Skip to content

release v0.1.159 - #1477

Merged
ranxianglei merged 1 commit into
masterfrom
2026-09-27_release-v0.1.159
Sep 27, 2026
Merged

ranxianglei merged 1 commit into
masterfrom
2026-09-27_release-v0.1.159

Conversation

@github-actions

Copy link
Copy Markdown
Contributor

Auto-fallback from the one-click release workflow: direct push to master was blocked by branch protection. Merging this PR publishes via the standard release.yml flow.

Changes since v0.1.158

@github-actions

Copy link
Copy Markdown
Contributor Author

📦 Built Package Artifact

Branch: 2026-09-27_release-v0.1.159 (6030174)

Option A — Install this exact build (recommended)

npm install -g billion-context@0.0.1400-pr.1477.a6e9497-a2

This exact version is immutable and reproducible. The rolling pr tag tracks the newest test build across all PRs.

Option B — Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr1477.tgz
npm install -g package

Install instructions are refreshed on each push.

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 Release audit (v0.1.158 → v0.1.159) — verified, ready to merge.

Release commit & base

  • 60301741 release v0.1.159 changes ONLY package.json version (0.1.158 → 0.1.159) + matching package-lock.json lines. No content bundled into the release commit.
  • Base is current master (a262be3e); npm view billion-context version = 0.1.158, so the drift guard is satisfied and 0.1.159 is unpublished.
  • PR state: mergeable: true, mergeable_state: clean.

CI — all 13 check runs on head 60301741 are completed success: full test matrix (ubuntu/windows × node 22/24), version-guard, e2e (real codex), e2e-fake-codex lanes (pi/opencode/dsh), hermetic local-registry e2e, Windows codex overlay, build-artifact.

Local pre-flight on the PR head (independent of CI): npm ci ✓ · typecheck ✓ · npm test → 2893 pass / 0 fail / 12 skipped · build ✓ (+ dist-import-annotations OK).

Window regression audit (v0.1.158..HEAD, 43 commits, 60 files, +2525/−356)

本版本引入回归:无。 Behavior changes introduced this cycle (all intentional, all disclosed here old→new):

  1. Stream-failure error shape default flipped (上游截断在 compress 内部重发路径无法恢复:重试闸门将 ping 计为已转发输出,且 emitError 合成 end_turn 伪完成 #1455): old = upstream stream failures delivered inside a legacy synthesized successful completion; new default compat.streamErrorShape: "protocol" = protocol-native error frames (anthropic/openai wires). Escape hatch: compat.streamErrorShape: "completion" or BILI_STREAM_ERROR_SHAPE=completion restores the old shape. Exit-matrix docs + CONFIGURATION aligned.
  2. Truncation-retry gate widened (上游截断在 compress 内部重发路径无法恢复:重试闸门将 ping 计为已转发输出,且 emitError 合成 end_turn 伪完成 #1455): inert pings now count as zero-side-effect, and anthropic reasoning-only rounds are retried (state-keyed start suppression + dangling-block close). More upstream truncations get retried than before — intended fix.
  3. Cache-economics reports now priced in real money (feat: expose compress.priceProfile {w,r,q} config (cache-economics report hardcoded to Anthropic weights) #1279 follow-up): new compress.priceProfile {w,r,q} (three-level cascade). Per-request stamp precedence: user config > models.dev catalog price for the request's model/host > kernel defaults (w=1, r=0.1, q=4). Registry fetch moved models.dev/models.json → models.dev/catalog.json (cost rows live there); disk-cache format {models, costs} with legacy flat maps still readable. Reports for stamped sessions show absolute $/Mtok instead of Anthropic-ratio guesses; unstamped sessions stay byte-identical to pre-feat: expose compress.priceProfile {w,r,q} config (cache-economics report hardcoded to Anthropic weights) #1279. Report-only — no trigger or wire impact.
  4. CCR v3 planning-aware search (feat: CCR v3 — planning-aware retrieval for search_context (compress.search.planAware) #1419): new compress.search.planAware, off by default — absent/false gives byte-identical output to plain lexical ranking. Opt-in only.
  5. Feature: CCR v3 — planning-aware retrieval (rank candidates against todo/task state, LineRetriever-style) #1336 retrieval-quality counters round-trip through restart (persist.ts): wholeBlockRestores/wholeBlockRestoresPreciseAvailable previously silently reset on reload; now persisted.
  6. dashscope host → real alibaba catalog provider, comfly dead-end dropped (HOST_TO_PROVIDER maps dashscope/comfly hosts to provider ids absent from the models.dev catalog → no registry window/price stamps for those hosts #1434): context-window resolution for dashscope model ids now hits real catalog entries.
  7. Electron/GUI node-runtime resolution ([Bug] Electron 宿主(DSH 桌面版)下 resolveNodeRuntime 找不到 node:代理起不来、5 个工具全丢、会话被撑爆到模型拒答 #1429): launcher finds node under Electron hosts instead of failing.
  8. Debug: ACP_DUMP_SSE now also captures loop-originated re-requests/retries (observability only).
  9. CI: rolling single pr tag + monotonic 0.0.<run>-pr dev versions (Design: PR artifact publishing — stop dist-tag growth with single rolling tag + monotonic 0.0.x-pr versions #1385) — artifact naming only.

Verdict: release commit clean, base fresh, no drift, CI fully green, no regressions in the window, all behavior deltas disclosed above. Merge when convenient — CI publishes automatically.

中文摘要:审核了 v0.1.159 发版 PR(含 v0.1.158 以来全部 43 个提交的区间回归审计)——release 提交干净、基线最新、CI 全绿、本地测试 2893 全过,无回归;本版本有 9 项有意行为变更(最重要:流式失败默认改为协议原生错误帧、缓存经济报告改用 models.dev 真实定价、CCR v3 计划感知检索默认关闭),均已按旧值→新值披露,可以合并。

@ranxianglei
ranxianglei merged commit b780916 into master Sep 27, 2026
13 checks passed
@ranxianglei
ranxianglei deleted the 2026-09-27_release-v0.1.159 branch October 15, 2026 05:32
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant