[finding] check-cross-package-test-inputs.mjs --union-into appends to packages.items but leaves packages.count stale, so the file it hands to the shard partitioner is self-contradictory #10046

Activity

1. os-steve

   added

   finding

   on Aug 19, 2026
2. claude

   added

   tooling

   pm:queue

   domain:devx

   pm:dispatched

   and removed

   pm:queue

   on Aug 19, 2026
3. claude commented on Aug 19, 2026

   

   claudebot

   on Aug 19, 2026 – with Claude

   Contributor

   More actions

   Claim: PM loop round 24
   Session: session_01XqDQYVU5smx29ts9pAErja
   Branch: claude/issue-10046-union-into-count
   Worktree: objectstack-issue-10046
   Domain: domain:devx
   File surface: scripts/check-cross-package-test-inputs.mjs (the --union-into writer), possibly scripts/partition-test-shards.mjs (reader) (stop on breach; explain in the report)
   Container & model: S, mode:subagent, model: opus (--tier on these paths returns "no path-derived mandate … This line is a FLOOR, never a clearance"; content limb no — the emitted document gains a correct count; no gate's accept/reject moves)
   Clause-②: no
   Serial constraints cleared: swept all open PRs for any touching cross-package-test-inputs or partition-test-shards — none. Predecessors on this file, PR #9826 (#9763) and PR #10018 (#9977), are both MERGED.

   ---

   Generated by Claude Code
4. claude

   assigned

   os-steve

   on Aug 19, 2026
5. os-steve commented on Aug 19, 2026

   

   os-steve

   on Aug 19, 2026 – with Claude

   CollaboratorAuthor

   More actions

   {
     "issue": 10046,
     "status": "done",
     "branch": "claude/issue-10046-union-into-count",
     "pr": "https://github.com/objectstack-ai/objectstack/pull/10059",
     "premise_still_valid": true,
     "summary": "RULING 1 = MAINTAIN, on measured evidence about whose schema `count` is: `turbo ls --output=json` (turbo 2.10.10) emits it itself and keeps count === items.length in every form measured (bare 77/77, --filter 1/1, --affected 77/77). It is turbo's field, not this script's invention, so it is not ours to delete -- deleting it would make our hand-mutated document stop being a valid `turbo ls` payload, the opposite of what partition-test-shards.mjs's assert-the-upstream-shape-loudly posture exists for. Both halves ship together: the writer reconciles count inside `serializePackageList()`, the ONLY thing in unionInto() that turns the document into bytes (a `reconcile(); write();` pair would re-create the defect the first time someone added a second write path); the reader `readPackageItems()` refuses a payload whose count disagrees with items.length, naming the contradiction. RULING 2: a future disagreement is now caught in production -- the reader runs on every Test Core shard, so it is a red step naming the cause, not a self-test hope. A payload with NO count is accepted deliberately (a redundant field's absence cannot mis-shard; its disagreement can), and a present-but-wrong count is rejected, never repaired. RULING 3 verified, not assumed: old vs new partitioner over the real 77-package document, all 3 shards, produce IDENTICAL package lists, weights and bin totals (23/29/24, 781/780/780). Rulings 4 and 5 hold: no governed surface touched, no changeset, `skip-changeset` applied and read back. ONE CHANGE BEYOND THE MINIMAL DIFF, declared: `partition-test-shards.mjs --self-test` was wired to NOTHING -- not package.json, not any workflow -- so every assertion in it, including the partitioner's pre-existing determinism/coverage/balance pins, evaluated never. Adding reader-side pins to an unrun self-test would have been dead code, and the card asked for a pin that 'keeps it from rotting back', which an unrun pin cannot do. It is now a step in Lint & Repo Gates, invoked as `node ... --self-test` rather than a `pnpm check:*` alias because that alias would live in root package.json, inside the #9465 fence -- matching the release-rehearsal self-test step directly above it.",
     "tests": "All at final commit a0b947783d (worktree, turbo 2.10.10, node 22.22.2).\n\nH1 REPRODUCED then FIXED. Pre-fix on 20b9a9ce1b, changed-file list = exactly scripts/sync-template-versions.mjs: `packages.count = 0 / items.length = 2`. Post-fix: `count 2 / items 2 / AGREE true`, packageManager 'pnpm9' untouched. The old contradictory document is now refused: exit 1, `packages.count is 0 but packages.items holds 2 -- the payload contradicts itself about its own size ... Refusing to shard it.` Mutations confirmed on disk independently of the edit tool: injected tokens grepped with counts (serializePackageList 5, readPackageItems 12, 'Refusing to shard it' 1) AND removed text confirmed absent (reconcilePackageCount 0, `writeFileSync(listPath, JSON.stringify(parsed))` 0).\n\nGATE UNION re-derived with `node scripts/pm/dispatch-gates.mjs` (no paths passed; it reads the merge base itself) -- it named SIX families the dispatch brief did not, all pulled in by the lint.yml edit. Exit codes captured by redirect-then-capture, never through a pipe; verdict lines quoted from the gates' own output: check:cross-package-test-inputs 0 'OK: 12 package(s) read outside themselves, all declared, and turbo.json hashes every declared glob.' | check:node-version 0 'OK (30 setup-node step(s) across 27 workflow(s), all on Node 22).' | check:required-contexts 0 | check:shard-attestation 0 | check:type-check-coverage 0 'OK -- 64/77 workspace packages type-checked (plus the root), 13 in the DEBT ledger' | check:workflow-status-functions 0 'OK (scanned 27 workflow file(s), 46 job(s), 24 job-level if: expression(s))' | check:nul-bytes 0 'OK (scanned 6358 text file(s) ... 0 untracked-not-ignored)' | partition-test-shards.mjs --self-test 0 'partition-test-shards: self-test OK' | check-cross-package-test-inputs.mjs --self-test 0 'All 57 self-test cases passed.'\n\nDECLARED NARROWING (one): check:type-check-debt exits 1 here, refusing to --re-measure because 55 workspace dependencies have no built dist/*.d.ts. NOT caused by this branch -- the identical refusal reproduces on the unmodified shared origin/main checkout with none of this diff present, and the diff touches ZERO TypeScript files (3 files: 2 .mjs, 1 workflow YAML). The half of that gate that reads lint.yml passed and printed its OK line before the closure precondition stopped the re-measurement. Building the 77-package closure would hold the container's shared heavy-verify lock ~10 min to re-measure a TS debt ledger this change cannot move; CI builds the closure before that step.\n\nABLATION -- REBUILD STATEMENT: no rebuild leg applies, and that is checked rather than assumed. Both scripts import only `node:` builtins, are executed directly as source by `node scripts/*.mjs`, appear in no vitest/test-source-alias ledger, and are reached through no package build -- there is no dist/ between the edit and the measurement, so the dist preflight is inapplicable rather than skipped. Both legs run from the COMMITTED state and restored with `git checkout HEAD -- <path>`; tree AND index confirmed clean afterwards via `git status --porcelain` (not `git diff HEAD`). Direction predicted before each run. WRITER leg: predicted red on exactly 2 of the 5 new cases (the other 3 already have a correct count, so a missing assignment is invisible to them) -- observed exactly that, '2/57 self-test case(s) failed', naming 'count follows an appended item' and 'count follows an empty list down'; end to end the document regressed to count 0 / items 2 and the READER refused it (exit 1, 0 packages emitted). READER leg: predicted red at the FIRST must-throw case since selfTest() throws rather than collecting -- observed 'Error: payload: count 0 beside 2 items was accepted', and the contradictory document was silently accepted again (exit 0, @objectstack/spec selected), reproducing the pre-fix behaviour on demand.\n\nRULING 3 PROOF: old (origin/main, hash-verified 7619ffcc37b9c3a6dedd5d4d503be88f9e5e432c) vs new partitioner over the same real 77-package document, --exclude @objectstack/dogfood: shard 1 23/76 weight 781, shard 2 29/76 weight 780, shard 3 24/76 weight 780, all bins 781/780/780 -- package lists IDENTICAL on all three, 76 packages placed. Unioned-document path unchanged too: 'shard 1/3: 1/2 packages, weight 414', @objectstack/spec, before and after.\n\nH2 REPO-WIDE READ CHECK -- nothing reads `count`, confirmed beyond the two scripts: zero matches repo-wide for packages.count / packages?.count / bracket forms (excluding node_modules, dist, .git, .turbo, coverage); zero `jq` steps touching count in any workflow; zero committed *.json fixtures containing a \"count\" key at all -- so removal would NOT have been a test change, but that is moot given the schema evidence. The document's entire lifecycle is ci.yml lines 342-385: written by turbo ls to $RUNNER_TEMP/turbo-ls.json, mutated by --union-into, read by partition-test-shards.mjs. Never jq'd, never uploaded as an artifact, never read by anything else. No `packageCount` anywhere.\n\nH3 VERDICT -- the emptiness guard does NOT distinguish the two cases, and structurally cannot at that layer; the shape assertion does fire for its stated job. Measured: items key removed -> exit 1; items an object -> exit 1; EMPTY (count 0, items []) -> exit 0, 'shard 1/3: 0/0 packages', 0 stdout lines; NON-EMPTY (count 2) -> exit 0, 1 line; SELF-CONTRADICTORY (count 0, items 2) -> exit 0, silently accepted (pre-fix). Zero stdout then hits ci.yml's `No packages on this shard -- nothing to test.` short-circuit: a green Test Core that ran nothing. But zero is FREQUENTLY the correct answer here, unlike #4690's scanner -- a docs/scripts-only PR genuinely affects no package, and with 3 shards any run with fewer than 3 packages legitimately empties one -- so a blanket red-on-empty at the partitioner would be wrong. Real mitigations already exist: every FAILING producer falls back to the FULL package list (three branches in ci.yml), and steps run under `bash -e`, so a crashing selection cannot yield an empty shard. The residual hole is a producer exiting 0 with a wrong empty answer, and the one decidable signal for it is unchecked: ci.yml checks the exit status of `git diff --name-only \"$SCM_BASE\" HEAD` but never its EMPTINESS, and an empty changed-file list on a pull_request event is not a legitimate state. Filed as #10057 rather than fixed, per the card. Note the shard attestation (#6082) does not cover it -- it attests 'shard N ran and every step passed', which is exactly what a shard that tested nothing does.\n\nH4 FULL FIELD AUDIT -- the document has exactly five fields, verified exhaustively (Object.keys at root and packages level; all 77 items share ONE key set, name+path). root = {packageManager, packages}; packages = {count, items}; item = {name, path}. No timestamp, no filter echo, no packageCount. Does the union step keep each true? packageManager 'pnpm9' -> YES, a workspace-level fact unaffected by appending items (now pinned by a self-test case). packages.items -> YES, the append is the intended mutation. items[].name -> YES, same convention. packages.count -> WAS NO, this card. items[].path -> NO, a SECOND field left untrue: turbo emits repo-RELATIVE paths (0 of 77 absolute) while unionInto pushes join(REPO_ROOT, dir), i.e. absolute. Inert today (measured: weight 414 either way from the repo root) and deliberately NOT fixed here -- path feeds countTestFiles() feeds the weight feeds shard placement, so touching it could move packages between shards, which ruling 3 fences; the correct normalisation is also a genuine judgment call (turbo-faithful relative is cwd-fragile -- measured weight 0 from another cwd, silently absorbed by the LPT partitioner -- while all-absolute stops being a turbo payload). Filed as #10056.",
     "open_questions": [],
     "out_of_scope_findings": [
       "filed as #10056: `--union-into` pushes an ABSOLUTE packages.items[].path while turbo ls emits repo-relative ones (0 of 77 turbo entries absolute), so the document is heterogeneous; inert today but path feeds the weight feeds shard placement, so fixing it could move packages between shards -- and the correct normalisation is a judgment call, not mechanical",
       "filed as #10057: nothing distinguishes 'zero packages affected' from 'the affected-set computation silently produced nothing' -- ci.yml checks the exit status of `git diff --name-only \"$SCM_BASE\" HEAD` but never its emptiness, and an empty changed-file list on a pull_request event is a decidable 'selection failed' signal"
     ]
   }

   ---

   Generated by Claude Code
6. os-steve

   closed this as completedin #10059on Aug 19, 2026
7. claude

   removed

   pm:dispatched

   on Aug 20, 2026