@object-ui/fields 17.3.0 → 17.6.0

package/src/widgets/__tests__/TextAreaField.characterCount.i18n.test.tsx

@@ -1,287 +0,0 @@
-/**
- * ObjectUI
- * Copyright (c) 2024-present ObjectStack Inc.
- *
- * This source code is licensed under the MIT license found in the
- * LICENSE file in the root directory of this source tree.
- */
-
-/**
- * `TextAreaField`'s character counter announces itself in the session locale —
- * objectui#3406.
- *
- * The counter block is rendered only when the field declares `maxLength`, and
- * what a sighted user sees (`{n}/{max}`) is digits, so nothing about this was
- * visible on screen. The block carries `aria-live="polite"`, and its accessible
- * name was the English literal `Character count: ${n} of ${max}` — so a zh/ja/ar
- * session had an English sentence read out on every keystroke, and only screen
- * reader users could perceive it.
- *
- * Unlike objectui#3404 (keys existed, this path did not consume them), no
- * `characterCount` key existed in any of the ten packs. `fields.textarea.characterCount`
- * is new here, in all ten, with `{{count}}` / `{{max}}`.
- *
- * ## What the cases assert, and why in this shape
- *
- * - **`en` is a NO-OP pin, not a defect reproducer.** The default in
- *   `FIELD_DEFAULTS` is byte-identical to the literal it replaced, so the
- *   English case was green before this change too. It is here to prove the
- *   swap did not alter English rendering — only a positive assertion sees that
- *   direction. Its exact expected string is repeated verbatim in
- *   `TextAreaField.characterCount.no-provider.test.tsx`; together the two pin
- *   pack-default equality across a boundary the packages cannot import across.
- * - **Non-`en` positive AND English-literal negative, together.** A re-inlined
- *   literal sitting next to a translated sibling still satisfies a
- *   positive-only assertion.
- * - **`ja` carries the load-bearing ORDER check.** Its translation puts the cap
- *   BEFORE the count ("of {{max}} characters, {{count}}"). An implementation
- *   that assembled the sentence from parts in code — "label + count + separator +
- *   max" — cannot produce that order, so it passes a zh-only check by luck and
- *   fails here. This is the case that justifies one interpolated key over a
- *   per-part assembly.
- * - **`ru` and `ar` are the plural-resolution check.** `count` is not an
- *   ordinary interpolation variable to i18next: passing it activates plural
- *   lookup (`key_one` / `key_few` / `key_many` / …) BEFORE the base key. Both
- *   packs declare the base key only, and both languages have multi-form plural
- *   rules, so these two prove the base-key fallback actually resolves rather
- *   than leaving the user with a raw key. `{{count}}` is the spelling this
- *   repo already uses for counts (`table.selected`, `lookup.recordCount`), and
- *   this pins that it keeps working for a pack with no plural forms declared.
- * - **Recompute-on-input.** The name is derived per render. Pinned so a future
- *   memo cannot freeze it at the mount-time length while the visible digits
- *   move on.
- *
- * ## objectui#3408 — same key, different place, plus a sibling
- *
- * The sentence is no longer an `aria-live` region's `aria-label`. It is the
- * textarea's accessible DESCRIPTION (`aria-describedby`), read once on focus
- * instead of re-announced on every keystroke, and the cases above resolve it
- * through that association. The expected strings did not change: the ten packs
- * still serve them, and this file still guards the same drift (backfilled
- * English, a code-assembled sentence that cannot reproduce ja's word order, a
- * raw key surfacing where plural lookup fails).
- *
- * `fields.textarea.charactersRemaining` is the sibling #3408 added — the only
- * thing the widget still SPEAKS while typing, and therefore the one the
- * English-backfill risk now applies to. Its cases sit at the bottom, with fake
- * timers, because reaching it requires crossing the near-limit threshold and
- * then pausing.
- */
-import React from 'react';
-import { describe, it, expect, vi, afterEach } from 'vitest';
-import { render, screen, fireEvent, act } from '@testing-library/react';
-import '@testing-library/jest-dom';
-import { I18nProvider } from '@object-ui/i18n';
-
-import { TextAreaField } from '../TextAreaField';
-import type { FieldMetadata } from '@object-ui/types';
-
-const textareaField = (extra: Record<string, unknown> = {}) =>
-  ({ name: 'notes', type: 'textarea', ...extra }) as unknown as FieldMetadata;
-
-function renderIn(language: string, element: React.ReactElement) {
-  return render(
-    <I18nProvider config={{ defaultLanguage: language, detectBrowserLanguage: false }}>
-      {element}
-    </I18nProvider>,
-  );
-}
-
-/**
- * Where the sentence lives moved in objectui#3408: it was the `aria-label` of
- * the visible counter, which was ALSO the `aria-live` region — so the whole
- * sentence was re-announced on every keystroke. It is now the textarea's
- * accessible DESCRIPTION, read once when focus lands.
- *
- * These helpers therefore resolve it the way an assistive technology does —
- * follow `aria-describedby`, concatenate the referenced nodes — instead of
- * reading an attribute off a node found by selector. That keeps the
- * ASSOCIATION under assertion here too, and it is why the cases below did not
- * need their expected strings changed: the same ten-pack key still produces
- * them, in a place a screen reader reaches without typing.
- *
- * The digits stay queryable separately (`visibleCounter`) because several
- * cases assert that the visible half is unchanged. The behaviour half of
- * #3408 — how often the live region speaks — is pinned in
- * `TextAreaField.characterCount.announcements.test.tsx`.
- */
-const visibleCounter = () => document.querySelector('[data-testid="textarea-character-count"]');
-const describedText = () => {
-  const box = screen.getByRole('textbox');
-  const ids = (box.getAttribute('aria-describedby') || '').split(/\s+/).filter(Boolean);
-  return ids.map((id) => document.getElementById(id)?.textContent ?? '').join(' ').trim();
-};
-
-describe('TextAreaField character counter is translated (objectui#3406)', () => {
-  it('renders the English sentence under an en provider', () => {
-    renderIn(
-      'en',
-      <TextAreaField value="hello" onChange={vi.fn()} field={textareaField({ maxLength: 200 })} />,
-    );
-
-    // Byte-identical to the literal this replaced, so `en` is a no-op change.
-    expect(describedText()).toBe('Character count: 5 of 200');
-    expect(visibleCounter()).toHaveTextContent('5/200');
-  });
-
-  it('renders the sentence in Chinese under a zh provider', () => {
-    renderIn(
-      'zh',
-      <TextAreaField value="hello" onChange={vi.fn()} field={textareaField({ maxLength: 200 })} />,
-    );
-
-    expect(describedText()).toBe('已输入 5 个字符，最多 200 个');
-    // The literal, asserted absent — a re-inlined one would still satisfy a
-    // positive-only check on a sibling case.
-    expect(describedText()).not.toMatch(/character count/i);
-    // The digits stay language-independent; only the accessible name moved.
-    expect(visibleCounter()).toHaveTextContent('5/200');
-  });
-
-  it('renders the sentence in Japanese, with the cap interpolated BEFORE the count', () => {
-    renderIn(
-      'ja',
-      <TextAreaField value="hello" onChange={vi.fn()} field={textareaField({ maxLength: 200 })} />,
-    );
-
-    // `{{max}}` precedes `{{count}}` in this pack. Code-side concatenation
-    // cannot produce this order — that is the point of the assertion.
-    expect(describedText()).toBe('文字数: 200 文字中 5 文字');
-    expect(describedText()!.indexOf('200')).toBeLessThan(describedText()!.indexOf('5'));
-    expect(describedText()).not.toMatch(/character count/i);
-  });
-
-  it('resolves the base key under ru, whose plural rules i18next consults first', () => {
-    renderIn(
-      'ru',
-      <TextAreaField value="hello" onChange={vi.fn()} field={textareaField({ maxLength: 200 })} />,
-    );
-
-    expect(describedText()).toBe('Количество символов: 5 из 200');
-    // The failure this guards: plural lookup missing and no base-key fallback
-    // would surface the raw key instead of a sentence.
-    expect(describedText()).not.toContain('fields.textarea');
-  });
-
-  it('resolves the base key under ar as well', () => {
-    renderIn(
-      'ar',
-      <TextAreaField value="hello" onChange={vi.fn()} field={textareaField({ maxLength: 200 })} />,
-    );
-
-    expect(describedText()).toBe('عدد الأحرف: 5 من 200');
-    expect(describedText()).not.toContain('fields.textarea');
-  });
-
-  it('recomputes the spoken count as the user types', () => {
-    // Derived per render, not memoized at mount. A frozen name would keep
-    // announcing "5" while the visible digits read 11.
-    const Host = () => {
-      const [v, setV] = React.useState('hello');
-      return <TextAreaField value={v} onChange={setV} field={textareaField({ maxLength: 200 })} />;
-    };
-    renderIn('zh', <Host />);
-    expect(describedText()).toBe('已输入 5 个字符，最多 200 个');
-
-    fireEvent.change(screen.getByRole('textbox'), { target: { value: 'hello world' } });
-
-    expect(describedText()).toBe('已输入 11 个字符，最多 200 个');
-    expect(visibleCounter()).toHaveTextContent('11/200');
-  });
-
-  it('keeps reading the legacy snake_case max_length spelling', () => {
-    // The dual-read predates this change (framework#1878 §3). Keying the
-    // sentence must not narrow which spelling reaches it.
-    renderIn(
-      'zh',
-      <TextAreaField value="hello" onChange={vi.fn()} field={textareaField({ max_length: 80 })} />,
-    );
-
-    expect(describedText()).toBe('已输入 5 个字符，最多 80 个');
-  });
-
-  it('renders no counter at all when the field declares no maxLength', () => {
-    // The block is `{maxLength && …}`. Translating it must not make it
-    // unconditional — an always-present live region would announce on every
-    // keystroke in every long-text field in the app.
-    renderIn('zh', <TextAreaField value="hello" onChange={vi.fn()} field={textareaField()} />);
-
-    expect(visibleCounter()).toBeNull();
-  });
-
-  it('renders no counter in the readonly branch', () => {
-    // The translation hook had to move ABOVE the readonly early return
-    // (rules-of-hooks). This pins that the hoist changed nothing a user sees
-    // on that branch.
-    renderIn(
-      'zh',
-      <TextAreaField
-        value="hello"
-        onChange={vi.fn()}
-        readonly
-        field={textareaField({ maxLength: 200 })}
-      />,
-    );
-
-    expect(visibleCounter()).toBeNull();
-    expect(screen.getByText('hello')).toBeInTheDocument();
-  });
-});
-
-/**
- * The near-limit warning — the sentence that IS spoken mid-typing
- * (objectui#3408). Reaching it needs the value inside the last 10% / 20
- * characters of the cap AND a pause, so these render already-near-full values
- * and advance fake timers rather than typing 490 characters.
- */
-describe('TextAreaField near-limit warning is translated (objectui#3408)', () => {
-  afterEach(() => vi.useRealTimers());
-
-  // Scoped to its own container and unmounted before returning: some cases
-  // call this twice, and a document-wide query would read the FIRST mount's
-  // region — green for the previous language, not the one under test.
-  const nearLimit = (language: string) => {
-    vi.useFakeTimers();
-    const { container, unmount } = renderIn(
-      language,
-      <TextAreaField value={'x'.repeat(190)} onChange={vi.fn()} field={textareaField({ maxLength: 200 })} />,
-    );
-    act(() => { vi.advanceTimersByTime(1000); });
-    const spoken = container.querySelector('[aria-live="polite"]')?.textContent ?? '';
-    unmount();
-    vi.useRealTimers();
-    return spoken;
-  };
-
-  it('renders the English sentence under an en provider', () => {
-    expect(nearLimit('en')).toBe('Characters remaining: 10');
-  });
-
-  it('renders the sentence in Chinese under a zh provider', () => {
-    expect(nearLimit('zh')).toBe('还可输入 10 个字符');
-    // A backfilled English string is invisible at runtime — the whole defect
-    // class is English reaching a non-English screen reader.
-    expect(nearLimit('zh')).not.toMatch(/characters remaining/i);
-  });
-
-  it('renders the sentence in Japanese', () => {
-    expect(nearLimit('ja')).toBe('残り 10 文字');
-  });
-
-  it('resolves the base key under ru and ar, whose plural rules i18next consults first', () => {
-    // Same mechanism as `characterCount` above: `count` sends i18next to
-    // `key_one` / `key_few` / … before the base key, and these packs declare
-    // the base key only. A raw key reaching a screen-reader-only string is
-    // invisible to everyone who could report it.
-    expect(nearLimit('ru')).toBe('Осталось символов: 10');
-    expect(nearLimit('ar')).toBe('الأحرف المتبقية: 10');
-    expect(nearLimit('ru')).not.toContain('fields.textarea');
-    expect(nearLimit('ar')).not.toContain('fields.textarea');
-  });
-});
-
-// The provider-LESS leg lives in `TextAreaField.characterCount.no-provider.test.tsx`.
-// It cannot share this file: mounting `I18nProvider` installs its instance as
-// react-i18next's GLOBAL default, so after any case above there is no "no
-// provider" state left in this module graph to observe (measured in
-// objectui#3404 — see that file's header).

package/src/widgets/__tests__/TextAreaField.characterCount.no-provider.test.tsx

@@ -1,160 +0,0 @@
-/**
- * ObjectUI
- * Copyright (c) 2024-present ObjectStack Inc.
- *
- * This source code is licensed under the MIT license found in the
- * LICENSE file in the root directory of this source tree.
- */
-
-/**
- * `TextAreaField`'s character counter with NO `I18nProvider` mounted
- * (objectui#3406) — standalone/embedded hosts, and most of this package's own
- * bare-render tests.
- *
- * ## Why this leg constrains the implementation
- *
- * Routing an accessible name through i18n must not turn it into a raw key
- * where nothing is configured — for a screen-reader-only string that failure
- * is invisible to everyone who could report it. Bare `useObjectTranslation()`
- * with no provider returns the key verbatim AND drops interpolation, which is
- * strictly worse than the English literal being replaced (measured on the
- * pre-change tree in objectui#3404, same hook, same package). So the widget
- * consumes `useFieldTranslation()` — `createSafeTranslation` plus English
- * defaults, which does its own `{{k}}` substitution.
- *
- * The expected sentence below is repeated verbatim in the `en`-provider case
- * of `TextAreaField.characterCount.i18n.test.tsx`. `packages/i18n` cannot
- * import `packages/fields` (the dependency runs the other way), so this pair of
- * literals is what pins pack and code default to the same bytes — and both to
- * the literal that existed before this change.
- *
- * ## Why a separate FILE and not another `describe`
- *
- * Mounting `I18nProvider` calls `initReactI18next`, which installs that
- * instance as react-i18next's GLOBAL default. After any sibling case has done
- * so there is no "no provider" state left in the module graph, and these
- * assertions would read that leaked instance instead of the fallback — green
- * for the wrong reason. Measured in objectui#3404: cases written inside the
- * provider file stayed GREEN under a reverse check that correctly reddened a
- * separate file. vitest's per-file isolation is what makes the observation
- * honest. Same split, same reason as
- * `FullscreenFieldEditor.no-provider.test.tsx` and
- * `TagsField.placeholder.no-provider.test.tsx`.
- */
-import { describe, it, expect, vi, afterEach } from 'vitest';
-import { render, screen, act } from '@testing-library/react';
-import '@testing-library/jest-dom';
-
-import { TextAreaField } from '../TextAreaField';
-import type { FieldMetadata } from '@object-ui/types';
-
-const textareaField = (extra: Record<string, unknown> = {}) =>
-  ({ name: 'notes', type: 'textarea', ...extra }) as unknown as FieldMetadata;
-
-const visibleCounter = () => document.querySelector('[data-testid="textarea-character-count"]');
-const statusRegion = () => document.querySelector('[aria-live="polite"]');
-
-/**
- * objectui#3408 moved the sentence off the live region's `aria-label` and onto
- * the textarea's accessible DESCRIPTION. Resolved here the way an assistive
- * technology resolves it — follow `aria-describedby` — so the association is
- * under assertion alongside the copy.
- */
-const counterLabel = () => {
-  const box = screen.getByRole('textbox');
-  const ids = (box.getAttribute('aria-describedby') || '').split(/\s+/).filter(Boolean);
-  return ids.map((id) => document.getElementById(id)?.textContent ?? '').join(' ').trim();
-};
-
-describe('character counter — English fallback survives with no i18n configured (objectui#3406)', () => {
-  afterEach(() => vi.useRealTimers());
-
-  it('renders the English default, not a raw key', () => {
-    render(
-      <TextAreaField value="hello" onChange={vi.fn()} field={textareaField({ maxLength: 200 })} />,
-    );
-
-    // Byte-identical to the literal this change replaced.
-    expect(counterLabel()).toBe('Character count: 5 of 200');
-    // The measured failure mode, stated directly — a key reaching a user.
-    expect(counterLabel()).not.toMatch(/fields\.textarea/);
-  });
-
-  it('substitutes BOTH placeholders in the fallback path', () => {
-    // `createSafeTranslation`'s fallback runs its own `{{k}}` replacement.
-    // Bare `useObjectTranslation` would leave the sentence unsubstituted
-    // (in fact would not return a sentence at all).
-    render(
-      <TextAreaField value="hi" onChange={vi.fn()} field={textareaField({ maxLength: 40 })} />,
-    );
-
-    expect(counterLabel()).toBe('Character count: 2 of 40');
-    expect(counterLabel()).not.toContain('{{');
-  });
-
-  it('announces zero for an empty value rather than a hole in the sentence', () => {
-    // `0` is falsy, so an interpolation that guarded on truthiness would drop
-    // it and speak "Character count:  of 40" — the one input length where the
-    // substitution can fail on its own.
-    //
-    // Deliberately `''` and not `undefined`: `FieldWidgetComponentProps`
-    // declares `value: T`, non-optional. The widget's `value || ''` guard
-    // predates this change and covers a JS host that ignores the type; pinning
-    // an out-of-contract shape here would be the lenient-consumer move
-    // AGENTS.md #0.1 rules out, and is not what this issue is about.
-    render(<TextAreaField value="" onChange={vi.fn()} field={textareaField({ maxLength: 40 })} />);
-
-    expect(counterLabel()).toBe('Character count: 0 of 40');
-  });
-
-  it('keeps a live region, now SEPARATE from the visible digits', () => {
-    // objectui#3406 pinned `aria-live="polite"` on the counter itself and said
-    // so explicitly: the per-keystroke re-announce was a behaviour question,
-    // filed separately, deliberately untouched. objectui#3408 is that filing,
-    // and this pin is updated in the direction it decided — the live region
-    // still exists (that attribute is why anything is ever spoken) but it is a
-    // visually-hidden node of its own, and the digits it used to share an
-    // element with are now `aria-hidden`.
-    render(
-      <TextAreaField value="hello" onChange={vi.fn()} field={textareaField({ maxLength: 200 })} />,
-    );
-
-    expect(statusRegion()).toHaveAttribute('aria-live', 'polite');
-    expect(statusRegion()).not.toBe(visibleCounter());
-    expect(visibleCounter()).toHaveAttribute('aria-hidden', 'true');
-    expect(visibleCounter()).not.toHaveAttribute('aria-live');
-  });
-
-  it('renders the English near-limit warning too, not a raw key', () => {
-    // The second half of the fallback contract: `charactersRemaining` is the
-    // only sentence the widget still speaks while typing, so a key leaking
-    // here is the same invisible failure — heard by exactly the users who
-    // cannot see that it is a key. 190 of 200 is inside the warning band; the
-    // 1s pause is what releases it.
-    vi.useFakeTimers();
-    render(
-      <TextAreaField
-        value={'x'.repeat(190)}
-        onChange={vi.fn()}
-        field={textareaField({ maxLength: 200 })}
-      />,
-    );
-    act(() => { vi.advanceTimersByTime(1000); });
-
-    expect(statusRegion()).toHaveTextContent('Characters remaining: 10');
-    expect(statusRegion()!.textContent).not.toMatch(/fields\.textarea/);
-    expect(statusRegion()!.textContent).not.toContain('{{');
-  });
-
-  it('the code-shipped default carries no CJK (Commandment #-1)', () => {
-    // With no provider mounted, whatever renders here came from CODE
-    // (FIELD_DEFAULTS). Escaped ranges on purpose: a literal CJK class would
-    // itself violate the commandment this asserts (same shape as
-    // `FullscreenFieldEditor.no-provider.test.tsx`).
-    render(
-      <TextAreaField value="hello" onChange={vi.fn()} field={textareaField({ maxLength: 200 })} />,
-    );
-
-    expect(counterLabel()).not.toMatch(/[\u3000-\u30ff\u4e00-\u9fff]/);
-  });
-});

package/src/widgets/__tests__/TextAreaField.fullscreenCharacterCount.i18n.test.tsx

@@ -1,161 +0,0 @@
-/**
- * ObjectUI
- * Copyright (c) 2024-present ObjectStack Inc.
- *
- * This source code is licensed under the MIT license found in the
- * LICENSE file in the root directory of this source tree.
- */
-
-/**
- * The fullscreen dialog's character-count description follows the session
- * language — objectui#3417.
- *
- * The sentence is the SAME `fields.textarea.characterCount` key the inline
- * surface has used since objectui#3406, and reusing it rather than minting a
- * `fields.textarea.fullscreenCharacterCount` twin is the point: one form-level
- * setting (`ObjectFormSchema.mobile.fullscreenLongText`) projects onto both
- * surfaces of the same field, so two copies of this copy could only drift, and
- * a twin would mean editing ten locale packs to say what they already say.
- * (Same reasoning, same shape, as objectui#3404's decision to have
- * `FullscreenFieldEditor` consume the existing `form.fullscreen.*` keys.)
- *
- * ## Why this needs its own cases at all
- *
- * `TextAreaField.characterCount.i18n.test.tsx` already pins the key and its ten
- * packs — including the load-bearing `ja` word-order case, which no code-side
- * concatenation can satisfy. What it cannot see is whether the FULLSCREEN
- * surface routes through `t()` or re-inlines an English literal on the way to a
- * node only a screen reader reads. A footer that renders
- * `Character count: 5 of 500` in a zh session looks correct to everyone who
- * could file the bug. So: a positive translated assertion AND the English
- * literal asserted absent, on the fullscreen surface specifically.
- *
- * `ja` is carried over for the same reason it exists on the inline side — its
- * pack puts the cap BEFORE the count, so an implementation that assembled the
- * sentence from parts passes a zh-only check by luck and fails here.
- *
- * The provider-LESS leg lives in `TextAreaField.fullscreenCharacterCount.test.tsx`
- * and cannot share this file: mounting `I18nProvider` installs its instance as
- * react-i18next's GLOBAL default, so after any case below there is no "no
- * provider" state left in this module graph to observe (measured in
- * objectui#3404 — cases written inside the provider file stayed GREEN under a
- * reverse check that correctly reddened a separate file).
- */
-
-import React from 'react';
-import { describe, it, expect, vi } from 'vitest';
-import { render, screen, fireEvent } from '@testing-library/react';
-import '@testing-library/jest-dom';
-import { I18nProvider } from '@object-ui/i18n';
-
-import { TextAreaField } from '../TextAreaField';
-import type { FieldMetadata } from '@object-ui/types';
-
-const fieldMeta = (extra: Record<string, unknown> = {}) =>
-  ({
-    name: 'description',
-    label: 'Description',
-    type: 'textarea',
-    mobile_fullscreen: true,
-    ...extra,
-  }) as unknown as FieldMetadata;
-
-/**
- * Render, open the dialog, and return the dialog textarea's accessible
- * DESCRIPTION resolved the way an assistive technology resolves it — follow
- * `aria-describedby`, concatenate the referenced nodes. Scoped to its own mount
- * and unmounted before returning: every case here renders a different language,
- * and a document-wide read after a second mount would report the FIRST one.
- */
-const fullscreenDescription = (language: string, maxLength = 200, value = 'hello') => {
-  const { unmount } = render(
-    <I18nProvider config={{ defaultLanguage: language, detectBrowserLanguage: false }}>
-      <TextAreaField value={value} onChange={vi.fn()} field={fieldMeta({ maxLength })} />
-    </I18nProvider>,
-  );
-  fireEvent.click(screen.getByTestId('textarea-fullscreen-toggle'));
-
-  const input = screen.getByTestId('textarea-fullscreen-input');
-  const ids = (input.getAttribute('aria-describedby') || '').split(/\s+/).filter(Boolean);
-  const text = ids
-    .map((id) => document.getElementById(id)?.textContent ?? '')
-    .join(' ')
-    .trim();
-  const digits =
-    document.querySelector('[data-testid="textarea-fullscreen-character-count"]')?.textContent ??
-    null;
-
-  unmount();
-  return { text, digits };
-};
-
-describe('the fullscreen counter description is translated (objectui#3417)', () => {
-  it('renders the English sentence under an en provider', () => {
-    // A no-op pin: `FIELD_DEFAULTS` and the `en` pack are byte-identical, so
-    // this is here to prove routing through `t()` did not change English
-    // rendering. Only a positive assertion sees that direction.
-    const { text, digits } = fullscreenDescription('en');
-
-    expect(text).toBe('Character count: 5 of 200');
-    expect(digits).toBe('5/200');
-  });
-
-  it('renders the sentence in Chinese under a zh provider', () => {
-    const { text, digits } = fullscreenDescription('zh');
-
-    expect(text).toBe('已输入 5 个字符，最多 200 个');
-    // The literal, asserted absent — a re-inlined one would still satisfy the
-    // positive assertion on a sibling case, and in a screen-reader-only string
-    // nobody who can see it is affected.
-    expect(text).not.toMatch(/character count/i);
-    // The digits stay language-independent; only the description is keyed.
-    expect(digits).toBe('5/200');
-  });
-
-  it('renders the sentence in Japanese, with the cap interpolated BEFORE the count', () => {
-    // `{{max}}` precedes `{{count}}` in this pack. Code-side concatenation
-    // cannot produce this order — that is the point of the assertion.
-    const { text } = fullscreenDescription('ja');
-
-    expect(text).toBe('文字数: 200 文字中 5 文字');
-    expect(text.indexOf('200')).toBeLessThan(text.indexOf('5'));
-    expect(text).not.toMatch(/character count/i);
-  });
-
-  it('resolves the base key under ru, whose plural rules i18next consults first', () => {
-    // `count` sends i18next to `key_one` / `key_few` / … BEFORE the base key,
-    // and these packs declare the base key only. A raw key reaching a
-    // screen-reader-only string is invisible to everyone who could report it.
-    const { text } = fullscreenDescription('ru');
-
-    expect(text).toBe('Количество символов: 5 из 200');
-    expect(text).not.toContain('fields.textarea');
-  });
-
-  it('recomputes the sentence as the DRAFT changes, not the committed value', () => {
-    // Derived per render from the draft. A memo frozen at open time would keep
-    // describing the value the user opened the dialog with while the digits
-    // beside it moved on.
-    render(
-      <I18nProvider config={{ defaultLanguage: 'zh', detectBrowserLanguage: false }}>
-        <TextAreaField value="hello" onChange={vi.fn()} field={fieldMeta({ maxLength: 200 })} />
-      </I18nProvider>,
-    );
-    fireEvent.click(screen.getByTestId('textarea-fullscreen-toggle'));
-
-    const input = screen.getByTestId('textarea-fullscreen-input');
-    const describedText = () => {
-      const ids = (input.getAttribute('aria-describedby') || '').split(/\s+/).filter(Boolean);
-      return ids
-        .map((id) => document.getElementById(id)?.textContent ?? '')
-        .join(' ')
-        .trim();
-    };
-
-    expect(describedText()).toBe('已输入 5 个字符，最多 200 个');
-
-    fireEvent.change(input, { target: { value: 'hello world' } });
-
-    expect(describedText()).toBe('已输入 11 个字符，最多 200 个');
-  });
-});