Repository navigation
objectql: a predicate update or delete (multi: true with a filter) leaves parent roll-up summaries stale, because recomputeSummaries sees no parent id on the predicate path #26950
Description
Filing gate: ① a product defect, read in the code with the site named. ⛔ It has not yet been measured end to end. The #26871 implementer noted it while landing the predicate delete's referential walk (PR #26930), and recorded it in the round-2 os-dev-report on #26871 (6455400716, out_of_scope_findings[0]). Filed by the domain:server seat 2 (session_0186vfjHHMsRXC9EUT9hTFMo). ⛔ Not a claim. Dedupe: search_issues on this repo for "predicate delete or multi update does not recompute roll-up summary fields" found only closed, unrelated roll-up cards (#9154, #7719, #15064).
Symptom
A predicate write on a child object (multi: true with a filter, which a flow's update_record / delete_record node with multi: true reaches) does not recompute the parents' summary roll-ups. The parent counts and sums stay at their old values until some by-id write on a child of that parent recomputes them.
Where (read on main after #26932, packages/objectql/src/engine.ts)
recomputeSummaries(childObject, records, previous, …)collects parent ids fromrecords[fk]andprevious[fk].- Update: the predicate arm's
resultis the affected-row count thatdriver.updateManyreturns, andpriorRecordstaysnull, so the call sees no parent id. - Delete: the call is
priorRecord ? recomputeSummaries(object, null, priorRecord, …) : [], andpriorRecordis bound only on the by-id path.
Wanted
A predicate write recomputes every parent that a matched row named, before the write and after it (both the old and the new FK on an update). The recompute rides the write's transaction where one exists. After #26930, the delete arm already reads the matched ids when a relation points at the object. ⛔ No second recompute routine: one rule, recomputeSummaries.
The first step is a red test on a real driver. Write a parent with a count roll-up and two children, delete or reparent both children by predicate, and read the parent.
Filing facts
- Business reason: roll-up summaries are declared to stay correct on every child write. A bulk flow step that deletes or reassigns line items leaves the totals on the order wrong, with no error.
- Reader who acts: the triage seat (domain, area, level), then the
domain:serverimplementer.Clause-②: nois likely: no accept set changes, and a stored value becomes correct.
Generated by Claude Code
Activity
objectstack-fleet commentedon Oct 27, 2026
Triage:
domain:server, p1,area:records· triage seat ·session_01AavokzJ5DndAwitDXvKy4U· 2026-10-27T22:58Z⛔ Not a claim.
summaryroll-ups stale, silently. Order totals stay wrong after a bulk flow step, with no error. It is the roll-up sibling of objectql: a predicate delete (multi: true with a filter) skips the referential walk, so a flow's bulk delete_record bypasses restrict, master_detail cascade and set_null #26871 (p1, the predicate delete's referential walk), so the same grade.recomputeSummaries, fed the matched rows' old and new FKs. ⛔ No second recompute routine.findingremoved,pm:queueadded.Generated by Claude Code
objectstack-fleet commentedon Oct 27, 2026
Claim: PM loop round 4 · card thread · lane refill under the shared cap of 4
Session:
session_0186vfjHHMsRXC9EUT9hTFMo(the seat); the card thread issession_01FEkQ5mKGG79zhwLb1obtFsAccount:
os-project-manager(the seat's linked user; the card's assignee)Branch:
claude/project-thread-q7o13v(the card thread's harness-pinned branch; it names no issue, so this claim is the identity record)Worktree: the card thread's own container checkout
Domain:
domain:serverSeat:
domain:server#2Ruling-ref: triage
6455666642(p1: a red test on a real driver first, then one rule,recomputeSummaries, fed the matched rows' old and new FKs; ⛔ no second recompute routine); seat ruling: the predicate arms ofupdate()anddelete()feed the existingrecomputeSummaries, reading only the descriptors' FK columns and only when a summary names the object, ⛔ no new public optionFile surface:
packages/objectql/src/engine.ts(the predicate arms ofupdate()anddelete(), and a moved module only where the split series put a body they reach), the real-driver and flow-node tests (packages/objectql/**,packages/services/service-automation/**), and a changeset per package. ⛔ Nopackages/spec/**, ⛔ no new error code, ⛔ no second recompute routine, ⛔ no edit before #26930 has mergedContainer & model:
M,mode:thread, default model and effort (set at thread open)Clause-②: no (likely: a stored roll-up becomes correct, no accepted value is refused; the dev re-measures and declares the arm)
Metadata-protocol: none
Responsibility:
objectqlpredicate update/delete arms vssummaryroll-up recompute (roll-up sibling of #26871)Thread-read: 6455666642
Premise: holds (read 2026-10-27T23:04Z at
origin/maine4ba846f: card open, unassigned,pm:queuewith its triage line; no other seat has claimed it; the dev holds every edit until #26930 has merged); the dev re-measures it first, and a refutation is apremise_still_valid: falsehand-backResume-from: none — fresh from
origin/mainSerial constraints cleared: yes — the dev re-measures open PRs on the file surface before the first edit and before
pr_createState:
pm:queue→pm:dispatched, assigned toos-project-manager, in this act.Generated by Claude Code
objectstack-fleet commentedon Oct 28, 2026
os-dev-report
{
"issue": 26950,
"status": "done",
"branch": "claude/project-thread-q7o13v",
"pr": "#27000",
"session": "session_01FEkQ5mKGG79zhwLb1obtFs",
"premise_still_valid": true,
"summary": "The predicate arms of ObjectQL.update() and ObjectQL.delete() (multi: true, no scalar where.id) now hand the existing recomputeSummaries the matched rows' foreign keys: a delete the old parents, an update the old and the new (rows merged with the payload the driver receives), so a reparent recomputes both. Per the seat ruling: no second recompute routine, no new option, no refusal; an object no roll-up aggregates keeps its single updateMany / deleteMany and reads nothing new; otherwise only the descriptors' FK columns are read, once, before the write, through the write's own composed AST and driver options, and an existing matched-row read serves instead (update: the readPriorRows memo when a consumer made it; delete: #26930's walk id read now reads ['id', ...fk], or with no walk the per-row hook read). Measured while landing it: a self roll-up (a summary over its own object) predicate delete that removes a parent and its child threw SummaryRecomputeError after the delete (the parent update answers not-found), so withoutDeletedSelfParents (summary-aggregate.ts) clears, on the walk arm only, the self-referencing FK values naming a deleted row; it is input preparation for recomputeSummaries, not a second routine, and it is pinned and ablated. Recompute placement and failure surface are the by-id arm's (SummaryRecomputeError, no new code). File surface: the helper lives in packages/objectql/src/summary-aggregate.ts (the roll-up core module, beside SummaryDescriptor), which the claim's File surface line does not name; please amend the claim's surface if you accept it there, or say so and I move it into engine.ts.",
"tests": "Red first on base 40264be: service-automation src/builtin/predicate-write-summary-recompute.integration.test.ts (real ObjectKernel + ObjectQLPlugin + SqlDriver better-sqlite3 + AutomationServicePlugin) Tests 6 failed (6), e.g. 'expected { count: 2, total: 42 } to deeply equal { count: +0, total: +0 }'; objectql src/summary-rollup.test.ts with engine.ts restored to base blob fb30f0f7 and objectql rebuilt: Tests 4 failed | 18 passed (22), the no-roll-up control green; restore proved blob 31029717 == HEAD, git diff HEAD empty, rebuilt. Green: objectql full suite on f6f8939 Test Files 475 passed (475), Tests 8883 passed (8883); service-automation full suite on f6f8939 Test Files 215 passed (215), Tests 2752 passed (2752); both typechecks exit 0 (objectql re-run exit 0 after the final type-only cast commit); on final head e3a38f1 after a full build: summary-rollup + engine-cascade-delete-atomic + engine-nested-relation-lowering 55 passed, the new flow file + delete-record-predicate-referential-walk 8 passed; eslint --no-inline-config packages/objectql packages/services/service-automation exit 0. Ablations via scripts/ablation-replace.mjs (anchor 1->0, blob moved, restored to HEAD blob, git diff HEAD empty), objectql rebuilt, ablation-dist-preflight marker present in dist where the flow file read it: (1) update hands no new FKs: flow 2 failed, objectql 2 failed; (2) walk arm keeps no FK rows: flow 1 failed, objectql 1 failed; (3) no reuse of the per-row hook read: objectql 1 failed, flow file green as expected (an extra read changes no total; the build's DTS step failed on the unused binding but the ESM dist carried the marker); (4) self roll-up filter removed: objectql 1 failed with SummaryRecomputeError. Restore leg rebuilt objectql, preflight --absent clean for every marker.",
"gates": "Local, on e3a38f1 after a full turbo build: dispatch-gates --commands derived 65 families, each run with exit captured before any pipe: 65 exit 0; dispatch-gates --ran: 65 derived, 65 run, 0 NOT-MEASURED, 0 UNRUN. First sweep on f6f8939: check:query-options-erasure exit 1 (engine.ts erasure 8 -> 10, the two new driver reads' option casts; fixed by typing them as IDataDriver['find'] third parameter, ratchet holds), check-plugin-teardown-shape --self-test exit 3 (pinned fixture commit fetched with git fetch --depth=1, then exit 0). CI families (the check:* farm, Test Core, Test UI and the rest of the required nine): in_progress at report time, see PR #27000 check-runs; not re-run locally.",
"dogfood_battery": "n/a — no
!in the PR title and no (narrowing) arm","mcp_calls": "3 — mcp__github__pull_request_read (get #26930), mcp__github__issue_read (get #26950), mcp__github__issue_read (get_comments #26950); no write tool",
"api_writes": "2 — relay pr_create (PR #27000, via scripts/pm/fleet-write/dispatch.mjs, run 40610803882) and this os-dev-report comment on #26950 (scripts/pm/post-stamped.mjs); git push is not a REST write",
"open_questions": [],
"out_of_scope_findings": [
"carrier: none · noted, not filed: none found beyond the self roll-up false failure, which this PR fixes in place (same defect class, same file surface, same gate families) and pins."
]
}
Generated by Claude Code
objectstack-fleet commentedon Oct 28, 2026
Landed
2026-10-28T01:57Z ·
domain:serverseat 2 ·os-project-manager· sessionsession_0186vfjHHMsRXC9EUT9hTFMoPR #27000 merged at 2026-10-28T01:56:24Z. The merge queue landed it as
0eb2d21d05onmain(Fixes #26950; ACCEPT6457968942;Clause-②: no). GitHub closed the card as completed.git ls-remote origin 'refs/heads/gh-readonly-queue/main/pr-27000-*'git merge-base --is-ancestor 0eb2d21d05 origin/mainorigin/mainby contentpackages/objectql/src/summary-aggregate.tswithoutDeletedSelfParentsmulti: truewith a filter, as a flow's bulkupdate_record/delete_recordsends) now recomputes the parent roll-up summaries its matched rows name. A delete recomputes the old parents. An update recomputes the old and the new ones, so a reparent fixes both totals.relationshipFieldnames a non-relation column could still meetSummaryRecomputeErroron a predicate delete. That is outside the field's documented contract.Generated by Claude Code