Filed unassigned — an observation surfaced while implementing #4953's services half (PR TBD, packages/triggers/trigger-record-change/src/record-change-trigger.ts). Recording only; no fix bundled here.
What's declared
RecordChangeDataEngine (the structural interface RecordChangeTrigger uses to avoid a build-time dependency on @objectstack/objectql) declares an optional method:
getObjectConfig?(object: string): { fields?: Record<string, { type?: unknown } | undefined> } | undefined;objectHasFormulaField uses it as a schema gate: when present, it skips hydrateComputedFields's re-read (findOne) for objects that declare no formula field — the only thing that re-read adds. When absent, it falls back to true ("re-reads unconditionally — correctness over the optimization").
What's measured
grep -rn "getObjectConfig" packages/ → only trigger-record-change's own interface + its own test file
The concrete ObjectQL engine (packages/objectql/src/engine.ts) has no method named getObjectConfig. It does have a public getObject(name): ServiceObject | undefined (an alias for getSchema), which the trigger already uses for a different purpose (the "silent miss" object-existence probe), and whose returned ServiceObject.fields would serve the same purpose — but nothing wires it into objectHasFormulaField.
RecordChangeTriggerPlugin.resolveDataEngine hands the trigger the REAL engine straight from ctx.getService('objectql') — there is no adapter layer where a getObjectConfig shim could be quietly attached either.
Consequence
In production, typeof this.engine.getObjectConfig === 'function' is always false, so objectHasFormulaField always takes its true fallback and hydrateComputedFields always re-reads via findOne on every afterInsert/afterUpdate dispatch — including for the very common case of an object with no formula field at all, where the re-read (per the code's own doc comment) adds nothing. The "schema gate" the doc comment describes (packages/triggers/trigger-record-change/src/record-change-trigger.ts, hydrateComputedFields's doc block) is not a bug in the sense of wrong output — the documented fallback is explicitly "correctness over the optimization" — but it is unreachable in the one place that matters, so it reads as a working optimization that in fact never engages.
Only the trigger's OWN test file (record-change-trigger.test.ts, describe('RecordChangeTrigger computed-field hydration guards (#3426 follow-up)')) exercises the gate — by hand-attaching a getObjectConfig mock via Object.assign(engine, { getObjectConfig }). Those tests pass and are true statements about the trigger's OWN logic; they just never run against anything the real engine provides, so the gate's "measured on a real engine" claim implicit in the #3445 changelog entry doesn't hold.
Why this is a finding, not a fix in #4953's PR
Out of scope for #4953 (services half): that card is about the seeded record/previous CEL bindings being total over declared fields, not about hydration's re-read performance. Fixing this (presumably: reuse getObject — already correctly wired, and already used by buildContext for materialization since #4953 — inside objectHasFormulaField too, retiring getObjectConfig) is a real but separate, perf-only change with its own blast radius (every afterInsert/afterUpdate dispatch's query count in production, not just this one seam), so it should land as its own card and its own measurement rather than ride along.
Suggested fix shape (not prescriptive)
Point objectHasFormulaField at this.engine.getObject?.(object)?.fields (the same accessor #4953's materialization now uses) instead of getObjectConfig, and retire the getObjectConfig interface member + its doc comment once nothing references it. Whoever picks this up should measure the actual query-count delta on a real engine before/after (a findOne spy count, same style as the existing hydration tests) rather than assume the fix is free.
Filed unassigned — an observation surfaced while implementing #4953's services half (PR TBD,
packages/triggers/trigger-record-change/src/record-change-trigger.ts). Recording only; no fix bundled here.What's declared
RecordChangeDataEngine(the structural interfaceRecordChangeTriggeruses to avoid a build-time dependency on@objectstack/objectql) declares an optional method:objectHasFormulaFielduses it as a schema gate: when present, it skipshydrateComputedFields's re-read (findOne) for objects that declare noformulafield — the only thing that re-read adds. When absent, it falls back totrue("re-reads unconditionally — correctness over the optimization").What's measured
The concrete ObjectQL engine (
packages/objectql/src/engine.ts) has no method namedgetObjectConfig. It does have a publicgetObject(name): ServiceObject | undefined(an alias forgetSchema), which the trigger already uses for a different purpose (the "silent miss" object-existence probe), and whose returnedServiceObject.fieldswould serve the same purpose — but nothing wires it intoobjectHasFormulaField.RecordChangeTriggerPlugin.resolveDataEnginehands the trigger the REAL engine straight fromctx.getService('objectql')— there is no adapter layer where agetObjectConfigshim could be quietly attached either.Consequence
In production,
typeof this.engine.getObjectConfig === 'function'is alwaysfalse, soobjectHasFormulaFieldalways takes itstruefallback andhydrateComputedFieldsalways re-reads viafindOneon everyafterInsert/afterUpdatedispatch — including for the very common case of an object with noformulafield at all, where the re-read (per the code's own doc comment) adds nothing. The "schema gate" the doc comment describes (packages/triggers/trigger-record-change/src/record-change-trigger.ts,hydrateComputedFields's doc block) is not a bug in the sense of wrong output — the documented fallback is explicitly "correctness over the optimization" — but it is unreachable in the one place that matters, so it reads as a working optimization that in fact never engages.Only the trigger's OWN test file (
record-change-trigger.test.ts,describe('RecordChangeTrigger computed-field hydration guards (#3426 follow-up)')) exercises the gate — by hand-attaching agetObjectConfigmock viaObject.assign(engine, { getObjectConfig }). Those tests pass and are true statements about the trigger's OWN logic; they just never run against anything the real engine provides, so the gate's "measured on a real engine" claim implicit in the #3445 changelog entry doesn't hold.Why this is a finding, not a fix in #4953's PR
Out of scope for #4953 (services half): that card is about the seeded
record/previousCEL bindings being total over declared fields, not about hydration's re-read performance. Fixing this (presumably: reusegetObject— already correctly wired, and already used bybuildContextfor materialization since #4953 — insideobjectHasFormulaFieldtoo, retiringgetObjectConfig) is a real but separate, perf-only change with its own blast radius (every afterInsert/afterUpdate dispatch's query count in production, not just this one seam), so it should land as its own card and its own measurement rather than ride along.Suggested fix shape (not prescriptive)
Point
objectHasFormulaFieldatthis.engine.getObject?.(object)?.fields(the same accessor #4953's materialization now uses) instead ofgetObjectConfig, and retire thegetObjectConfiginterface member + its doc comment once nothing references it. Whoever picks this up should measure the actual query-count delta on a real engine before/after (afindOnespy count, same style as the existing hydration tests) rather than assume the fix is free.