Uh oh!
There was an error while loading. Please reload this page.
fix(objectql): a CEL defaultValue stores the declared type's contract shape, not a raw Date (#7373) - #7469
Merged
Conversation
…ct shape, not a raw `Date` (#7373) `applyFieldDefaults` produces a default three ways and only two honoured the stored-value contract: the `NOW()` token routes through `resolveNowDefault`, a literal is checked against `valueSchemaFor(def, 'stored')` at author time (#7127) — and the expression envelope's result was assigned verbatim. The temporal stdlib returns a JS `Date` (ADR-0053 D1: `today()`/`daysFromNow(n)`/ `daysAgo(n)` are UTC-midnight of the reference-tz calendar day, `now()` the raw instant), so `{ dialect: 'cel', source: 'daysFromNow(7)' }` on a `datetime` put a `Date` OBJECT in the column while `valueSchemaFor` names an ISO-8601 STRING. `validateRecord` accepts a `Date` on `date`/`datetime` by explicit decision, so nothing refused the write and the divergence was silent — while `os migrate value-shapes` walks stored values against that same schema and reports such a row as a violation by the platform's own scan. The expression branch now routes a `Date` result through the SAME per-type table the `NOW()` token uses, rather than growing a second copy of the contract: `datetime` → `YYYY-MM-DDTHH:MM:SS.sssZ`, `date` → `YYYY-MM-DD`, `time` → `HH:MM:SS[.fff]`. Storage on SQL and MongoDB is byte-identical to before: `SqlDriver.formatInput` already coerced a `Date` through `canonicalUtcDatetime`/`toDateOnly`, and mongodb's `storageDatetimeValue`/`storageDateValue` do the same. What changes is the memory driver, which applies its temporal canon to filter comparands only (`coerceTemporalValue`) and stored writes as handed. Same declaration, different stored shape per datasource — the split #4597 / #4560 closed for the `NOW()` token, reappearing on the CEL branch and closed the same way, engine-side. Normalization rather than refusal: refusing a `Date` here would make the rule depend on WHO wrote the value — `validateRecord` accepts one from any caller, temporal types are not in ADR-0104's strict value-shape block, and the documented envelope (#7244) stores correctly on SQL today. Non-`Date` results pass through untouched, as does an `Invalid Date` (the totality the driver canons keep). No day can shift: the ADR-0053 `Date` is UTC-midnight OF the reference-tz day and is read back with UTC getters. Pins in `engine-cel-default-temporal-shape.test.ts` assert the stored value against `valueSchemaFor` itself, and assert the TYPE as well as the text — `JSON.stringify` renders a `Date` as its ISO string, which is what made the original defect read as correct. Reverse-verified: the four subject pins fail against the pre-fix assignment; the four controls (`NOW()` token, literals, non-date CEL results, caller-supplied values) pass in both states. Refs #7373, #7244, ADR-0053, ADR-0104. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012tjfhdVGuYU9KH7SNbor56
The latest updates on your projects. Learn more about Vercel for GitHub. 1 Skipped Deployment
|
Contributor
📓 Docs Drift CheckThis PR changes 1 package(s): 15 hand-written doc(s) reference the affected code and may need an implementation-accuracy re-verification:
⛔ 1 release-owned page(s) also reference the affected code. These are read-only:
|
os-zhuang
marked this pull request as ready for review
August 10, 2026 15:10
This was referenced Aug 10, 2026
Uh oh!
There was an error while loading. Please reload this page.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for freeto join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes#7373.
The defect
applyFieldDefaultsproduces a default three ways, and only two of them honoured the stored-value contract:NOW()tokenresolveNowDefault→ the form the declared type storesvalueSchemaFor(def, 'stored')at author time (#7127)The temporal stdlib returns a JS
Date(ADR-0053 D1:today()/daysFromNow(n)/daysAgo(n)are UTC-midnight of the reference-tz calendar day;now()the raw instant), so{ dialect: 'cel', source: 'daysFromNow(7)' }on adatetimeput aDateobject in the column whilevalueSchemaFornames an ISO-8601 string.validateRecordaccepts aDateondate/datetimeby explicit decision, so nothing refused the write — andos migrate value-shapes, which walks stored values against that same schema, reports such a row as a violation by the platform's own scan.The fix
The expression branch routes a
Dateresult through the same per-type table theNOW()token uses — one table, both branches, not a second copy of the contract:datetime(and anything else)YYYY-MM-DDTHH:MM:SS.sssZdateYYYY-MM-DDtimeHH:MM:SS[.fff]Measurement behind the choice
In-tree CEL
defaultValuecensus — thedialect: 'cel'corpus is almost entirelyexpression:(formula fields, virtual — never stored),condition:/when:(rule predicates). Exactly two aredefaultValue, one temporal:engine-write-formula-hydration.test.ts:157now()datetimeDateobjectengine-select-option-default.test.ts:210'approved'textstring(unaffected)Per-driver behaviour when handed a
Date— measured by reading each driver's write seam:datetimedatecanonicalUtcDatetime→toISOString()toDateOnly→YYYY-MM-DDstorageDatetimeValuekeeps BSONDate; an ISO string parses to the identicalDatestorageDateValue→YYYY-MM-DDcoerceTemporalValue(filter comparands); stores writes as handedDateobject → contract shapeSo SQL and MongoDB already stored exactly what the engine now produces; the only stored-shape change is on the memory driver, which is where the defect was visible. This is the split #4597 / #4560 closed for the
NOW()token, reappearing on the CEL branch and closed the same way — engine-side, so one answer serves every driver.Why normalize rather than refuse. Refusing a
Dateout of a CEL default would make the rule depend on who wrote the value:validateRecordaccepts aDateondate/datetimefrom any caller (if (value instanceof Date) return null), and temporal types are not in ADR-0104's strict value-shape block at all — so there is no strict path that already refuses this on a user write for a defaults path to be bypassing. Refusal would also break the documented envelope (#7244) on precisely the SQL backends where it stores correctly today.No day-shift. ADR-0053 D1's
Dateis UTC-midnight of the reference-tz calendar day, and it is read back withtoISOString()— UTC getters, the samegetUTC*the ADR names for the driver filter path. Reading those parts in local time is the move that would shift a day, and nothing here does it. Pinned directly: at an instant whose UTC day andAmerica/Los_Angelesday differ, adatefield defaulted bytoday()stores the LA day.Non-
Dateresults pass through untouched (a CEL default's result type is otherwise a runtime concern), as does anInvalid Date— the same totality the driver canons keep.Pins
packages/objectql/src/engine-cel-default-temporal-shape.test.ts— 8 tests. They assert stored values againstvalueSchemaForitself rather than a hand-copied shape, and assert the type as well as the text:JSON.stringifyrenders aDateas its ISO string, which is exactly what made the original defect read as correct.Subjects:
datetime,date,timeCEL defaults; the reference-tz day-shift guard.Controls:
NOW()token byte-identical, literal defaults untouched, non-date CEL results passed through, caller-supplied values untouched.Reverse-verified by restoring the pre-fix
out[f.name] = result.value— the 4 subject pins fail, the 4 controls pass in both states.Verification
packages/objectqlsuite: 177 files / 3135 tests passcheck:adr-anchors,check:durability-log-level,check:engine-double-contract,check:stack-collection-maps,scripts/check-engine-split-ratio.mjs,check:nul-bytescheck-empty-changeset(1 declaring changeset added),check-changeset-no-major(no major)tsc --noEmitandeslintclean on the changed filesChangeset
patch, on@objectstack/objectqlalone. Argued from the gate text:check-changeset-no-major.mjsrecords that the launch window ships even breaking changes asminor, somajor(and the ADR-0087 trio) is out regardless. Betweenpatchandminor, the observable change is confined to correcting values the platform's own value-shape scan already classified as violations, on the one backend that stored them — no API, type, or message-contract change. That matches the recent precedent for contract-correcting behaviour fixes (nested-plugin-view-container-expansion,driver-sql-json-column-operator-refusal, bothpatch).No
packages/spectouch — deliberately.resolveNowDefaultkeeps its name so the reference inpackages/spec/src/data/default-value-shape.tsstays accurate; only its parameter is renamednow→instant, sincedaysFromNow(7)is not "now" and takes the identical per-type treatment.Refs #7373, #7244, #7127, #4597 / #4560, ADR-0053, ADR-0104.
Generated by Claude Code