Skip to content

test: extend offline clean-wheel agent acceptance to sharded datasets - #262

Merged
codeasier merged 4 commits into
mainfrom
test/253-offline-agent-shards
Oct 11, 2026
Merged

codeasier merged 4 commits into
mainfrom
test/253-offline-agent-shards

Conversation

@codeasier

@codeasier codeasier commented Oct 11, 2026 •

Copy link
Copy Markdown
Owner

Summary

  • Add --events-per-slice to the offline clean-wheel CLI/skills harness. Preserve standalone import/reuse/query acceptance, then exercise both pt-snap-native-v2 and compatibility-v1 through the installed public CLI.
  • Require multiple real shards. Check directory and manifest discovery, import provenance, stable fingerprints, explicit devices, full-device peak parity, opposite-shard point lookups, boundary-crossing event parity, and unchanged project focus.
  • Install the bundled helper and peak diagnostic under isolated HOME; verify complete-dataset instructions, catalog support, command availability, and exact installed/bundled content.
  • Audit the wheel against source and verify the reviewed fixture manifest before installing/importing. Clear inherited Python, product and pip overrides. Optional --wheelhouse installs dependencies offline into a venv without system packages and checks their installed paths.
  • Run --events-per-slice 4096 in the existing Python 3.13 packaging gate, with a regression for the exact invocation and fail-closed harness guards.

Related issue

Stacked directly on #248 infrastructure commit 68d89670b188c93e41b1a5d8f41903057e5ada06; all infrastructure gates are retained.

Refs #253 item 6 and #248. This completes the offline sharded clean-wheel portion only; it leaves #253 open.

Verification

  • Stacked integration focused suite: 378 passed, including fixture provenance, harness/skills, governance/release, wheel audit, CI tiers and fail-closed dataset acceptance runner.
  • Ruff, Black and git diff --check: passed.
  • basedpyright: 0 errors; existing warnings remain informational.
  • Exact committed wheel: source audit and actual offline standalone/sharded acceptance passed. Each format used two physical shards with 8,180 real events.
  • On the original byte-identical harness, deliberately inherited tiny timeout/one-row limits were cleared during the real sharded run.
  • Real one-shard negative (--events-per-slice 10000) failed as required on the original byte-identical harness; its fail-closed focused regression also passed on the stacked integration.
  • Original implementation 468166d: full pytest 2,658 passed, 15 skipped (12 browser, one opt-in scale case, and two intentionally unsupported slice-leak combinations), in 357.32 seconds. The stacked integration preserves all runtime/harness/skill bytes and only resolves infrastructure workflow/doc context; the full suite was not repeated for that union.

Both local scenario runs used CPython 3.12 with --wheelhouse; default system-site-packages dependency mode and hosted CI were not executed locally.

The locally verified wheel SHA-256 is 00efc9a82a15e1931b50ca4c390fe1d00a1826635c49a1f93bd9c7dca6cd9af0.

Remaining boundary

No live agent/model runner was implemented or executed. No real user agent session or model provider was connected. Baseline grading continues to score stored records; filesystem skill installation and CLI success do not establish host loading, fresh model execution, or decision quality. The live-agent portion of #253 remains incomplete and requires separate approval.

No new committed pickle fixture or snapshot runtime file was introduced. Default acceptance still shares system dependencies; fully isolated offline dependencies require --wheelhouse.

Copy link
Copy Markdown
Owner Author

Latest-head hosted CI verified on 0fb83847b926805d804147540d554bbe29849c0b: Test run 38147794549 succeeded, with seven applicable jobs passed. Python 3.13: 2691 passed, 14 skipped, 1 deselected, coverage 93.18%; wheel audit and offline clean-wheel CLI/skills acceptance passed. The installed-wheel dataset/nightly job was skipped for this PR event, not executed or passed. Actual local scale evidence is separate; no live-agent/model result is claimed. Remote commit/tree matches the independently reviewed local source.

Copy link
Copy Markdown
Owner Author

检视结论:approve

对照 #253 agent/wheel 分片:--events-per-slice;双格式多 shard;3.13 packaging 门禁。定向 43 passed。与 runner REQUIRED 元组冲突较少,建议放在扩展 nightly 的 #260→#259→#263 串行之后最后合。本评论不 approve/merge。

@codeasier codeasier left a comment

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM,已转为正式 PR。GitHub 不允许作者账号提交 APPROVE。建议放在 #260→#259→#263 串行之后最后合。

Refs #253. Preserve seven required generic gates, independent 200-seed step, four Windows partitions and resource evidence portability.

Copy link
Copy Markdown
Owner Author

已最后接入实际 main 91b06d6f746ece707cb93c3aeaef49fcbbc990a9,新 head 4fb33017b9e26c2a1cf821a1726cd5676ac9d377,树 729c947e86037e6fe612fa3dcca938c30e1a9681。非 force、expected-head lease 更新成功。

六处历史基础设施冲突按已独立审阅的最终并集解决;本 PR 对当前 main 仅7文件差异。保留7个 REQUIRED generic节点、同一 nightly job 中独立200-seed步骤、4个Windows分片,以及#263最新测试证据路径修复。offline harness和对应回归测试内容与原审阅PR字节一致,3.13 packaging继续强制4096分片。

当前树与更新后本地预览 f3dc9816b347ba5d9df36c7532995665962ab730 完全一致,119项focused通过。之前独立最终预览(尚未含#263两文件路径修复)另实跑7/7 generic和双格式offline;该旧证据不作为新远端head已执行的声明。新head完整托管CI待验证;最终实际main还将重新运行整体验收。

@codeasier
codeasier merged commit 07bdf66 into main Oct 11, 2026
13 checks passed

Copy link
Copy Markdown
Owner Author

最终实际 main 验收完成,9个PR已按批准顺序合入。

  • Source 07bdf660f10f03a9c13ab313fae149adf62355f8;tree 729c947e86037e6fe612fa3dcca938c30e1a9681。
  • 本次重新构建并源码审计的 wheel SHA-256:7ac005c04dda8572bbb689b494875a147b554a1a732d79c09f9e428a77fc9787。
  • 实际 main 的 Test CI 和 Platform CI 均成功,涵盖Python3.10–3.13、浏览器、4个Windows分片和macOS。
  • 独立Linux实际main:2922 passed;启用分支统计的综合coverage 93.19%,关键文件门槛全部通过。Ruff/Black通过;basedpyright零错误、1745 warnings。
  • 新鲜Python3.12隔离wheel环境:7/7 generic必需节点实跑成功,无失败/错误/skip,required/executed身份完全一致。含1,000,000事件、500,002短尾、3个route-million及2个resource节点。
  • 独立200-seed runner:400/400双格式case通过,28,864次比较、零unexpected error、2,128次scratch build。
  • offline 4096验收双格式通过;各8180真实事件、2个物理shard,验证reuse/discovery/parity及focus不变。
  • 最终源码再次确认7节点并集及同一nightly job内独立200-seed步骤;Linux collection证明4个730-case分区互斥且完整覆盖2920个非slow case。Windows实际执行以以上托管CI为准。

范围说明:本地source run明确排除12个受云端socket限制的浏览器case(托管浏览器CI已通过),另有2个预期不支持的slice leak skip。push事件的nightly job按设计跳过,以上7/7与400/400是绑定实际main的新本地实跑证据;未宣称远端nightly已运行。所有相关issue #248–#253仍保持open;无live-model/真实agent-session验收。

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