Review history drops older reports that still fit

未关闭
  • 哪里不对？
#596 2 条评论 0 个 reaction 已指派 1 人 在 GitHub 查看

维护者通常 1 天内回复

@HellHBBD 已经在做这个了。

开始于 2026年11月6日。

评估

这个 Issue 还没有评估数据。

描述

Observed behavior

The browser-local review archive prematurely drops older reports, even when the 500-entry limit and the 164 KiB byte budget would allow them to be retained.

In web/history.js, boundedReviews() limits the input to the first 500 candidates before checking whether each report fits the remaining byte budget. A skipped report still consumes an input position, preventing an older report from being considered.

In a deterministic reproduction, saving three large reports into an initially full 500-entry archive leaves only 499 reports after the third save. An older report (old-497) disappears even though retaining it would keep the archive within both limits.

This affects browser-local report history. The reproduction does not demonstrate account-database report loss.

Expected behavior

A report that cannot fit within the remaining byte budget should be skipped without consuming a retained-report slot.

Older candidates should continue to be considered until either 500 reports have been retained or all available candidates have been examined.

In the reproduced scenario, the archive should retain exactly 500 reports, including old-497, totaling 160,917 bytes, below the 164 KiB (167,936 bytes) budget.

Existing newest-first ordering, UTF-8 byte accounting, code stripping, and short-history retention should remain unchanged.

Reproduction

Reproduced using the real history persistence module on upstream main at commit b2c2bc9aca5e9e70f28ff2a16e4bd0b2d5e87a2f, with Node.js v26.12.1, in-memory storage, and a mocked signed-out session.

Steps to reproduce
  1. Seed codetrial_review_history with 500 small reports, old-0 through old-499.
  2. Seed codetrial_history with the newest 20 entries.
  3. Call saveReportHistory() three times with distinct reports containing 56,000-character summaries.
  4. Inspect the retained review archive after each save.
Observed results
  • First save: 500 retained reports, 104,904 bytes; oldest retained: old-498.
  • Second save: 500 retained reports, 160,917 bytes; oldest retained: old-497.
  • Third save: 499 retained reports, 160,819 bytes; oldest retained: old-496.

After the third save, old-497 is missing. Restoring it would produce 500 reports totaling 160,917 bytes, below the 164 KiB (167,936 bytes) limit.

The failure is deterministic with this setup. The reproduction uses mocked browser storage rather than an interactive interview.

CodeTrial version

codetrial 0.1.0 (b2c2bc9)

Environment

OS: Arch Linux
Browser: Not applicable; reproduced under Node.js v26.12.1 with mocked browser storage.
Language: Not applicable.
Interview loop: Not applicable; no interview session was started.
Problem: Not applicable; browser-local review-history persistence.
REACTO or STAR step: Not applicable.
Reconnect: Not applicable.

Reproduction environment: Real web/history.js persistence module, in-memory storage, and a mocked signed-out /api/session response.

Sanitized evidence

The regression assertion fails on the unmodified upstream implementation:

AssertionError: Archive should retain 500 reports when another older report fits

499 !== 500

Relevant implementation (web/history.js, lines 462-489):

https://github.com/sysprog21/codetrial/blob/b2c2bc9aca5e9e70f28ff2a16e4bd0b2d5e87a2f/web/history.js#L462-L489

for (const row of reviews.slice(0, REVIEW_HISTORY_CAP)) {

The input is truncated before the byte-budget check. Candidates skipped for exceeding the remaining budget still occupy positions within the first 500 inputs, so older candidates are not considered.

The existing oversized-review test in tests/browser/account.test.js (lines 864-880) does not cover this interaction between a full 500-entry archive and byte-budget rejection.

https://github.com/sysprog21/codetrial/blob/b2c2bc9aca5e9e70f28ff2a16e4bd0b2d5e87a2f/tests/browser/account.test.js#L864-L880

Related work: #221 and PR #321 address final-code retention, not this premature review-archive eviction.

The reproduction uses synthetic report identifiers and summaries. No credentials, session tokens, recordings, or personal interview content are included.

Public data confirmation
  • I removed credentials, session tokens, candidate recordings, and personal interview content from this public report.
主要语言
Rust
星标
223
派生
148
平均合并
1 天 10 小时
30 天内合并 PR
227

环境准备

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

sysprog21/codetrial 的其他 Issue

查看 sysprog21/codetrial 的全部 Issue

相似的 Issue

更多 Rust Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。