Repository navigation
refactor(app-shell): Explain access picks its object and operation with the shared Select (objectui#11865) - #11990
Merged
objectstack-fleet[bot] merged 2 commits intoOct 8, 2026
Conversation
…th the shared Select (objectui#11865) The Explain access panel's two native selects, Object (the package's objects) and Operation, become one local ExplainPicker on the shared Select. Items carry their option's index, so the "select an object" option (value '') needs no empty SelectItem, and an object the package does not list is shown in the trigger instead of the first option. Claude-Session: https://claude.ai/code/session_01MgfduSkFrfM3eorB3UGfAU Co-authored-by: Claude <noreply@anthropic.com>
…quest they send (objectui#11865) A new suite pins both pickers as the Radix trigger named by their labels, every option against the explain request the native control's change produced, re-pick setting nothing, an object outside the package shown in the trigger, and Enter opening and selecting. The package-scoped dropdown test in AccessExplainPanel.test.tsx picks through the trigger instead of a native select. Patch changeset for @object-ui/app-shell. Claude-Session: https://claude.ai/code/session_01MgfduSkFrfM3eorB3UGfAU Co-authored-by: Claude <noreply@anthropic.com>
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
|
objectstack-fleet
Bot
deleted the
claude/issue-11865-explain-access-select
branch
October 8, 2026 16:52
This was referenced Oct 8, 2026
akarma-synetal
pushed a commit
to akarma-synetal/objectui
that referenced
this pull request
Oct 9, 2026
…ed Select (objectstack-ai#12012) Part of objectstack-ai#11865 Clause-②: no Slice h of claim `6069665700`, the console's own pages: four browser-native select elements on three pages become the shared Radix `Select` (`packages/components/src/ui/select`, used, not edited), following objectui#11976, objectstack-ai#11990, objectstack-ai#11995, objectstack-ai#12000 and objectstack-ai#12005. This PR converts these files and no others: - `apps/console/src/pages/developer/PublicFormsPage.tsx`: "FormView" in *Publish a FormView* and "After submit" in the edit dialog, through one module-private `FormsPicker`. - `apps/console/src/pages/developer/FlowRunsPage.tsx`: the flow picker, through a module-private `FlowPicker`. - `apps/console/src/pages/system/ProfilePage.tsx`: "Preferred language", through a module-private `LanguagePicker`. Each picker follows the card's earlier pickers. Items carry their option's index, so the two options whose value is the empty string ("— Select a FormView —", "Use the deployment default") need no empty `SelectItem`. A value no option carries gets an item of its own and is shown as itself. No picker is exported, and no listing logic on Public Forms moved (objectui#11545 is untouched). ## Hypotheses, measured - **H1, value parity: holds.** A scratch probe (not committed) on base `902fbe190`, with the native selects and the same fixtures as the pins, read what every option led to. The pins compare against those readings as JSON text: - Public Forms, publish: each view writes the same `meta.saveItem` call (`["view","lead.intake",{…,"sharing":{"enabled":true,"allowAnonymous":true,"publicLink":"/forms/contact-us"}}]`, and the same for the view with no label or object). "— Select a FormView —" leaves Publish disabled with no call. - Public Forms, "After submit": 12 rows (stored behaviour unset, `redirect` with `/done`, or `continue`, by each of the four options), each the same full `saveItem` call, `submitBehavior` included. - Flow Runs: each flow, picked, is the one Run Flow executes (`[["nightly_sync",{"params":{}}]]`) and whose runs are listed (`[["nightly_sync",{"limit":20}]]`), and the same for `reassign_wizard`. - Profile: 13 rows (stored tag none, `ja` or `pt-BR`, by each option), each the same `adapter.update` call; "Use the deployment default" still saves `{"locale":null}`. Re-picking the current option leaves Save disabled, as before. - **H2, names, keyboard and read-only: holds.** - Names, read on base through `getByRole('combobox', { name })`: "FormView", "After submit" and "Preferred language" come from each field's Label htmlFor. Each trigger now carries that `id`, and the name pins are green on base and head by design. - The flow picker had no label, `id` or `aria-label`, so its name was empty. It is still empty, and is pinned so (see Acceptance notes). - Keyboard: Enter opens and Enter on an option selects, pinned once per page. - Read-only: only the Profile card has such a state. Its trigger is the primitive's disabled trigger (objectui#11781), pinned by `ProfilePage.language`'s "degrades to read-only" case. Public Forms' two dialogs open only from the page's own buttons, and Flow Runs has no such state. - **H3, a value no option carries: holds.** On base the native control showed its first option in each case. Head shows the value itself, pinned three times: - Flow Runs, a picked flow that a Refresh no longer lists: base showed "Reassign (reassign_wizard)" while the page held `nightly_sync` and showed no runner. - "After submit" with a stored kind none of the four names: base showed "Show a thank-you panel" with no thank-you fields. - Profile, a pick the offered list has since dropped: base showed "Use the deployment default" while Save wrote `zh`. - **H4, first-load bytes: neutral.** - Measured before the first edit, on base: the three pages ship in lazy chunks (`PublicFormsPage-*`, `FlowRunsPage-*` and `ProfilePage-*`), none among the 336 members of `dist/eager-closure.json`. The positive controls were the entry chunk and `ui-components`, both eager. - On head: eager gzip 3,391,326 B on base and 3,391,327 B on head, and eager raw 11,496,190 B on both, so the delta is +1 B gzip and 0 B raw. - The one eager member that changed is the entry chunk. It names the three lazy chunks, whose hashes moved. `ui-components` keeps its file name, so the shared `Select` was already in it. ## Tests New pins (one per page): `PublicFormsPage.sharedSelect-11865.test.tsx`, `FlowRunsPage.sharedSelect-11865.test.tsx`, `ProfilePage.sharedSelect-11865.test.tsx`. Suites that drove or mount these pages, changed only in their harness: - `PublicFormsPage.redirect`: the editor helper picks "Redirect to a URL" through the primitive. - `ProfilePage.language`: picks through the primitive. Its adapter mock is now one object, as the app's provider hands out. A fresh one per render re-read the row and reset the pick before Save could be reached. The native control's synchronous change event used to win that race. - `developerMetadataEnvelope.contractEnvelope-6917` and `consoleFaces.displayLocale-9909`: `getClient()` now returns one client, as the real adapter's does. With a fresh client per call, Flow Runs re-ran its loads on every render. Measured with that harness shape: 251 `meta.getItems` calls in 600 ms on the base page and 73 on this branch, against 2 with one client. The native option text was readable between reloads; the shared trigger's value never settled, so their Flow Runs cases went red. Reverse leg, from the committed head `699224076`: - The three page files were put back on their base blobs (hash-checked on disk) and the new pins run: `Tests 42 failed | 3 passed (45)`. The three passing are the name pins, green on both sides by design. - Restored with `git checkout HEAD`. The blobs on disk equal HEAD's, and `git diff HEAD` is empty. A trap restored on EXIT, INT and TERM. ## Gates (head `f0ec3325f`, worktree clean) - Repo-root `pnpm exec vitest run` over the 3 new pins, the 8 suites that mount these pages, `submitRedirect.test.ts`, `column-identity.ratchet.test.ts` and `one-authority-per-exported-name-6273.test.ts`: `Test Files 14 passed (14)`, `Tests 180 passed (180)`. - `pnpm --filter @object-ui/console type-check`, after `turbo run build --filter='@object-ui/console^...'` (`Tasks: 34 successful, 34 total`): exit 0. `tsc --listFilesOnly` lists every touched file. - `pnpm exec eslint` on the 10 touched ts/tsx files: 0 errors and 32 warnings, all pre-existing rule kinds. The base blobs carry 33, and `PublicFormsPage` loses one `no-explicit-any`. - `pnpm check:eager-closure` on the head build: "Console eager closure is 3311.8 KB gzipped across 336 of 2458 chunks (budget: 3312.0 KB, headroom: 0.2 KB)." - Each exit 0: - `node scripts/check-changeset-presence.mjs`: "10 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s)". - `node scripts/check-changeset-no-major.mjs`. - `pnpm check:new-line-citations`: "0 new citation(s)". - `pnpm check:control-bytes`. - Derived from the touched paths: `check:vi-mock-specifiers`, `check:vi-mock-inherit`, `check:vi-mock-override-shape`, `check:test-path-roots`, `check:i18n-keys`, `check:changeset-claims`, `check:pending-changeset-literals`, `check:phantom-deps`, `check:metadata-write-doors`, `check:unreferenced-sources` and `check:handler-key-reads`. Changeset: `.changeset/11865-h-console-pages-shared-select.md`, `'@object-ui/console': patch`. Docs: no guide or README describes these controls. A `git grep` of `content/docs` and the console README for the page names and labels finds only `public-forms.md`, which documents the route and the redirect contract, not the dialog's controls. So no doc change. ## Acceptance notes None of these is filed. Each is noted with no carrier. - The flow picker has no accessible name, before and after: no label, `id` or `aria-label` is associated with it. Kept for parity, as the claim asks. Naming it is a separate change. - Picking a flow reloads the flow list, because `loadFlows` depends on the picked name. The picker unmounts while the list loads, so focus does not come back to it after a pick. Read in jsdom, the same before and after. - Flow Runs keys its flow and run loads on the identity of `getClient()`. The real adapter returns one client, so this is dormant in the app; only a harness that builds a fresh client loops. `FlowRunsPage.test.tsx` still builds one per call and passes. - A stored `submitBehavior.kind` outside the spec's four is saved away: Save writes the view without `submitBehavior`, before and after. The spec refuses such a kind, so reaching this needs an off-spec stored view. - The FormView option for a view with no object keeps the trailing space the native option had. Session: `https://claude.ai/code/session_01MgfduSkFrfM3eorB3UGfAU`. --- _Generated by [Claude Code](https://claude.ai/code/session_01MgfduSkFrfM3eorB3UGfAU)_ --------- Co-authored-by: Claude <noreply@anthropic.com>
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 free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Part of #11865
Clause-②: no
The Explain access slice of objectui#11865 (claim 6063378622): the panel's two browser-native select elements, Object and Operation, become the shared Radix
Select. The card stays open for the other native selects its body counts. The earlier slices are objectui#11934 (the Group panel) and objectui#11976 (the validation editor).What changed
packages/app-shell/src/views/metadata-admin/AccessExplainPanel.tsx: one module-privateExplainPicker, drawn with the sharedSelectfrom@object-ui/components, renders both pickers. It follows the validation editor'sRulePicker: items carry their option's index, so "Select an object…" (value'', whichSelectItemrefuses) needs no stand-in value, and a value none of the options carries gets an item of its own. The trigger carries the control'sid, so each label'shtmlForstill names it.Inputfallback when the package lists no objects, the user picker, the record selector, the explain request, the props and the exports. No i18n key is added; the panel's set ofengine.*keys is identical to the base's (diffed).AccessExplainPanel.sharedSelect-11865.test.tsx(new), andAccessExplainPanel.test.tsx, whose package-scoped dropdown test now opens and picks through the trigger. It still checks the same three things: the options listed, the free-text placeholder gone, and the object the request sends..changeset/11865-explain-access-shared-select.md:@object-ui/app-shell: patch.Why Select and not Combobox for the object list
Triage allows
Comboboxwhere a list is long, and a package's object list is open-ended, so it was built first and measured. Both readings come from Chromium, on a scratch page that mounted this panel with a package of 25 objects. The scratch page was deleted and never committed.Combobox(built, then kept out of the branch): a 400 px mouse wheel over its open list leftscrollTopat 0, with the list'sscrollHeight840 andclientHeight300. The panel lives in aSheet, and there the list did not scroll under the wheel.Select: the same wheel moved its viewport from 0 to 208, the maximum forscrollHeight840 andclientHeight632. Typingshowcase_tin the open list jumped toshowcase_task.The
Comboboxalso reports a re-pick of its current option as''. So the object list is aSelect, like the operation list.Dispatch hypotheses
H1, value parity: held. The base panel was read first with a scratch probe (not committed): a
changeon each native option, then Explain. The new suite pins every option of both pickers against the request body read there:One caveat: the probe's
changeevent fires even for the current option, so the re-pick rows pin the primitive's behaviour, not a browser reading of the native one.H2, labels and keyboard: held.
getByRole('combobox', { name: 'Object' })and{ name: 'Operation' }find the two triggers (the base probe found the native controls by the same names).StudioDesignSurface) passesopen,onOpenChangeandpackageId, and it writes no metadata. So objectui#11781 has nothing to reach here.H3, a value outside the options: held.
crm_lead) showed "Select an object…" (selectedIndex0), while Explain sentcrm_lead.crm_lead, and it is listed first. Picking it changes nothing, and Explain still sendscrm_lead.defaultObject.H4, first-load bytes: measured, −1 gzip byte.
apps/consolevite build, thennode scripts/check-eager-closure-budget.mjs, on this head and on the same tree with the panel at its base blob5aade6a7, the only bundle input the branch changes.indexchunk: 0 raw bytes, −1 gzip byte.studioBuilderchunk on both legs.Tests and gates, on head
0c5300d7cpnpm exec vitest run --maxWorkers=2over the new suite,AccessExplainPanel.test.tsxandAccessExplainPanel.principalPerson-11862.test.tsx, the only suites that mount the real panel (theStudioDesignSurfacesuites stub it): "Test Files 3 passed (3)", "Tests 28 passed (28)".packages/core/src/utils/__tests__/column-identity.ratchet.test.tsandscripts/__tests__/one-authority-per-exported-name-6273.test.ts: "Test Files 2 passed (2)", "Tests 18 passed (18)".pnpm --filter @object-ui/app-shell type-check(echoestsc --noEmit && tsc -p tsconfig.test.json) exits 0. It ran after the@object-ui/app-shell^...closure build (29 projects, exit 0).--listFilesOnlyshows the panel in the main program and all three panel suites in the test program.pnpm exec eslinton the three touched files exits 0. The JSON format reports 3 files, 0 errors and 8 warnings. The base blobs of the two existing files report the same 8 (no-explicit-anyandset-state-in-effectin untouched code), and the new suite reports none.node scripts/check-changeset-presence.mjs("3 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s)");node scripts/check-changeset-no-major.mjs;pnpm check:new-line-citations("0 new citation(s)");pnpm check:control-bytes;check-changeset-fixed,check-changeset-overwrite,check-changeset-claims;check:pending-changeset-literals,check:test-path-roots;check:vi-mock-specifiers,check:vi-mock-inherit,check:vi-mock-override-shape;check:phantom-deps,check:unreferenced-sources,check:i18n-keys,check:i18n-designer-parity,check:metadata-write-doors.pnpm testfarm and repo-widepnpm lint.Reverse leg. The implementation was committed first (
a551c491f). The panel was then written from its base blob, and the landing was proven: 2 native select elements, 0ExplainPicker, and a hash equal to the base blob. The new suite andAccessExplainPanel.test.tsxthen read "Tests 18 failed | 5 passed (23)":That is the direction the suite header predicts. The trap restored the file with
git checkout HEAD --. The restore was proven: the hash equals the HEAD blob andgit diff HEADis 0 bytes.Acceptance notes
Comboboxinside aSheet. Its open list did not scroll under the mouse wheel, measured once in Chromium on the scratch page above. This bears on this card's own remaining work: a later slice that moves a long list inside a dialog or sheet toComboboxwould meet it. It is not filed from here, because no public entry onmainwas measured.ObjectRefFieldandRecipientPickerFieldimport theCombobox, and whether either one renders inside a modal was not read.Combobox's search reads the option value only. It passes no keywords to cmdk, so an option's label (a translated object label, for example) is not searchable. This was read from source and not measured.defaultObjecthas no caller in the tree. In the tree, the outside-the-options state is reached by text typed into the free-text input before the package's object list arrives (read from source, not measured).Generated by Claude Code