Repository navigation
·
10 commits
to master
since this release
[Version: v2.0.0-alpha.6] - 2026-11-17
Full Changelog: v2.0.0-alpha.5...v2.0.0-alpha.6
发布范围
- ACP 2.0 Alpha 性能维护更新,减少请求前 token usage 估算中的重复工作。
功能优化
- 复用 tokenizer 实例,避免每次计数都重新初始化和释放 WASM tokenizer;保留原有文本归一化、特殊 token 处理及异常回退行为。
- 在 batch cleanup 与裁剪前统计之间共享同一消息投影的 token 计数,避免重复扫描;裁剪后的统计仍重新计算。
兼容性与升级前置条件
- 要求 OpenCode >=2.1.4 且 <3。
- 不改变配置、持久化状态格式、message refs 或 compression block IDs;从 alpha.5 升级无需进行这些数据的迁移。
- 不引入文本内容缓存,不改变历史 provider usage 的处理方式。
已知问题
- 本版针对 token 计数中的重复工作进行优化,不代表长历史会话的整体等待问题或图片压缩问题已全部解决,见 #51。
[Version: v2.0.0-alpha.6] - 2026-11-17
Full Changelog: v2.0.0-alpha.5...v2.0.0-alpha.6
Release scope
- An ACP 2.0 Alpha performance maintenance update reducing repeated work in pre-request token usage estimation.
Improvements
- Reuse the tokenizer instance instead of initializing and freeing the WASM tokenizer for every count, preserving text normalization, special-token handling, and fallback behavior.
- Share the token count for the same message projection between batch cleanup and pre-pruning accounting to avoid duplicate scans. Post-pruning accounting is still recomputed.
Compatibility and upgrade prerequisites
- Requires OpenCode >=2.1.4 and <3.
- No changes to configuration, persisted state formats, message refs, or compression block IDs. Upgrading from alpha.5 requires no migration of these data.
- No text-content cache is introduced, and historical provider usage handling is unchanged.
Known issues
- This release targets repeated token-counting work. It does not resolve all long-history delays or image-compression issues tracked in #51.