Skip to content

[PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan - #6667

Open
ranxianglei wants to merge 1 commit into
masterfrom
2026-11-14_bounded-init-snapshot
Open

ranxianglei wants to merge 1 commit into
masterfrom
2026-11-14_bounded-init-snapshot

Conversation

@ranxianglei

Copy link
Copy Markdown
Owner

Summary

snapshotInit() in src/context-analysis.ts (the strictly-advisory per-session observability module) allocated one InitMessageEntry per message and sorted the complete array before retaining only 100. On a long first-request history that added O(n) temporary memory and O(n log n) synchronous work to the first proxied request, even though the visible sample is capped at INIT_PER_MSG_CAP (100).

This replaces the allocate-all + sort-all approach with a bounded top-K min-heap (worst entry at the root) that keeps only the 100 largest entries in memory while scanning. Aggregate counts (totalChars / estTokens) still cover every message.

Behavior

No behavior change. The emitted perMessage sample is byte-identical to the old whole-array stable sort:

  • descending estTokens
  • earlier-message precedence for equal sizes (the old stable sort's tie order)
  • aggregate counts over all messages
  • thinking tokens still part of the ranking (countMessageTokens unchanged)

Only the memory footprint and synchronous work of the snapshot drop (1M synthetic messages, local run):

before after
~1313 ms / ~85 MiB ~48 ms / ~13 MiB

The residual is the unavoidable per-message token count; the retained live memory is now 100 entries regardless of history length.

Tests

tests/context-analysis.test.ts:

  • new: bounded top-100 over a 300-message history — asserts the sample equals the unbounded "sort everything" top-100 (estTokens + index tie-break), pins the boundary positions, and asserts aggregates cover all 300 messages.
  • new: thinking tokens rank a small-text/large-thinking message above a larger visible-text message.

Existing init-profile tests still pass unchanged. No config surface change. No kernel change. No wire/persistence change.

Closes #6655

snapshotInit allocated one InitMessageEntry per message and sorted the
whole array before retaining 100, adding O(n) memory and O(n log n)
synchronous work to the first proxied request of a long session even
though the visible sample is capped at 100.

Replace it with a bounded top-K min-heap (worst entry at the root) that
keeps only the INIT_PER_MSG_CAP largest entries in memory while scanning.
The output is byte-identical to the old whole-array stable sort:
descending estTokens, earlier-message precedence for equal sizes,
aggregate counts over every message, and thinking tokens still ranked.

Closes #6655
@github-actions

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-11-14_bounded-init-snapshot (39acb45)

Latest build — install this exact version

npm install -g billion-context@0.0.9902-pr.6667.28ea99b

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

Recent builds (commit → version)

commit version build
39acb45 billion-context@0.0.9902-pr.6667.28ea99b build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr6667.tgz
npm install -g package

Install instructions are refreshed on each push.

@ranxianglei ranxianglei changed the title fix: bound init-snapshot memory/time with a top-100 min-heap scan [PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan Nov 14, 2026
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

查重 (duplicate screening): #6667 duplicates #6656. Two open PRs both Closes #6655, both fixing the same defect with the same strategy (keep a bounded top-100 sample while scanning instead of allocating+sorting the whole history). #6656 (external, opened 06:19 UTC) uses a sorted-insert bounded sample; #6667 (opened 07:28 UTC) uses a min-heap. Two open PRs against one thread is the merge-stepping-on-itself trap — the owner needs to pick one and close the other. I have not closed either (owner-only close). Details at the end.

Technical review of #6667 — the change is correct and clean.

  • Correctness (verified by execution, not just reading): I extracted the exact old (master) and new (#6667) snapshotInit and ran a differential harness — 200 random trials per size across 0/1/2/50/99/100/101/150/300/1000/5000/20000 messages (with estTokens ties, thinking tokens, tool names) plus an all-equal-value tie stress. Output is byte-identical to the old whole-array stable sort in every case. The min-heap (worst-at-root, initNodeKeeps = estTokens desc then index asc) selects exactly the same 100, and the final explicit estTokens desc || index asc sort reproduces the old stable-sort tie order — and is actually more deterministic (no longer relies on V8 sort stability).
  • Perf claim confirmed: synthetic 1M-message run — old 1127 ms / +85 MiB → new 20 ms / +15 MiB, outputs identical. Matches the PR/issue numbers.
  • All consumers safe: perMessage is read only as perMessage[0] (the largest: line in formatContextAnalysis, src/context-analysis.ts:272) and as the full list by the web card (src/web/sessions-data.ts:1580); aggregates (totalChars/estTokens/messageCount) are unchanged. The output shape (InitMessageEntry) is unchanged — index is internal to the scan and stripped before return.
  • Tests have teeth: the new bounded-top-100 test compares against an independent unbounded-sort oracle and pins the boundary positions (t149/t299 at the top, t250 at the bottom) — it would catch a tie-break or off-by-one. The thinking-token test's math checks out against the real countMessageTokens (ceil(len/4) for ASCII + thinkingTokens).
  • Diff clean: 2 files, no unrelated churn, no lockfile/whitespace. Naming/comments follow the file's style; no new deps; snapshotInit stays under the 80-line bar.
  • Labels (p2 / perf / scope/ui) are appropriate — advisory perf issue, not a cache leak, so not p1.

One process note (not a blocker): the PR base is 17 commits behind current master, but none of those 17 touch src/context-analysis.ts, tests/context-analysis.test.ts, or the kernel token surface, and the PR is mergeable: true — a rebase is trivial and behavior-neutral. CI is still running (no failures so far); the full-suite verdict is for CI on the pushed head.

8问 (结构化审核):

  1. 是什么问题:观测模块 snapshotInit() 为了只保留 100 条样本,却给每条消息都分配一个 entry 并对整个数组排序——长会话的首个代理请求被加上 O(n) 临时内存 + O(n log n) 同步耗时,而可见样本本有 100 上限。
  2. 什么原因导致的:实现是「全量分配 + 全量排序 + 截断」而非有界 top-K 扫描;超额内存/耗时的产生者就是无界 entries 数组和全量 sort。
  3. 影响范围与严重程度:任何带长历史的会话的首个代理请求;严重度为性能/延迟 + 一次性临时内存(1M 消息约 1s / 85MiB)。该模块纯观测、无决策路径读取,不影响正确性与可用性 → p2(可用,但长首请求有可避免的开销)。
  4. 怎么修复的:用「最坏元素在根」的有界 top-100 最小堆替换全量分配+排序,扫描期只保留最大的 100 条;聚合计数仍覆盖全部消息;输出与旧稳定排序逐字节一致(降序 estTokens,同值取更早 index,且不再依赖 V8 排序稳定性)。
  5. 是否有回归问题:无。perMessage 逐字节一致(独立差分验证 200 组×多尺寸 + 全同值压力均一致);所有消费方(perMessage[0] 报告行、web 卡片、聚合计数)读到的形状/值不变;无配置/线协议/持久化/kernel 变更;PR base 落后 master 17 提交,但那 17 个提交未触碰本 PR 改动文件或 kernel token 面,无语义冲突。
  6. 是否破坏缓存:不涉及。纯进程内观测,不碰前缀缓存/重试/增量游标/去重键。
  7. 是否改了设计方向:不涉及。同模块、同表面、同样纯观测定位,仅内部选择算法改变。
  8. 是否引入了新的配置:不涉及。

收工判定: #6667 本身可合(正确、干净、测试有牙、perf 属实)。唯一待办是重复 PR 的取舍:

🔶 需要决策 — #6656 与 #6667 同修 #6655,请二选一合并、关闭另一个(我不代关)。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — #6667 的冲突不是机械冲突,是「同一修复的两个等价实现,其一已合入 master」,需要你拍板留哪个。我已打 needs-decision 标,未擅自解冲突、未代关 PR。

现状(本轮 live 核实,非凭记忆):

为什么我不直接解: 两边都是正确且等价的实现,「取哪边」= 选内部算法 + 选保留哪个贡献,是设计/归属决策,不是「取一边」的机械冲突。floor 3 的 reviewer 已把「#6656/#6667 二选一」标成 🔶 需要决策、你还没拍板;且当时 #6656 未合,现在它已合入并 closed,情况变了。我不擅自替你选。

选项:

推荐 A;不决定的默认 = C(master 已修好,#6667 挂着无副作用,你随时可关或改选 B)。你回一句 A/B/C 我就执行(A=我留关单说明、由你关;B=我 rebase 取 min-heap 侧 force-push 后看 CI)。

标签: needs-decision(等待人工决策——同一修复的两个等价实现取舍,非机械冲突)。

8问(结构化审核 — 本轮 CI 车道):

  1. 是什么问题:[PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 与 master 冲突,但非普通文本冲突——是同一修复(Initial context analysis allocates and sorts the whole history to retain only 100 samples #6655)的两个等价实现相撞:master 已合 [PR] perf: bound the initial context profile sample #6656 sorted-insert,本 PR 是 min-heap。
  2. 什么原因导致:Initial context analysis allocates and sorts the whole history to retain only 100 samples #6655 有两个 PR([PR] perf: bound the initial context profile sample #6656 外部 / [PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 你自提)并行;[PR] perf: bound the initial context profile sample #6656 先经 batch6-wave6 合入 master 并 closed,后到的 [PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 在 snapshotInit 同一段上产生语义冲突。
  3. 影响范围与严重程度:仅影响 [PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 能否合并;master 本身已修好 Initial context analysis allocates and sorts the whole history to retain only 100 samples #6655(状态健康)。无用户面影响、无正确性风险(两实现输出逐字节等价)。严重度=流程性(重复 PR 取舍),非功能缺陷。
  4. 怎么处理的:核实 live 状态(两 PR 状态、master 已含 [PR] perf: bound the initial context profile sample #6656、dry-run rebase 确认冲突点、CI 无红项)→ 判定为设计/归属决策而非机械冲突 → 打 needs-decision 标 + 列选项等你拍板,未擅自 force-push / 未代关 PR。
  5. 是否有回归问题:不涉及(本轮未改任何代码)。若走 B 回退 [PR] perf: bound the initial context profile sample #6656,需重验 min-heap 版在 master 当前树的测试(dry-run 已确认测试可 auto-merge,B 的实际合并验证待决策后跑 CI)。
  6. 是否破坏缓存:不涉及。纯进程内可观测,不碰缓存/重试/游标/去重键。
  7. 是否改了设计方向:不涉及(未改代码);决策是「保留哪个等价实现」,不改变模块定位。
  8. 是否引入新配置:不涉及。

收工判定: 本轮 CI 车道唯一 blocker(语义冲突)已定位并升级为 🔶 需要决策,已打 needs-decision 标。等你在 A/B/C 拍板后我执行对应动作。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending, no change this run

Re-verified live (not from memory):

Nothing has changed since my floor-6 escalation and the owner has not chosen A/B/C, so there is no mechanical action to take this run. The full options + recommendation are in floor 6 — I'm not re-listing them. I did not force-push, modify the PR, or close it; the needs-decision label remains applied.

Holding for your A/B/C. Reply with the letter and I'll execute it.

8问(紧凑版,同 floor 6,无变化):

  1. 是什么问题:[PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 与 master 的语义冲突——同一修复 Initial context analysis allocates and sorts the whole history to retain only 100 samples #6655 的两个等价实现(master 已合 [PR] perf: bound the initial context profile sample #6656 sorted-insert vs 本 PR min-heap)在 snapshotInit 同段相撞。
  2. 什么原因:[PR] perf: bound the initial context profile sample #6656 先合入 master 并 closed,[PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 后到,两者重写了同一段。
  3. 影响范围与严重程度:仅 [PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 能否合并;master 已修好,无用户面/正确性风险。
  4. 怎么修复:live 复核确认无变化 → 仍属设计/归属决策 → 维持 needs-decision + 等你拍板,未擅动。
  5. 是否有回归:不涉及(本轮未改代码)。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入新配置:不涉及。

收工判定: 无新变化,决策仍待 owner;本轮仅复核 + 维持升级状态,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (4th re-trigger), no change

Re-verified live this run (not from memory):

The owner has not chosen A/B/C, so there is still no mechanical action to take this run. The full options + recommendation are in floor 6 — I'm not re-listing them. I did not force-push, modify, or close the PR.

Holding for your A/B/C. Reply with the letter and I'll execute it.

8问(紧凑版,同 floor 6/11,无变化):

  1. 是什么问题:[PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 与 master 的语义冲突——同一修复 Initial context analysis allocates and sorts the whole history to retain only 100 samples #6655 的两个等价实现(master 已合 [PR] perf: bound the initial context profile sample #6656 sorted-insert vs 本 PR min-heap)在 snapshotInit 同段相撞。
  2. 什么原因:[PR] perf: bound the initial context profile sample #6656 先合入 master 并 closed,[PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 后到,两者重写了同一段。
  3. 影响范围与严重程度:仅 [PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 能否合并;master 已修好,无用户面/正确性风险。
  4. 怎么修复:live 复核确认无变化 → 仍属设计/归属决策 → 维持 needs-decision + 等你拍板,未擅动。
  5. 是否有回归:不涉及(本轮未改代码)。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入新配置:不涉及。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (5th re-trigger), no change

Re-verified live this run: master f648217 / PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" — all unchanged since floor 15. The owner has not chosen A/B/C, so there is no mechanical action to take this run. Holding; I did not force-push, modify, or close the PR. Full options + recommendation are in floor 6.

8问: 同 floor 6/11/15,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (6th re-trigger), no change

Re-verified live: master advanced to de895aa (no new commit touched the conflict files); PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" — all unchanged. Owner has not chosen A/B/C. Holding; no action taken this run. Full options + recommendation in floor 6.

8问: 同 floor 6/11/15/16,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

2 similar comments
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (7th re-trigger), no change

Re-verified live: master 301e1a5 (no new commit touched the conflict files); PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" / needs-decision label — all unchanged. Branch head is clean (no [ework:ci-held] marker, so no prior run died mid-session). Owner has not chosen A/B/C. Holding; no action taken this run. Full options + recommendation in floor 6.

8问: 同 floor 6/11/15/16/20,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (8th re-trigger), no change

Re-verified live: master 301e1a5 / PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" — all unchanged. Owner has not chosen A/B/C. Holding; no action taken this run. Full options + recommendation in floor 6.

8问: 同 floor 6/11/15/16/20/27,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

3 similar comments
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (9th re-trigger), no change

Re-verified live: master 301e1a5 / PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" — all unchanged. Owner has not chosen A/B/C. Holding; no action taken this run. Full options + recommendation in floor 6.

8问: 同 floor 6/11/15/16/20/27/29,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (10th re-trigger), no change

Re-verified live: master 77fe0b1 (no new commit touched the conflict files) / PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" — all unchanged. Owner has not chosen A/B/C. Holding; no action taken this run. Full options + recommendation in floor 6.

8问: 同 floor 6/11/15/16/20/27/29/35,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (11th re-trigger), no change

Re-verified live: master a259e0b (no new commit touched the conflict files) / PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" — all unchanged. Owner has not chosen A/B/C. Holding; no action taken this run. Full options + recommendation in floor 6.

8问: 同 floor 6/11/15/16/20/27/29/35/38,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (12th re-trigger), no change

Re-verified live: master 4b77eab (no new commit touched the conflict files) / PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" — all unchanged. Owner has not chosen A/B/C. Holding; no action taken this run. Full options + recommendation in floor 6.

8问: 同 floor 6/11/15/16/20/27/29/35/38/43,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (13th re-trigger), no change

Re-verified live: master a446a3c (no new commit touched the conflict files) / PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" — all unchanged. Owner has not chosen A/B/C. Holding; no action taken this run. Full options + recommendation in floor 6.

8问: 同 floor 6/11/15/16/20/27/29/35/38/43/47,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (14th re-trigger), no change

Re-verified live: master a446a3c (unchanged) / PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" — all unchanged. Owner has not chosen A/B/C. Holding; no action taken this run. Full options + recommendation in floor 6.

8问: 同 floor 6/11/15/16/20/27/29/35/38/43/47/51,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (15th re-trigger), no change

Re-verified live: master 04ecbb4 (no new commit touched the conflict files) / PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" — all unchanged. Owner has not chosen A/B/C. Holding; no action taken this run. Full options + recommendation in floor 6.

8问: 同 floor 6/11/15/16/20/27/29/35/38/43/47/51/55,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

1 similar comment
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (16th re-trigger), no change

Re-verified live: master 6453fee (no new commit touched the conflict files) / PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" — all unchanged. Owner has not chosen A/B/C. Holding; no action taken this run. Full options + recommendation in floor 6.

8问: 同 floor 6/11/15/16/20/27/29/35/38/43/47/51/55/59,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (17th re-trigger), no change

Re-verified live: master 4649935 (no new commit touched the conflict files) / PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" — all unchanged. Owner has not chosen A/B/C. Holding; no action taken this run. Full options + recommendation in floor 6.

8问: 同 floor 6/11/15/16/20/27/29/35/38/43/47/51/55/59/63,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (18th re-trigger), no change

Re-verified live: master 496e058 (no new commit touched the conflict files) / PR 39acb45 / CI 0 failure (12 cancelled · 15 skipped · 3 success) / mergeable_state "dirty" — all unchanged. Owner has not chosen A/B/C. Holding; no action taken this run. Full options + recommendation in floor 6.

8问: 同 floor 6/11/15/16/20/27/29/35/38/43/47/51/55/59/63/67,无变化——语义冲突(同一修复 #6655 的两个等价实现)待 owner 拍板;本轮未改代码,无回归/缓存/设计方向/配置影响。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

🔶 需要决策 — still pending (19th re-trigger), no change

Re-verified live this run (not from memory):

  • master advanced to b10c1a0 (538 commits since the PR merge-base 1ff89e8); no new commit touched src/context-analysis.ts / tests/context-analysis.test.ts — the most recent commit touching them is still 3be4ed0756 ([PR] perf: bound the initial context profile sample #6656's sorted-insert, confirmed still on master).
  • PR [PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 head 39acb45 unchanged (min-heap).
  • Dry-run merge confirms the conflict is unchanged: CONFLICT (content) in src/context-analysis.ts (the snapshotInit body — master sorted-insert vs PR min-heap); tests/context-analysis.test.ts auto-merges.
  • CI: no red (0 failure; 12 cancelled / 15 skipped / 3 success); mergeable_state still "dirty".
  • needs-decision label confirmed still applied.

Same situation as floor 6 — the conflict is not mechanical, it's two byte-identical-equivalent implementations of the same fix (#6655) colliding in snapshotInit, one already merged to master. Choosing which to keep is a design/ownership decision, not a mechanical resolution.

Options (full detail in floor 6):

Owner has not chosen A/B/C. Holding; no action taken this run. I did not force-push, modify, or close the PR. Reply with the letter and I'll execute it.

8问 (紧凑版,同 floor 6,无变化):

  1. 是什么问题:[PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 与 master 的语义冲突——同一修复 Initial context analysis allocates and sorts the whole history to retain only 100 samples #6655 的两个等价实现(master 已合 [PR] perf: bound the initial context profile sample #6656 sorted-insert vs 本 PR min-heap)在 snapshotInit 同段相撞。
  2. 什么原因:[PR] perf: bound the initial context profile sample #6656 先合入 master 并 closed,[PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 后到,两者重写了同一段。
  3. 影响范围与严重程度:仅 [PR] fix: bound init-snapshot memory/time with a top-100 min-heap scan #6667 能否合并;master 已修好,无用户面/正确性风险(两实现输出逐字节等价)。
  4. 怎么修复:live 复核确认无变化 → 仍属设计/归属决策 → 维持 needs-decision + 等你拍板,未擅动。
  5. 是否有回归:不涉及(本轮未改代码)。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入新配置:不涉及。

收工判定: 无新变化,决策仍待 owner;本轮仅复核,未做任何改动。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[ework] CI fix verified green ✓

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Initial context analysis allocates and sorts the whole history to retain only 100 samples

1 participant