Repository navigation
[PR] test(#5219): pin the JSON writer on so the decode-cap gate is host-hermetic - #6569
ranxianglei wants to merge 3 commits into
Conversation
…rmetic withUnified set BILI_STORAGE_UNIFIED=full but not the writer: a host config file with storage.jsonWriter=off (the pure-sqlite default since #6480, where the file value always wins over the 'full' dual-write default) silently leaves the disk empty, every boot decodes nothing, and the warm-set/small-corpus gates read 0. CI is clean-host so it never saw this; the env pin makes the gate independent of the host's storage knobs, matching what the test actually measures.
📦 Built Package ArtifactBranch: Latest build — install this exact versionnpm install -g billion-context@0.0.9688-pr.6569.fb43a5cThis exact version is immutable and reproducible. The rolling Recent builds (commit → version)
Download artifact
tar xzf billion-context-pr6569.tgz
npm install -g packageInstall instructions are refreshed on each push. |
[bot] 🏷 已核对 PR #6569(分支 修法核对(与 issue 描述一致):test-only,+11/-3,只动 CI 状态(实时读镜像缓存,刚查):30 项,0 红,7 项仍 queued( 判定:代码正确、无回归面(test-only,不碰 |
The suite baseline preload (preload-storage-baseline.mjs) pins BILI_STORAGE_UNIFIED="legacy" for the broad corpus. After #6480 flipped the storage default to pure-sqlite, the notice-registry suite's visibleNotices assertions expect the pure-sqlite default notice, which is gated on unifiedSelectionIsSqlite (unified "sqlite" or absent) — the preload's "legacy" made the notice silent, so the suite was born-red on master. Pin the selection explicitly (#4011 lesson: a test that depends on the production default passes locally but fails in CI).
(ai: vllm-qwen/qwen3.8-27b) CI red triage (auto-fix lane 1/3) — re-verified live state at head This PR is a clean test-only change (+11/-3 in 2 known inherited master mechanisms (both tracked in #4599):
ci-fixes is a no-op here: Per CI coordination policy (owner 2026-10-24): master-level fixes converge to ci-fixes (sentinel-maintained); I do NOT bundle an unrelated master-level test fix into this PR (one-scope + unrelated-changes discipline). The branch is already at master tip + 1 commit (merge ci-fixes = no-op), so I am leaving it untouched and waiting for the sentinel to converge the visibleNotices hermetic fix into ci-fixes. Reported to #4599 this run. 八问:
判定:本 PR 红 = 100% 继承性 master 红,非本 PR 代码问题;本分支不动,等哨兵 ci-fixes 收敛后重跑即绿。 |
[bot] 🏷 CI red triage (auto-fix lane 1/3) — outcome: Re-verified live check state at head Two known master mechanisms (both tracked in #4599):
ci-fixes is a no-op (0 ahead / 5 behind master, a007974); the sentinel's hermetic fix has not converged (stalled since #4599 floor 1309, disk-full). Per CI coordination policy (owner 2026-10-24), I left this branch untouched and reported to #4599. Once the sentinel converges the fix, Full triage with 八问 posted above. |
什么问题
#5634 合入的
tests/issue5219-bounded-boot-decode-cap.test.ts门禁只在BILI_STORAGE_UNIFIED=full下跑,没钉 JSON writer。而 #6480(同批合入)把存储默认切成 pure-SQLite 后,宿主配置文件里显式storage.jsonWriter=off会直接压过 full 的 dual-write 默认——这种宿主上writeNow不落 JSON 信封,boot 时磁盘零文件,门禁 3 条里 2 条必然红(本机复现:0 !== 10)。CI 是干净环境所以绿,属于环境性假红。修法(test-only,+11/-3)
withUnified 同时钉
BILI_STORAGE_JSON_WRITER=1(finally 还原),和该 suite 已有的 env 钉法同一模式。门禁测的是「有信封可读时的解码上限」,writer 开关不是它的变量。验证
本机(jsonWriter=off 宿主): 钉前 1/3 → 钉后 3/3;干净环境行为不变(env 优先级高于文件,CI 本来就 unset)。