Skip to content

SmartPerfetto v1.15.1

Latest
Latest

Choose a tag to compare

View all tags
@Gracker Gracker released this 06 Oct 10:09
v1.15.1
d963668

SmartPerfetto v1.15.1

中文

1.15.1 继承 1.15 的功能更新，并补齐安装诊断和发布验收。doctor 现在真实打开、查询并关闭内存 SQLite，AI 关闭时也会检查原生绑定；HTTP 测试在导入运行时前隔离私有数据目录并完成资源清理。npm 公开安装验收由消费者项目明确批准经过源码审查的依赖脚本 pins，再查询固定 Trace，不能用 version/help 成功替代真实执行。

本版把源码与文档知识库选择放到每轮输入框旁，统一五个运行时的最终交付与核验，补齐进程实例隔离、呈现节奏和 CPU 簇负载测量。

  • 本轮上下文：源码按需查找与读取，无需预建完整索引；源码深度控制预算，发送授权仍以披露范围为准。文档文件夹可以作为独立知识库，Web 和 CLI 都支持管理与授权。
  • 答案与停止：支持的运行时先显示临时草稿，正文交付后继续核验；草稿不保存。Web 正文交付后的第一次停止只结束复核，第二次进入有界强停。CLI 第二次 Ctrl-C 会取消整轮且不保存；JSON/NDJSON 的第一次 Ctrl-C 即取消整轮。
  • 分析语义：呈现节奏与帧耗时并列，保留 Buffer Stuffing 和候选上屏关联的证据限制；点击、CPU、启动详情支持 exact UPID 与重启实例隔离探针。新增堆转储类增长、进程状态、后台电源状态和持锁方区间。
  • 本次修复：CPU 簇负载的分子、分母统一扣除休眠；转换缺失时整体使用墙钟。多设备同号 CPU 或歧义元数据返回不可用原因。完整休眠窗口不计活跃核心。Linux gate 安装真实 ripgrep，源码选择测试使用真实文件；帧方差阈值测试通过生产依赖链执行，保留语义断言与完整滑动覆盖。

HTTP 测试改用自有 IPv4 listener；会话、健康检查和路由集成使用专属数据库与日志目录，避免测试触碰默认用户数据。

源码部署安全修复：开发模式的 /uploads 静态入口已移除，Trace 下载统一经过 API 认证和归属检查。默认 Docker 与便携包配置原本不暴露这一入口。

升级说明

旧内置 Android Internals Wiki Pack 与 smp knowledge-pack 已移除，请注册普通文档知识库。/api/perfetto-sql/*、/api/template-analysis/*、/api/sql/* 返回 410，部分入口无直接替代；/api/perfetto/* 返回 404。模型调用类 Skill 步骤与旧 Claude verifier timeout/verification toggle 已移除。

CLI 要求 Node.js 24，便携包自带运行时。本轮启用源码或文档知识库时，报告与分析快照仅创建者可读。

已知限制

真实 Provider 验收仍不完整：本次 DeepSeek 的 OpenAI/Pi 启动报告未通过声明或报告检查，OpenAI 滑动报告未通过声明事实验收。OpenCode 启动实测返回了声明但没有可用正文，最终核验拒绝交付；现有材料无法确定正文补全的具体失败分支。确定性测试不能视作所有运行时的真实长报告或准确性验收。

English

1.15.1 retains the 1.15 feature update and adds installation diagnostics and release verification. doctor opens, queries and closes an in-memory SQLite database even with AI disabled. HTTP fixtures isolate private data before runtime imports and settle owned resources before cleanup. The public npm smoke explicitly approves source-reviewed script pins in its consumer project and queries a committed trace; version/help alone does not establish executable package health.

This release adds per-turn source and document context beside the composer, unifies final delivery across five runtimes, and improves process-instance isolation, presentation cadence and CPU cluster measurements.

  • Context: search and read source on demand without a complete index; source depth controls budgets, while disclosed provider-send consent controls authorization. Ordinary document folders are independently selectable knowledge bases with Web and CLI management.
  • Delivery: supported runtimes show live-only drafts, then a readable answer while review continues. The first Web stop after delivery ends review; the second uses the bounded force-stop path. CLI's second Ctrl-C aborts without saving; JSON/NDJSON abort on the first Ctrl-C.
  • Evidence: presentation cadence remains distinct from frame duration and speculative response associations. Click, CPU and startup detail support exact UPID, same-package and restart isolation probes. Heap class growth, process state, background power state and lock-holder evidence retain observation limits.
  • Corrections: CPU Running spans and window capacity use one clock basis, excluding suspend consistently. Missing conversions cause a common wall-clock fallback. Ambiguous CPU identity produces null metrics and a reason; a fully suspended window has zero active cores. Linux CI installs ripgrep; source-selection success tests create real source files. The threshold regression uses its production dependency chain without relaxing assertions or removing full scrolling coverage.

Source deployment security: the development-only /uploads static route is removed. Trace downloads use authenticated, ownership-checked APIs. Default Docker and portable configurations were not exposed through this route.

HTTP tests use owned IPv4 listeners; session, health and route integration fixtures isolate their databases and logs from default user data.

Migration and limits

The built-in Wiki Pack, smp knowledge-pack, model-calling Skill steps and old Claude verifier options are removed. Retired SQL/template APIs return 410; /api/perfetto/* returns 404. Consult each route's migration information. The npm CLI requires Node.js 24; portable packages bundle their runtime. Enabled source/document context makes reports and snapshots creator-only.

Real-provider acceptance remains incomplete. DeepSeek OpenAI/Pi startup runs failed declaration or report checks; OpenAI scrolling did not pass declared-fact acceptance. OpenCode startup returned a declaration without a usable body and was rejected. The saved evidence does not identify its continuation failure branch. Deterministic tests do not establish successful real-provider long-report acceptance across all runtimes.

npm install -g @gracker/smartperfetto@1.15.1
smp --version
smp doctor --format json

If npm reports blocked or skipped dependency scripts, review the scoped approval and rebuild instructions in the CLI installation guide. A publisher's allowScripts does not grant consumer approval.

Full changes: CHANGELOG.md.

Assets 5
Loading