @@ window.__ModuleLoader__.load({
       var pollPair = useState(null)
       var poll = pollPair[0]
       var setPoll = pollPair[1]
+      var gpuPair = useState(function () { try { return localStorage.getItem('dsh-auto-memory.pyGpu') === '1' } catch (eG) { return false } })
+      var gpuPref = gpuPair[0]
+      var setGpuPref = function (v) {
+        gpuPair[1](v)
+        try { localStorage.setItem('dsh-auto-memory.pyGpu', v ? '1' : '0') } catch (eG) {}
+      }
+      var actGpu = function () {
+        if (busy) return
+        setBusy(true)
+        apiPost(API.pyDeps, { gpu: gpuPref }).then(function (d) { setData(d); setBusy(false); setPoll(true); var ph = d && d.phase; if (ph === 'ready' || ph === 'error' || ph === 'idle') setPoll(false) }).catch(function () { setBusy(false); setPoll(false) })
+      }
       var refresh = function () { apiGet(API.pyStatus).then(function (d) { if (d) setData(d) }).catch(function () {}) }
       useEffect(function () {
         var alive = true
@@ -1267,7 +1278,11 @@ window.__ModuleLoader__.load({
         stepRow(t('pyWizStep2'), data.venvOk ? true : 'run',
           (!data.venvOk && (data.pythons || []).some(function (p2) { return p2.status === 'ok' && !p2.isVenv })) ? h('button', Object.assign({}, btnCls, { onClick: function () { act(API.pyVenv, true) } }), t('pyWizCreate')) : null),
         stepRow(t('pyWizStep3'), data.depsOk ? true : 'run',
-          (data.venvOk && !data.depsOk) ? h('button', Object.assign({}, btnCls, { onClick: function () { act(API.pyDeps, true) } }), t('pyWizInstall')) : null),
+          (data.venvOk && !data.depsOk) ? h('span', { style: { display: 'flex', gap: '8px', alignItems: 'center' } },
+            h('label', { style: { display: 'flex', gap: '4px', alignItems: 'center', fontSize: 'calc(10.5px * var(--dam-scale))' } },
+              h('input', { type: 'checkbox', checked: gpuPref, onChange: function (e) { setGpuPref(e.target.checked) } }),
+              locale === 'zh' ? 'GPU 推理(NVIDIA,需已装 CUDA;否则自动回退 CPU)' : 'GPU inference (NVIDIA + CUDA; auto-falls back to CPU)'),
+            h('button', Object.assign({}, btnCls, { onClick: function () { actGpu() } }), t('pyWizInstall'))) : null),
         stepRow(t('pyWizStep4'), data.modelReady ? true : 'run',
           (!data.modelReady && !dlActive) ? h('button', Object.assign({}, btnCls, { onClick: function () { act(API.pyModel, true) } }), t('pyWizDownload')) : null,
           dlActive ? h('button', { 'data-dam-btn': '', onClick: function () { apiPost(API.pyCancel, {}).then(refresh).catch(function () {}) } }, t('pyWizCancel')) : null),
@@ -3047,6 +3062,7 @@ window.__ModuleLoader__.load({
               : 'The built-in JS semantic model (~135MB, one-time) can be downloaded in the background; the semantic tier enables itself once done:'))
             sdActions.push(h('button', { 'data-dam-btn': '', 'data-primary': 'true', style: close, onClick: function () {
               apiPost(API.semanticDownload, { action: 'start', mirror: 'auto' }).catch(function () {})
+              try { localStorage.setItem('dsh-auto-memory.semDetectSnoozeUntil', String(Date.now() + 24 * 3600 * 1000)) } catch (eSn2) {}
               closeDialog()
             } }, sdZh ? '立即下载' : 'Download now'))
           }
@@ -3421,7 +3437,7 @@ window.__ModuleLoader__.load({
                   h('option', { value: 'auto' }, t('mAuto')),
                   h('option', { value: 'cn' }, t('mCn')),
                   h('option', { value: 'intl' }, t('mIntl'))) : null,
-                ready ? h('button', { 'data-dam-btn': '', onClick: function () { setGuide(''); var n2 = Object.assign({}, cfg); n2.semanticEngineMode = guide; if (guide === 'js') { n2.activationSource = 'js'; n2.contextSinkMode = 'null' } else if (guide === 'python') { n2.activationSource = 'python'; n2.contextSinkMode = 'python' } setCfg(n2); setDirty(true) } }, locale === 'zh' ? '启用并继续' : 'Enable & continue') : null,
+                ready ? h('button', { 'data-dam-btn': '', onClick: function () { setGuide(''); var n2 = Object.assign({}, cfg); n2.semanticEngineMode = guide; if (guide === 'js') { n2.activationSource = 'js'; n2.contextSinkMode = 'null' } else if (guide === 'python') { n2.activationSource = 'python'; n2.contextSinkMode = 'python'; try { var gp = localStorage.getItem('dsh-auto-memory.pyGpu'); if (gp === '1') n2.pythonGpu = true } catch (eL) {} } setCfg(n2); setDirty(true) } }, locale === 'zh' ? '启用并继续' : 'Enable & continue') : null,
                 !ready && isJs && !dlActive ? h('button', { 'data-dam-btn': '', onClick: startDl }, (dl.phase === 'error' || dl.phase === 'cancelled') ? t('semDlRetry') : t('semDlStart')) : null,
                 !ready && isJs && dlActive ? h('button', { 'data-dam-btn': '', onClick: cancelDl }, t('semDlCancel')) : null,
                 h('button', { 'data-dam-btn': '', onClick: function () { setGuide('') } }, t('gotIt'))))

package/lib/index.js

@@ -2335,6 +2335,78 @@ class MemoryEngine {
     return true
   }
 
+  /**
+   * F1(M6-CONTRACT 外补丁,2026-08-23):anchor 开启时的记录级压缩。
+   * 按 sidecar 记录序整条保留/移除——今天(标题日期=插件今天)记录无条件保留,其余按字节配额
+   * 从最新往回保留(layer 上限 user 4000/note 3000);被移除记录整条原文写入归档文件
+   * (writeFullRaw,不经 anchor 事务,归档非检索语料)。重组经 store.replace 原子执行:
+   * 候选预检(clean)→被移除 id 自动删除→保留 id/版本不变。永不字符切片。
+   * @returns {Promise<boolean|'not-anchored'>} true=已腾位 false=无法腾位 'not-anchored'=无 sidecar 走旧路径
+   */
+  async compactAnchoredLayer(store, filePath, layer, today, paths) {
+    const cur = await this.readTextSafe(filePath)
+    if (!cur || !cur.trim()) return false
+    let parsed
+    try { parsed = parseAnchors(Buffer.from(cur, 'utf8')) } catch (_) { return 'not-anchored' }
+    if (!parsed || parsed.status !== 'clean') return false // 脏文件 fail closed:不压缩、不切片
+    const anchored = parsed.records.filter((r) => r.kind === 'anchored')
+    if (!anchored.length) return 'not-anchored'
+    const limit = layer === 'user' ? 4000 : 3000
+    const isTodayHeading = (heading) => /^\d{4}-\d{2}-\d{2}/.test(String(heading || '')) && String(heading).slice(0, 10) === today
+    const buf = Buffer.from(cur, 'utf8')
+    // 配额从尾回溯(F1):新的留;今天记录无条件保留
+    const keep = new Array(anchored.length).fill(false)
+    let keptBytes = 0
+    for (let i = anchored.length - 1; i >= 0; i--) {
+      const rec = anchored[i]
+      const spanBytes = rec.byteEnd - rec.markerByteStart
+      if (isTodayHeading(rec.heading) || keptBytes + spanBytes <= limit) { keep[i] = true; keptBytes += spanBytes }
+    }
+    const removedIdx = []
+    anchored.forEach((rec, i) => { if (!keep[i]) removedIdx.push(i) })
+    if (!removedIdx.length) return false // 无可移除 → 无法腾位(预算照常拒绝)
+    // 有序块重组:legacy 块与 preamble/tail 原样保留,gap 字节逐段拷贝,仅丢弃被移除记录的 [marker..contentEnd]
+    const blocks = []
+    for (const r of parsed.records) {
+      if (r.kind === 'anchored') blocks.push({ start: r.markerByteStart, end: r.byteEnd, keep: keep[anchored.indexOf(r)] })
+      else blocks.push({ start: r.byteStart, end: r.byteEnd, keep: true })
+    }
+    blocks.sort((a, b) => a.start - b.start)
+    const parts = []
+    let cursorB = 0
+    for (let i = 0; i < blocks.length; i++) {
+      const b = blocks[i]
+      if (b.start > cursorB) parts.push(buf.slice(cursorB, b.start).toString('utf8'))
+      if (b.keep) parts.push(buf.slice(b.start, b.end).toString('utf8'))
+      cursorB = b.end
+    }
+    if (cursorB < buf.length) parts.push(buf.slice(cursorB).toString('utf8'))
+    const newText = parts.join('')
+    // 归档:整条原文(writeFullRaw 绕开 anchor 事务;归档文件从不参与解析/索引)
+    const archiveFile = layer === 'user'
+      ? path.join(dshHome(), 'memory', 'archived-user.md')
+      : path.join(paths.projectDir, 'archive', 'notes-archived.md')
+    const archiveText = removedIdx.map((i2) => {
+      const r = anchored[i2]
+      return buf.slice(r.markerByteStart, r.byteEnd).toString('utf8')
+    }).join('\n')
+    const existingArchive = await this.readTextSafe(archiveFile)
+    await this.writeFullRaw(archiveFile, (existingArchive ? existingArchive.replace(/\s+$/, '') + '\n' : '') + archiveText + '\n')
+    const rr = await store.replace(filePath, newText)
+    if (!rr.ok) { console.error('[compacted-anchored] replace failed: ' + rr.reason); return false }
+    const freshText = await this.readTextSafe(filePath)
+    if (layer === 'user') this.state.userText = freshText
+    else this.state.notesText = freshText
+    if (this._budgets) {
+      const b = this._budgets.get('today')
+      if (b && b.date === this.memToday()) b[layer] = 0
+    }
+    this._lastCompactAt = Date.now()
+    this.state.loadedAt = Date.now()
+    console.log('[compacted-anchored] ' + layer + ': kept=' + (anchored.length - removedIdx.length) + '/' + anchored.length + ' archived=' + removedIdx.length + ' bytes=' + cur.length + '->' + newText.length)
+    return true
+  }
+
   // ---------- 注入渲染(同步,基于缓存) ----------
   // 动态记忆 → ctx.systemPrompt.context()(user-role 快照):内容变化才追加新快照,内容不变不重复注入(dsh-agent-loop project() 去重),
   // system prompt 不再包含动态内容 → 字节级稳定 → DeepSeek 前缀缓存全程命中(对比 section 方案:动态内容任何变化都从变化点起击穿整个前缀,含全部历史)
@@ -2381,6 +2453,17 @@ class MemoryEngine {
     if (cfg.waterLevelAdvisory !== false && !this.isUnattendedNow() && (s.waterLevelRatio || 0) >= (Number(cfg.waterLevelThreshold) || 0.8) && Date.now() - (s.waterLevelAt || 0) < 600000) {
       lines.push('\n' + L('snapshotWaterTitle') + '\n' + L('snapshotWaterBody', { pct: Math.round((s.waterLevelRatio || 0) * 100) }))
     }
+    // M-CM1 交接白板:PLAN.md 全貌+最新交接账本,动态快照首位(动态层内容,不碰静态字节;敏感段清洗+行边界硬预算截断)
+    if (cfg.handoffEnabled !== false && (s.planText || s.latestHandoffText)) {
+      const planPart = truncateLinesBounded(stripSensitiveSections(sanitizeForInjection(s.planText || '')), Math.max(Number(cfg.handoffPlanChars) || 1200, 200))
+      if (planPart) lines.push('\n' + L('snapshotPlanTitle') + '\n' + planPart)
+      const ledgerPart = truncateLinesBounded(stripSensitiveSections(sanitizeForInjection(s.latestHandoffText || '')), Math.max(Number(cfg.handoffLedgerChars) || 800, 200))
+      if (ledgerPart) lines.push('\n' + L('snapshotHandoffTitle') + '\n' + ledgerPart)
+    }
+    // M-CM4/M-CM5 水位建议:越阈注入交接+派子代理建议(advisory;10 分钟新鲜度;无人值守静默)
+    if (cfg.waterLevelAdvisory !== false && !this.isUnattendedNow() && (s.waterLevelRatio || 0) >= (Number(cfg.waterLevelThreshold) || 0.8) && Date.now() - (s.waterLevelAt || 0) < 600000) {
+      lines.push('\n' + L('snapshotWaterTitle') + '\n' + L('snapshotWaterBody', { pct: Math.round((s.waterLevelRatio || 0) * 100) }))
+    }
     let used = 0
     const part = (title, text, max) => {
       if (!text) return
@@ -4829,7 +4912,10 @@ export function apply(ctx, config) {
   {
     const pluginDir = path.dirname(fileURLToPath(import.meta.url))
     engine._jsSemantic = createJsSemanticEnginePre({ pluginDir })
-    engine._jsDownload = createSemanticDownloaderPre({ modelsRoot: path.join(pluginDir, 'models') })
+    // #15 后续/B:JS 模型下载落位用户目录(~/.dsh/models/js-semantic/)——包目录在 npm 更新时被重装,130MB 曾被冲掉
+    const userModelsRoot = path.join((process.env.DSH_HOME && process.env.DSH_HOME.trim()) || path.join(homedir(), '.dsh'), 'models', 'js-semantic')
+    mkdirSync(userModelsRoot, { recursive: true })
+    engine._jsDownload = createSemanticDownloaderPre({ modelsRoot: userModelsRoot })
     // 资产探测(semantic-status / 档位解析 / 引导卡共用):共享实现见 semantic-js-pre 的
     // probeJsSemanticAssets —— peer 探测与引擎加载走同一套 Node 解析(2026-09-02 issue 修正)。
     // _peerExtraDirs = 深度扫描热接入位(semanticDeepDetect 命中后 probe/加载/档位即时生效)。
@@ -5458,7 +5544,8 @@ export function apply(ctx, config) {
       handler: async (req, res) => {
         if (!isLoopbackRequest(req)) return writeJson(res, 403, { error: 'forbidden: loopback-only' })
         if ((req.method || 'POST') !== 'POST') return writeJson(res, 405, { error: 'method not allowed' })
-        try { writeJson(res, 200, await engine._pythonSetup.ensureDeps()) } catch (e) { writeJson(res, 500, { error: String(e && e.message || e) }) }
+        const body = await readJsonBody(req).catch(() => ({}))
+        try { writeJson(res, 200, await engine._pythonSetup.ensureDeps({ gpu: !!(body && body.gpu) })) } catch (e) { writeJson(res, 500, { error: String(e && e.message || e) }) }
       },
     },
     {

package/lib/python-setup.js

@@ -35,8 +35,9 @@ const MODEL_SPEC = {
   ],
 }
 
-/** venv 内 pip 依赖:与 bench 校准环境同源(transformers AutoTokenizer + onnxruntime;torch 为张量构造所需;池化/精度契约冻结自 R@5 0.925 基线,fastembed 注册表无 BGE-M3 且池化契约不同,不可用)。 */
-const PIP_DEPS = ['transformers', 'onnxruntime', 'torch']
+/** venv 内 pip 依赖:#19 实测(int8 档 encode_ids 全程 numpy+onnxruntime,tokenizer 走 Rust 快速路径)——基础集无 torch,venv 体积 ~400MB。GPU 推理开关追加 onnxruntime-gpu(CUDAExecutionProvider)。池化/精度契约冻结自 R@5 0.925 基线,fastembed 注册表无 BGE-M3 且池化契约不同,不可用。 */
+const PIP_DEPS_CPU = ['transformers', 'onnxruntime']
+const PIP_DEPS_GPU_EXTRA = ['onnxruntime-gpu']
 
 export function createPythonSetupPre(opts = {}) {
   const dshHomeOf = typeof opts.dshHome === 'function' ? opts.dshHome : () => opts.dshHome || path.join(homedir(), '.dsh')
@@ -110,7 +111,7 @@ export function createPythonSetupPre(opts = {}) {
     return {
       phase: st.phase, error: st.error,
       pythons: st.pythons, chosenPython: st.chosenPython,
-      venvOk: st.venvOk, depsOk: st.depsOk, modelReady: st.modelReady,
+      venvOk: st.venvOk, depsOk: st.depsOk, modelReady: st.modelReady, wantGpu: !!st.wantGpu,
       modelPath: modelPath(), venvPython: venvPython(),
       modelBytes: st.modelReady ? statSync(modelPath()).size : 0, modelExpectedBytes: MODEL_SPEC.bytes,
       dl: { ...st.dl },
@@ -136,14 +137,17 @@ export function createPythonSetupPre(opts = {}) {
     return snapshot()
   }
 
-  /** ③deps:venv 内 pip 装 transformers+onnxruntime+torch;官方源失败切清华镜像。 */
-  async function ensureDeps() {
+  /** ③deps:venv 内 pip 装 transformers+onnxruntime(CPU 基础集);opts.gpu=true 追加 onnxruntime-gpu(用户选 GPU 推理时)。官方源失败切清华镜像。 */
+  async function ensureDeps(opts2) {
+    const wantGpu = !!(opts2 && opts2.gpu)
     if (!st.venvOk) { st.error = 'venv 未就绪,先执行 venv 步骤'; st.phase = 'error'; return snapshot() }
     st.phase = 'deps'
+    st.wantGpu = wantGpu
+    const deps = PIP_DEPS_CPU.concat(wantGpu ? PIP_DEPS_GPU_EXTRA : [])
     const runPip = async (extra) => {
       try {
-        const r = await execFileP(venvPython(), ['-m', 'pip', 'install', '--quiet'].concat(PIP_DEPS).concat(extra || []),
-          { timeout: 600000, windowsHide: true, maxBuffer: 4 * 1024 * 1024 })
+        const r = await execFileP(venvPython(), ['-m', 'pip', 'install', '--quiet'].concat(deps).concat(extra || []),
+          { timeout: 900000, windowsHide: true, maxBuffer: 4 * 1024 * 1024 })
         return { ok: true }
       } catch (e) { return { ok: false, tail: String((e && (e.stderr || e.message)) || e).slice(-300) } }
     }
@@ -151,7 +155,18 @@ export function createPythonSetupPre(opts = {}) {
     if (!r.ok) r = await runPip(['-i', 'https://pypi.tuna.tsinghua.edu.cn/simple'])
     st.depsOk = !!r.ok
     if (!r.ok) { st.error = '依赖安装失败: ' + (r.tail || ''); st.phase = 'error'; return snapshot() }
-    diagOf('python-setup: deps installed into venv')
+    // GPU 偏好写入 embedding-config.json(worker 读 config['gpu'] 选 CUDA/CPU provider;read-modify-write 不覆盖既有键)
+    try {
+      const cfgDir = path.join(dshHomeOf(), 'memory', 'semantic')
+      mkdirSync(cfgDir, { recursive: true })
+      const cfgPath = path.join(cfgDir, 'embedding-config.json')
+      let cfg = {}
+      try { cfg = JSON.parse(await readFile(cfgPath, 'utf8')) } catch (_) {}
+      if (!cfg || typeof cfg !== 'object') cfg = {}
+      cfg.gpu = wantGpu
+      await writeFile(cfgPath, JSON.stringify(cfg, null, 2), 'utf8')
+    } catch (eCfg) { diagOf('python-setup: gpu pref write failed: ' + String(eCfg && eCfg.message || eCfg)) }
+    diagOf('python-setup: deps installed into venv (gpu=' + wantGpu + ')')
     return snapshot()
   }