Repository navigation
release v0.1.167 - #1587
release v0.1.167#1587
Conversation
📦 Built Package ArtifactBranch: Option A — Install this exact build (recommended)npm install -g billion-context@0.0.1524-pr.1587.73298b9This exact version is immutable and reproducible. The rolling Option B — Download artifact
tar xzf billion-context-pr1587.tgz
npm install -g packageInstall instructions are refreshed on each push. |
2202dd2 to
23800c9
Compare
[bot] 🏷 Review complete — mergeable. One fix applied directly on the branch during review (details below), everything re-verified on the new head. Fix applied during review (stale base)The branch was cut at
① Original issue (#1582 checklist step 1) — direction OKRoutine patch whose whole purpose is shipping the validated updater to the fleet so instances start polling ②b Version-window regression audit (v0.1.166..HEAD — 2 content changes)A. Advisory watcher re-land (#1481/#1483 via #1582) — full audit already done in #1582 floor 3 against byte-identical content (
B. Tag-tail drain (#1546 via #1549) — went through its own full review cycle (two agent rounds + your deep review, verdict "merge as-is"). Window-audit re-check of its touched paths: ordinary streams byte-identical with no held tail (every return path reduces to pre-fix value); ④ Test validity & pre-flight (rebased head
|
Step 1 of the post-merge checklist in #1582 ("Release 0.1.167 (routine)") — the routine patch after the advisory re-land; ships the new updater to the fleet so instances start polling
billion-context-advisories.Changes since v0.1.166
Two content changes (branch rebased onto current master
1facb6e7; diff vs master remains only the version bump):billion-context-advisories(published by CI fromadvisories/) every 180s, independent of auto-update; versions inside an entry'saffectedrange are force-installed to itstarget(rollback-capable), with the remote reason surfaced as a log warning + web banner +/__bili/status; fail-open on unreachable/malformed sources; source checkouts (auto-update overwrites tracked files when run from a git source checkout (findInstallDir resolves to repo root) #580) and host-managed copies (native模式下自动更新问题需要解决 #991) are never overwritten in place; the normal self-update loop defers while an advisory is active but NOT when the target is unresolvable (F1/F2).BILI_ADVISORY_CHECK=0/ config"advisoryCheck": false; document overrideBILI_ADVISORY_URL. Documented in CONFIGURATION.md (en+zh).hadText=true, suppressing the only path that flushes the tag-echo filter's held tail ahead of the current frame; a terminal frame carrying buffered tag-start text took the rebuilt path with no pre-flush — either way the tail stranded until[DONE]/stream-end, i.e. AFTER thefinish_reasonframe, which strict consumers abort on (OpenCode 2.0.18: "OpenAI Chat received content after the finish reason"). Fixed at the decision point inprocessOpenai+processGoogle: empty-string fields count as no-text; terminal frames fold their own fields' held tails into their deltas, then drain any remaining pending as synthetic chunks ahead of the frame. Wire ordering changes only in this edge case; ordinary streams byte-identical; tool-call arguments stay byte-exact ([bug] 插件透传会剥掉工具调用参数里的 render tag(arguments 流挂了 tag-echo 过滤器) #1039).Behavior changes (disclosed, intended): one background GET to the npm registry every ~3 min even when auto-update is off; additive
advisoryfield in/__bili/status; no change to existing self-update behavior when no advisory matches; plus the #1549 wire-ordering edge case (content previously emitted after the finish frame now emits before/at it).Pre-flight (local, rebased head
23800c99)npm ci✓ ·npm run typecheck✓ ·npm test✓ (3073 tests: 3061 pass / 0 fail / 12 e2e-gated skips) ·npm run build✓Note: the one-click fast path (
Release (one-click)) would serve this routine patch equally well — if you prefer that route, close this PR and trigger the workflow instead (the drift guard makes both safe).