Repository navigation
research(core): 保留历史内存与 H10000 Search 成本 (#164) - #176
Conversation
…164 Adds an EXCLUDE_FROM_ALL probe that reports working set, private commit, fill time and a full-document search at H=100/1000/10000 on the production Screen and TextSearch, with fixed results and evidence identity.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3ee3baf189
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
|
||
| - 这是**生产** `Screen`/`TextSearch` 的观测，不是 r164 fork 的 Scan/LazyTable 对照；两者不可直接比较。 | ||
| - 每行成本主要来自 `Cell` 的 `std::string text` 与完整属性结构（80 列 × 约 72 B），本轮未尝试压缩或共享存储；该数值是后续历史压缩/大历史实验的基线输入，**不是性能预算**。 | ||
| - 峰值工作集等于最终工作集（没有额外瞬时峰值），但进程基线含运行时与二进制映像；每行数字取自前后差值。 |
There was a problem hiding this comment.
The statement that peak working set equals final working set cannot be checked against the published evidence: validation.json omits peak_working for every sample, although the probe emits it. Moreover, that peak is captured before TextSearch runs, so it cannot establish the absence of a transient peak across the whole experiment. Preserve the per-sample peak values and explicitly restrict this conclusion to the fill phase, or remove the unsupported conclusion. AGENTS.md requires raw samples for every performance claim.
AGENTS.md reference: AGENTS.md:L12-L12
Useful? React with 👍 / 👎.
内容
补上 #164 最后一项里仍缺的生产内存与 H=10000 整文档 Search 观测：
history_memory_probe（EXCLUDE_FROM_ALL，默认不构建、不注册 CTest）：每个样本一个新进程，Parser(24,80,N)喂N+40行，用K32GetProcessMemoryInfo读工作集/私有提交，steady_clock记录填充与整文档 Search 的耗时/步数。experiments/core/results/2026-10-29-history-memory/（README + validation.json），并在 EVIDENCE 登记。结果（H=100/1000/10000，各 3 样本）
明确未做
Screen/TextSearch的观测，不是 r164 fork 的 Scan/LazyTable 对照，两者不可直接比较。Refs #164