Gap e2e: load-time reach of an owner file with no plan test is not modeled (exposed by per-law blocking)

未關閉
#186 0 則留言 0 個 reaction 已指派 0 人 在 GitHub 檢視

維護者通常 1 天內回覆

還沒有人認領這個 Issue。

評估

難度
5/5
預估耗時
一週以上
新手友好度
25/100
Issue 類型
缺陷
描述清晰度
基本清楚
活躍度
活躍
技術堆疊
javascript

研究方向

Start with src/sut.ts at D_OBJ:30 and the gap report replay in packages/oracles; compare the reported not-distinguished result with the two cited CI counterexamples. Read how the model and runner handle module-load reach, especially with isolate: false, then decide whether to model that rule or document the limit until the CLI records load reach per file. Done means the maintainer’s chosen behavior is captured and the counterexample is resolved or recorded as a stated limit.

由索引模型根據 Issue 內容生成。

描述

The end-to-end gap oracle on PR #176 (head 7f8e8104) found a counterexample. Its model cannot decide it, so this issue brings it to the maintainer.

  • Unit: random seed: a declaration fault: brokenOwner, seed 101
  • Difference: src/sut.ts:D_OBJ:30:field-drop#0: not-distinguished, owed indeterminate

Replay (from packages/oracles):

ORACLE_SEED=101 ORACLE_PATH=0:0:0:0:0:0:0 ORACLE_PROPERTY='oracle-gaps.report' ORACLE_UNIT='random seed: a declaration fault: brokenOwner' pnpm exec vitest --run --config oracle-gaps/vitest.config.mjs gap-report-oracle -t '^oracle-gaps\.report(?: | > )random seed: a declaration fault: brokenOwner$'

What the evidence shows:

  • For code that runs when the module loads (module data such as D_OBJ), the model judges by the plan's tests only (main's rule). An owner file whose only tests are not the plan's (the console test, the failing owner's passing pair) counts as no load reach.
  • Production does not follow one rule here:
    • In this counterexample, module data owned by files a and b, whose plan tests are all nondeterministic, was judged by some deterministic test that is not the plan's: not-distinguished.
    • In CI seed -555480425 of "a failing owner under its parent's config", module data owned only by file b, whose only tests were the failing owner's suite, was unreached.
    • In the fixed seed at fd32d67f, a model that counted those tests was wrong the other way (indeterminate, owed not-distinguished).
  • The runner runs with isolate: false, so the load's background row may depend on which file of a context loads the module first. The model does not represent that.

Before #176, the failing owner's file was blocked whole, so this was not visible. Block per law now lets that file's passing pair judge.

The question for the maintainer: should the model learn the runner's load-reach rule, or should this case be recorded as a stated limit of the e2e oracle until the CLI records load reach per file?

This issue was written by Isaac.

主要語言
JavaScript
星號
5
分支
1
平均合併
11 小時 18 分鐘
30 天內合併 PR
66

環境準備

這個專案沒有提供開發容器、Dockerfile 或貢獻指南,環境需要你自己搭建:先看它的 README,通用步驟見我們的新手貢獻指南。

從這裡開始

  1. 先讀完整個 Issue,再讀專案的貢獻指南。
  2. 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
  3. Fork 儲存庫,在一個分支上完成修改。
  4. 送出 Pull Request,並在描述裡引用這個 Issue 編號。

TanStack/oracles 的其他 Issue

查看 TanStack/oracles 的全部 Issue

相似的 Issue

更多 JavaScript Issue

把新 issue 寄到你的電子郵件信箱

精選適合新手參與的 GitHub issue 摘要。