test(export): assert post-#377 compress panel format — fix master-red (#391)

[ranxianglei]

Fixes #391

One-line fix: tests/export-cmd.test.ts:89 still asserted the pre-#377 panel /1 block/; #377 changed the format to blocks: b1=m00002 but its branch predated #272's merge, so neither PR's CI saw the other's tests. Master went red on merge; PR #380's CI failure is inherited from this.

Verified: 10/10 export-cmd + full suite green on this branch.

[github-actions]

📦 Built Extension Artifact

Branch: 2026-09-12_fix-export-cmd-panel-assert (00b3ff8)

Option A — Install from npm PR tag (recommended)

pi install npm:billion-context-pi@pr-392

Each push to this PR publishes a new version under the pr-392 npm tag.

Option B — Download artifact

1. Download the artifact from the Actions run
2. Extract the tarball and install:

tar xzf billion-context-pi-pr392.tgz
pi install ./package

---

This comment is automatically updated on each push.

[ranxianglei]

🤖 Powered by ework · qwen3.8-27b

Reviewed PR #392 end-to-end before recommending merge.

Verification (复现/查证)

• Reproduced on master (2bdc831): tests/export-cmd.test.ts fails 7/10 with AssertionError: compress created a block — line 89 still asserts the pre-feat: report new block ids and actual ref spans in compress result panel (#376) #377 /1 block/. Matches master red: export-cmd.test.ts asserts pre-#377 panel format ("1 block") — cross-PR regression between #272 and #377 #391 exactly.
• On this branch: export-cmd 10/10 green; full suite 725 tests → 722 pass / 0 fail / 3 skipped; tsc --noEmit clean.

Layer judgment (定位层次)
Root cause is correctly identified, not a symptom. #272 added the /acp-export test asserting the old panel /1 block/; #377 changed the live panel to blocks: b1=m00002. Confirmed in source: src/compress-tool.ts:498-501 builds …, blocks: <label>), and blockSpanLabel (:206-232) renders a single-message block covering m00002 as b1=m00002. Because #377 branched off pre-#272-merge, neither PR's CI saw the other's change → cross-PR regression surfaced only on merge. The fix updates the stale assertion to the real output; it does not mask any functional bug.

Solution validity (方案评估)
Correct and minimal. I grepped all of tests/* for other stale assertions on the old live-output format — none found. The remaining "… 1 block)" literals (view-recount.test.ts:104, compress-tool.test.ts:359, state-rebuild.test.ts:60) are intentional fixtures feeding the backward-compatible parser (compressPanelBlocks deliberately accepts both the legacy N blocks and the #376 blocks: span form), not stale assertions. So this single line is the complete fix — no sibling instances to chase.

Diff cleanliness: one line, exactly scoped to the described change. No unrelated files, no whitespace churn.

PR references its issue (Fixes #391) — trace requirement satisfied. Not merging (human-only); mergeable_state=blocked is the normal pre-merge state pending approval/CI.

中文摘要:修复了 #272/#377 交叉回归——master 上 export-cmd.test.ts:89 仍断言已被 #377 废弃的面板格式 /1 block/(实测 master 7/10 红),现已改为实际输出 /blocks: b1=m00002/;核验确认这是唯一的过期断言(其余 "1 block" 字样均为兼容解析器的测试夹具,非过期断言),改后全量 722 绿且 typecheck 通过,可以合并。

[ranxianglei]

已被 #393 取代(同一行字面量修复已随 #393 合并进 master,本 PR 变冗余且与 master 冲突)。建议关闭本 PR;若要消除格式耦合复发风险,合并 #390(把字面量正则换成生产谓词 isCompressSuccessText,已 rebase 到 master、9/9 CI 绿)。#391 → #393 因果链完整,无需其他动作。

[ranxianglei]

🤖 Powered by ework · qwen3.8-27b

Verified the supersession before acting, then closed #392.

Verification

• Live master is now bcd2770 (my first local fetch was stale at 2bdc831; the GitHub branch API + a forced refspec refetch confirm the real tip). Line 89 on master = assert.match(text, /blocks: b1=m00002/, …) — the fix is present.
• test(export): assert post-#377 panel format in setupSession #393 is genuinely merged (merge commit 93ace1d2, head d5a15a6). Its diff is byte-identical to test(export): assert post-#377 compress panel format — fix master-red (#391) #392: same file, same line, same old→new (index 7f49f7f..2ffc191). test(export): assert post-#377 compress panel format — fix master-red (#391) #392 therefore contributes nothing beyond what test(export): assert post-#377 panel format in setupSession #393 already landed → fully redundant.
• One precision on “与 master 冲突”: a trial git merge of test(export): assert post-#377 compress panel format — fix master-red (#391) #392 onto current master actually auto-resolves cleanly as a no-op (both sides made the identical change), so it's redundant rather than hard-conflicting. Same conclusion either way — nothing to gain from merging it.

Action taken

• Closed test(export): assert post-#377 compress panel format — fix master-red (#391) #392 without merging: state=closed, merged=false. Reopenable if needed.

On #390 (the robust variant)

• Confirmed it's rebased onto the current master tip bcd2770 and swaps the literal regex for the production predicate: assert.ok(isCompressSuccessText(text), …) (imports isCompressSuccessText from src/compress-tool.ts). That decouples the test from the exact panel string — the right long-term guard against this class of cross-PR format regression.
• Ran its tests/export-cmd.test.ts locally: 10/10 green.
• I did not merge it — PR merges are human-only here. It's open and ready; your call whether to land it in place of the literal assertion that test(export): assert post-#377 panel format in setupSession #393 introduced.

中文摘要:核实后确认 #393 已把同一行修复合入 master(现 tip bcd2770,line 89 已是 /blocks: b1=m00002/,且 #392 与 #393 逐字节相同、完全冗余),已按你的要求关闭 #392(未合并、可重开);补一句——试合并显示它对当前 master 是干净 no-op 而非硬冲突,但结论一致。#390 已 rebase 到最新 master、改用生产谓词 isCompressSuccessText(本地 export-cmd 10/10 绿),更抗格式回归,是否合并由你拍板(我按规则不代合并)。