Skip to content

[PR] batch-D: tests-only — absorb #7288 (#5558 receipt) + #7164 (#5631 cold-load retry); #7373 already pinned in master - #7447

Open
ranxianglei wants to merge 3 commits into
masterfrom
batch-D-2026-11-21
Open

ranxianglei wants to merge 3 commits into
masterfrom
batch-D-2026-11-21

Conversation

@ranxianglei

Copy link
Copy Markdown
Owner

What this is

A tests-only batch (zero production behavior change) that absorbs two small open test PRs so their tests land on master in one shot, plus a verification note on #7373.

Absorbed original PRs

Original PR Author What it does Absorbed commit
#7288 ranxianglei Adds the #5558 no-op receipt regression tests (carried from superseded PR #5564): a fully protected range returns an applied no-op instead of a failure, and an already-covered range reports the covering block as a no-op b780fa22 + efe0656a (wording fix for the already-covered assertion)
#7164 ranxianglei Bounded-retry on the #5631 post-restart cold-load test to stop the known load-flake family from red-ing CI on timing 58625041

Original authors preserved via cherry-pick -x (no --no-keep-author). The original PRs will be closed as superseded after this batch lands — they stay open for now.

#7373 — verified already pinned in master, no new test needed

Issue #7373 asks for a full-path regression: drive a watermark-HIT turn through prepareOpenai -> persist -> shadow and assert the verified prefix is preserved, failing if the fix line if (wpHit !== undefined) session.lastMessagesFolded = true; (src/server/prepare-openai.ts:675) were removed.

That test already exists in master: tests/issue7346-pf-window-ledger-fullpath.test.ts, merged 2026-11-20 via #7357 (about an hour after the issue was filed). Duplicate-screening per AGENTS.md §7.1, I did not add a third copy. Empirical re-verification on this branch's head:

#7373 can be closed as already covered by that merged test.

Test evidence (targeted, local)

node --import ./tests/preload-test-env.mjs --import tsx --test tests/issue5558-ext-summary-noop.test.ts   # 4/4 pass
node --import ./tests/preload-test-env.mjs --import tsx --test tests/issue5631-restart-survival.test.ts   # 3/3 pass
node --import ./tests/preload-test-env.mjs --import tsx --test tests/issue7346-pf-window-ledger-fullpath.test.ts  # 1/1 pass (with fix); fails without it, see above
npm run typecheck   # clean

The authoritative full-suite verdict is the CI test lanes on this head (AGENTS.md §7.2: a local full-suite run is environment noise, not evidence; the local run was additionally blocked by the machine-wide test-suite lock held by a concurrent batch run).

Issues

ework-agent and others added 3 commits November 21, 2026 15:12
…5564

Carries the regression tests from PR #5564 (superseded by #5560) that assert
observable no-op receipt fields not covered by #5560's existing tests:
- reason === 'noop' and blocksCreated === 0 on the ext-lane no-op path
- compressFailStreak not re-armed on a legal no-op
- genuine kernel errors (too small) keep the refused receipt
- the shared no-op receipt did not change the normal path (applyRanges)

Text assertions adjusted to match #5560's receipt wording
('protected' / 'already compressed' instead of the PR's original
'protected zone' / 'already covered').

Closes #5558 (behavior fix already in master via #5560; this adds the
regression coverage that PR #5564 carried but never landed).

(cherry picked from commit 914efb4)
…nel receipt wording

The carried-over test asserted res.text.includes("already compressed"), but the
kernel's no-op warning for an already-covered range reads "already covered by
active block(s)" (kernel/src/compress.ts:1864) — "already compressed" is the
wording of the refused/error path (gate:already-compressed), not the no-op path.
The assertion therefore failed against master. Align it with #5560's existing
already-folded test, which uses /covered by active block/.

(cherry picked from commit 0f94cbb)
…load-flake family)

The #5631 restart-survival dsh lane flakes under CI load: the first
request after a restart can arrive before the proxy finishes reloading
the folded session state from disk, so the proxy sees the full raw
context and fails the preflight window check (502/413
preflight_compress_failed). Documented 3+ times in #4599 (floors
971/981/1049) as the 'post-restart preflight race' load-flake family;
local targeted runs are always green.

This is the test-side bounded-polling fix the fleet's durable item
(floor 903) pointed at: sendTurn now retries the known race signature
(502/413 preflight_compress_failed) up to 6x with 1s backoff on the
main arm. A one-off race succeeds on a later attempt; a real
fold-state loss 502s on every attempt and still fails the test. The
control arm is untouched (it must never fold, so a 502 there is a real
bug, not a race).

Closes #7145 (scheduled full test red at 548e396, ubuntu 24 shard 2,
dsh lane cycle0/sigterm/post-restart 0).
#7108 (windows shutdown-signal test) is already fixed on master at
3186e54; referenced for the aggregation board.

(cherry picked from commit b4f7c7a)
@github-actions

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: batch-D-2026-11-21 (5862504)

Latest build — install this exact version

npm install -g billion-context@0.0.11178-pr.7447.b2d4725

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
5862504 billion-context@0.0.11178-pr.7447.b2d4725 build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr7447.tgz
npm install -g package

Install instructions are refreshed on each push.

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

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant