Uh oh!
There was an error while loading. Please reload this page.
ComponentMeta at the registry: derive it from the one declaration, extend it under a named type - #6297
Merged
Conversation
…aration `packages/core/src/registry/Registry.ts` carried its own thirteen-key `ComponentMeta`: nine members restated from `@object-ui/types`' `base.ts`, four registry-only keys added here, and `tags` / `description` absent — though both are declared on the canonical type and on the `ComponentMetaSchema` zod mirror. Two of the three authorities agreed and the registration surface did not, so those two keys were unwritable at exactly the declaration most component registrations import. Same two-key delta objectui#5893 closed inside `@object-ui/types`, on a third declaration. The nine shared members are no longer restated. `ComponentMeta` is now `CanonicalComponentMeta & RegistryComponentMetaExtras`, where the newly exported `RegistryComponentMetaExtras` carries `tier` / `namespace` / `skipFallback` / `labelling` — the four keys that are genuinely registry mechanics and have live consumers. Nothing is removed or renamed; `tags` and `description` are added. Pinned by key set rather than by assignability. Every member of both shapes is optional, so `extends` is mutually true across the diverged pair and an assignability assertion is green on the defect. Measured on the emitted `.d.ts` of both packages, `Exclude<keyof Canonical, keyof Core>` moved from `"tags" | "description"` to `never` while both `extends` readings stayed `true`. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Mn4BZ5AVDM81pvfij1WwM9
Contributor
✅ Console Performance Budget
The eager closure is every chunk the entry reaches through static imports — what the browser fetches and parses before the app renders. The entry chunk on its own is a small fraction of it. 📦 Bundle Size Report
Size Limits
|
os-litant
commented
Aug 25, 2026
CollaboratorAuthor
The out-of-scope finding named at the end of the description is filed as #6298 ( Also: the downstream consumer type-check row in the description was re-run at the final commit Generated by Claude Code |
Contributor
✅ Console Performance Budget
The eager closure is every chunk the entry reaches through static imports — what the browser fetches and parses before the app renders. The entry chunk on its own is a small fraction of it. 📦 Bundle Size Report
Size Limits
|
os-litant
marked this pull request as ready for review
August 25, 2026 10:53
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.
Fixes#6067
@object-ui/core'sComponentMetano longer restates the canonical declaration. It is now derived from it, andtags/descriptionreach the registration surface.1. Premise, re-derived on today's
main— it holdsRe-derived from scratch at merge-base
c456d91f4, not carried from the card.Declaration census. A repo-wide search for a
type/interface/classdeclaration of the exact name returns exactly two:packages/types/src/base.ts:511export interface ComponentMeta {— the canonical onepackages/core/src/registry/Registry.ts:37export type ComponentMeta = {— the structural copySo core's is the only remaining structural copy, exactly as the 04:18Z triage comment graded it, and the count the first triage comment (16:29Z) asked to be re-verified on the merged ref is confirmed: #5893 collapsed
packages/types' two onto one, and the card's title ("a THIRD declaration") is the pre-#5893 count.Control probe for the census, because the whole task is counting declarations and a zero-hit is not a reading on its own: the same pattern run against the sibling type
ComponentInputreturns exactly one declaration (packages/types/src/base.ts:379) — the state #5671 left it in — and the pattern matches both theexport typeand theexport interfacespellings, which is where the card's own filing probe went wrong.Key sets, re-counted — canonical 11, core 13, symmetric difference 6, all confirmed:
tags,descriptiontier,namespace,skipFallback,labellinglabel,icon,category,inputs,defaultProps,examples,isContainer,resizable,resizeConstraints— includingresizeConstraints, whose six members were identical2. The disposition: converge, not rename
The ruling admitted two resolutions. This PR converges, and the reason is that rename cannot be executed here without a contract break:
@object-ui/coreis published, andComponentMetais part of its export surface:src/index.tscarriesexport * from './registry/Registry.js', andpackage.jsonmaps"."→./dist/index.d.ts.The other rejected option was moving the four registry keys onto
@object-ui/types'ComponentMetaand re-exporting outright, the way #5671 handledComponentInput.skipFallbackandnamespacedescribe how the registry keys an entry, not what a component is, and@object-ui/types'ComponentMetais the general, plugin-facing, AI-facing type. They stay where they are read, under their own named type.Nothing is removed and nothing is renamed. All fifteen keys on the resulting type are writable; two of them are new.
3. The four registry-only keys ARE published behaviour with consumers
Requested by the dispatch as the measurement that decides whether converge-with-extension is safe. It is, and this is why a rename would not have been.
namespacepackages/**Registry.ts:247,Registry.ts:315— the bare-key fallback decisionskipFallbackaction-bar.tsx,action-button.tsx,record-approvals-renderer.tsx, …)Registry.ts:247,Registry.ts:315, same decisionlabellingFIELD_WIDGET_LABELLING(packages/fields/src/index.tsx:2934); 14 of them (10group+ 4display) write the key into the meta atindex.tsx:3009—'control'is spelled as absencepackages/components/src/renderers/form/form.tsx:446—ComponentRegistry.getMeta(…)?.labellingpackages/fields/src/index.tsx:2936,packages/app-shell/…/DashboardWidgetInspector.tsx:581andpackages/app-shell/…/widgets.tsx:2547all take an indexed accessComponentMeta['labelling']off the type published by@object-ui/coretierPUBLIC_BLOCKSlist;tier: 'public'is the documented per-registration opt-in (ADR-0080)Registry.ts:560,Registry.ts:564(getPublicConfigs),packages/sdui-parser/src/index.ts:148(publicOnly)labellingadditionally appears in three shipped CHANGELOGs as "@object-ui/core—ComponentMetagainslabelling?: 'control' \| 'group'", so it is published behaviour on the record, not merely in-repo usage.Direct importers of the type itself:
packages/fields/src/index.tsx:11,packages/app-shell/…/widgets.tsx:44,packages/app-shell/…/DashboardWidgetInspector.tsx:33.4. The measurement that shaped the pin:
extendsis a ghost hereEvery member of both shapes is optional, so a type with fewer optional members is assignable to one with more and vice versa. Mutual assignability therefore holds across the diverged pair — an
extends-based mirror assertion is green on the defect and could never have caught it.Measured with a TypeScript program resolved against the emitted
.d.tsof both packages (packages/types/dist/index.d.tsandpackages/core/dist/registry/Registry.d.ts), before and after:Core extends CanonicaltruetrueCanonical extends CoretruetrueExclude<keyof Canonical, keyof Core>"tags" | "description"neverExclude<keyof Core, keyof Canonical>"tier" | "namespace" | "skipFallback" | "labelling"Exactly one reading moved. That reading is the pin; the assignability pair is kept beside it in the test, labelled, as the control that demonstrates the contrast.
This also means the triage wording "asserting a mirror relationship that is false in both directions" is true of the key sets and false of TypeScript assignability. Recorded because acting on the assignability reading would have produced a test that cannot fail.
5. Pinning against the emitted declaration, not the source
The dispatch flagged the #6267 trap: a source-level assertion passing while the shipped
.d.tsis wrong. Measured rather than assumed for this package:@object-ui/core'sbuildis a baretsc, not a.d.tsbundler, so a module re-export survives into the emit rather than being inlined. Natural control:ComponentInput: re-export the one declaration instead of restating it three times #5671'sexport type { ComponentInput } from '@object-ui/types';appears verbatim atdist/registry/Registry.d.ts:31on the pre-change tree. After this PR,dist/registry/Registry.d.tslikewise carriesimport type { ComponentMeta as CanonicalComponentMeta } from '@object-ui/types';andexport type ComponentMeta = CanonicalComponentMeta & RegistryComponentMetaExtras;— so the shipped declaration references the canonical one by module specifier. There is one declaration of the shared eleven in the published artifact, not a copy of it.packages/core/tsconfig.test.json, which sets"paths": {}so@object-ui/typesresolves through the workspace dependency's builtdist/index.d.ts. That project is chained from this package'stype-checkscript, and turbo'stype-checktaskdependsOn: ["^build"], so CI's Type Check job builds the dependency before compiling it. The canonical half of every assertion is therefore read off the emitted declaration.packages/core/dist. CI'stestjob runspnpm test(rootvitest run) with no build ahead of it at all, so a dist-reading assertion would be absent-or-red depending on whether someone ran a build — the standard this repo already records insrc/actions/__tests__/actionKeys.types.test.ts: an assertion whose colour depends on whether someone ran a build is not a pin. Thepackages/core/disthalf is covered by the out-of-band measurement above and reported here rather than encoded as a green-by-luck test.6. Evidence — per-point ablation
Direction predicted before each run; both predictions held. Each ablation restored under
trap … EXIT INT TERMwith a cwd-independentgit -C … checkout HEAD --, and each restore printed an emptygit diff HEAD --stat. Mutations proved on disk by grepping the injected and the removed text separately, with anchor uniqueness asserted before the edit. Committed before ablating.A. Revert only the convergence (
Registry.tsback toorigin/main; test and changeset kept)Predicted: the key-set pin and the
tags/descriptioncounter-probe go red; the assignability control stays green.tsc -p tsconfig.test.json→ exit 2, six diagnostics, all naming the pin file:Root
vitest→ exit 1,Tests 2 failed | 5 passed (7):imports the canonical declaration instead of restating it,declares none of the canonical members locally.Still passing after the revert — reported because it is the point of the exercise:
is mutually assignable with the canonical declaration— no diagnostic at all undertsc, and green under vitest. This is the ghost, demonstrated on the real defect.adds exactly the four registry-only keys, named(line 113) — no diagnostic: the pre-convergence type also carried exactly those four, so this assertion is correctly indifferent to the convergence.still declares the four registry-only members here— green: the old declaration still declared them.tags/descriptioncounter-probe pass under vitest, because annotations are erased there.tscis their enforcement, which is why both legs are reported.B. Delete exactly one registry-only key (
labelling) fromRegistryComponentMetaExtrasPredicted: the named-keys assertion goes red while the canonical-key pin stays green — proving the two are independent instruments, not one assertion written twice.
Anchor asserted unique (1 occurrence) before the edit; landing site
1 file changed, 1 deletion(-).tsc -p tsconfig.test.json→ exit 2, three diagnostics:Root
vitest→ exit 1,Tests 1 failed | 6 passed (7):still declares the four registry-only members here.Still passing:
leaves no canonical key unreachable at the registration surface(line 101 — no diagnostic), the assignability control, and — informatively —extrasSupplyThem(line 122), which compares the extras type against itself and so shrinks with it. That contrast is exactly why the four keys are pinned to a literal union as well.7. Gates run locally
All exit codes captured before any pipe. Union run at
8ef81a944, the final commit.turbo run build --filter=@object-ui/core --force2 successful, 2 totalpnpm --filter @object-ui/core type-check(tsc --noEmit && tsc -p tsconfig.test.json)pnpm exec vitest run packages/core/(repo root)Test Files 100 passed (100),Tests 2023 passed (2023)pnpm exec vitest run packages/core/src/registry/__tests__/component-meta-derives-from-canonical.test.tsTests 7 passed (7)turbo run type-check --filter='...@object-ui/core' --filter='!@object-ui/site' --forceTasks: 71 successful, 71 totalnode scripts/check-changeset-presence.mjsnode scripts/check-changeset-no-major.mjsmajorbump."pnpm --filter @object-ui/core lintno-explicit-anywarnings; the new test file contributes none)node scripts/check-control-bytes.mjsnode scripts/check-type-check-coverage.mjsnode scripts/check-lint-coverage.mjsnode scripts/check-phantom-dependencies.mjsnode scripts/check-package-self-import.mjsnode scripts/check-vi-mock-specifiers.mjsnode scripts/check-doc-component-types.mjsnode scripts/check-shell-escape-residue.mjsFilter direction demonstrated, not asserted — the consumer direction is the prefix form, and it is the one used above:
Declared narrowings.
pnpm test/pnpm lint/pnpm type-checkrepo-wide,check:readme-exports,check:doc-snippets,check:esm-specifiers,check:node-esm-load,check:published-distandcheck:eager-closurewere not run locally; CI runs the farm regardless. The downstreamtype-checkrow above is the load-bearing one for a published-type change and it was run in full.@object-ui/siteis excluded from the downstream row for runtime; it consumes@object-ui/coreonly through documentation snippets, whichcheck:doc-snippetscovers in CI.8. Scope
packages/core/src/registry/Registry.ts,packages/core/src/registry/__tests__/component-meta-derives-from-canonical.test.ts,.changeset/6067-component-meta-derive-from-canonical.md. Nothing outsidepackages/core/src/registry/**plus the mandatory changeset.packages/typeswas not edited. Convergence did not require it: the canonical declaration already carries everything core needed, so importing it was sufficient. That package is held by fix(types): StackSchema ships its declared members instead of collapsing under BaseSchema's index signature #6267.content/docs/guide/component-registry.mdis the only doc that namesComponentMeta; it shows a registration literal and does not enumerate the key set, so nothing it states became false. Enumerating the newly reachable keys there would be a docs change outside this card's file surface.Out-of-scope finding, not fixed here
ComponentConfigis the same species one type down:packages/core/src/registry/Registry.ts:141declaresexport type ComponentConfig<T = any> = ComponentMeta & { … }whilepackages/types/src/base.ts:578declaresexport interface ComponentConfig extends ComponentMeta { … }— two same-named published declarations, differing in genericity and in whichComponentMetahalf they build on. This PR reduces it (the shared eleven are now single-sourced throughComponentMetaon both) without closing it. Filed separately, unassigned, as afinding. #4631 does not cover it — that card is about a component type's three disagreeing surfaces (schema type / registryinputs/ renderer reads), and namesComponentConfignowhere.Generated by Claude Code