Add support for React router v8 - #11289

React Router v8 merged react-router-dom into react-router and requires
React 19. Rather than adding a compatibility layer and partial
reimplementation inside ra-core (per review feedback on marmelab#11289), v8
support ships as a standalone, opt-in adapter package mirroring
ra-router-tanstack.

The adapter is a thin pass-through over react-router v8's native API:
the obsolete v6/v7 `future` flags are dropped (now defaults in v8) and
imports come from `react-router` instead of `react-router-dom`. ra-core
keeps its built-in react-router v6/v7 adapter as the default, so existing
apps are unaffected.

The package is ESM-only because react-router v8 is ESM-only.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

After merging the ra-router-react-router-v8 package, the in-core
react-router adapter no longer needs the compatibility shims:

- reactRouterProvider imports everything from `react-router`, with only
  `Link` and `createHashRouter` from `react-router-dom`
- Form.stories uses `HashRouter` from `react-router-dom` directly
- Remove CompatHashRouter, CompatLink and their tests
- Restore `react-router-dom` to the dependencies of ra-core,
  ra-ui-materialui and react-admin, and narrow the react-router range
  back to ^6.28.1 || ^7.1.1 (v8 is supported via the separate package)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

No code in ra-ui-materialui imports react-router-dom; all routing
imports come from react-router. ra-core declares its own
react-router-dom peer for its router adapter.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Individual zshy builds had rewritten the exports field to the flat
pre-normalization form. Restore the nested import/require form that
update-package-exports produces and that matches upstream.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…-exports

update-package-exports forces dual import/require exports on every
package; the ESM-only v8 adapter must keep its ESM-only exports (it has
no .cjs build). Also keep minor README/story/doc wording.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…pter

Drop the extra basenameRef; useNavigate now reads basename from the hook
and lists it in the useCallback dependency array, matching the tanstack
adapter's shape. The navigateRef (react-router-specific, issue marmelab#7634)
stays. Behavior unchanged; 107/107 in both adapters.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ters

Follow the tanstack adapter's useNavigate skeleton exactly: numeric
(go back/forward) branch first, then the resolvePath helper, then the
object branch, then the string path — with the same comments. Only the
call differs (navigateRef.current(to) instead of router.history.go(to))
since react-router's navigate handles all three input shapes. Keeps the
three router adapters diff-aligned.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Match the tanstack adapter's Link: branch on the object form first
(`if (typeof to === 'object' && to !== null)`) and fall back to
`resolvePath(to as string)` in the else, instead of a nested ternary.
Behavior unchanged; 107/107 in both adapters.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Build a resolvedPath string via the tanstack Navigate shape (string vs
object branch, append search/hash, then the inline basename-prepend
block) and render <Navigate to={resolvedPath} replace state />. Drops the
to.state merge since react-router's Navigate `to` (a Path) has no state
field — state stays a prop. 107/107 in both adapters.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Pass through the rest of the Navigate props ({...rest}) instead of
listing replace/state explicitly, and drop the "(like react-router does)"
aside from the basename comment. 107/107 in both adapters.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Rename the provider-local useProviderBasename to useBasename and publish
the basename via BasenameContext in both RouterWrapper branches. Fix
quote style to satisfy prettier. 107/107 in the react-router adapter.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Use fireEvent.click for navigation in both router adapter specs, matching
the tanstack provider spec's convention, and keep userEvent.setup/user.type
only in the two unsaved-changes form tests that need typing. 107/107 in
both adapters.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Restore the tanstack import ordering (PathlessLayoutRoutes before
NestedResourcesPrecedence) in both router adapter specs. No behavior
change; 107/107 in both adapters.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… parity)

The earlier blanket fireEvent.click conversion wrongly flipped the one
test that tanstack drives with user.click ("navigate back to parent app").
Match tanstack per-test: user.click + userEvent.setup there, fireEvent.click
everywhere else. 107/107 in both adapters.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Add the "// Wait for data to load before clicking" comment that tanstack
has between the two findByText calls in the navigate-within-nested-routes
test, in both router adapter specs.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Remove the extra explanatory comment atop the matchPath block and reword
the useCanBlock test to "should return true for React Router", matching
the tanstack spec wording. 107/107 in both adapters.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Rename to "should decode only path separator in URL-encoded splat values"
in both router adapter specs. 107/107.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Rename the test to "should decode only path separator in URL-encoded
params", restore tanstack's "// UTF-8 characters: 衣類/衣類 encoded" comment,
and drop the react-router-specific note. 107/107 in both adapters.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Rename the three tests to match the tanstack useBlocker wording: "should
block navigation when form is dirty", "should allow navigation when
clicking proceed", and "should not block navigation when form is not
dirty". 107/107 in both adapters.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The "should block navigation when form is dirty" test was conflating the
block (confirm fires) with the cancel outcome (declining keeps you on the
form). Split into separate "should block navigation when form is dirty"
and "should cancel navigation when clicking cancel" tests, ordered like
tanstack (block, proceed, cancel, not-dirty). 108/108 in both adapters.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Match the tanstack useBlocker tests, which have no inline comments in the
test bodies. 108/108 in both adapters.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…er-dom

Part of the React Router v8 migration: v8 merged react-router-dom into
react-router. Move Route/Routes/RouterProvider/BrowserRouter doc imports
to react-router, while keeping createBrowserRouter and Link on
react-router-dom for v6 compatibility.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

BrowserRouter is not exported from the react-router core package in v6
(it is DOM-only, like Link and createBrowserRouter). Revert the two
BrowserRouter imports back to react-router-dom; Routes/Route stay on
react-router.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

`NODE_OPTIONS=--experimental-vm-modules` is process-global, so setting it on
the shared `test-unit`/`test-unit-ci` scripts pushed the entire default
(CommonJS) jest project into ESM mode. Jest then honored the `"type": "module"`
field of transformed dependencies and refused to `require()` them, so every
suite that transitively imports an ESM-only dep (e.g. react-hotkeys-hook, pulled
in via the layout/button/form barrels) failed to load with "Must use import to
load ES Module". This took out 93 suites across ra-ui-materialui, react-admin,
ra-input-rich-text and ra-i18n-polyglot — the suites failed to run while their
individual tests never executed.

Keep the root config a single CommonJS project (no `projects` array, no flag)
and run ra-router-react-router-next's ESM/React 19 tests as a separate jest
invocation against its own jest.config.cjs, appended to the test scripts with
the flag set only for that run.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…-next

The package imports react-router at runtime, so it belongs in `dependencies`
(matching the sibling ra-router-react-router), not only in dev/peer deps.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

ra-core is provided by the react-admin meta-package at runtime; the sibling
ra-router-react-router does not declare it either.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Restore ra-core as a peerDependency of ra-router-react-router-next (it is
provided by the host app, like the other ra-router adapters), and declare
react-router as a peerDependency of ra-no-code.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

react-admin imports from react-router (via ra-router-react-router), not
react-router-dom directly, so the devDependency is unnecessary.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The auth spec asserted on the post-logout/post-login URL with
cy.url().then(url => expect(url).to.contain(...)), which evaluates the
assertion exactly once before the redirect has necessarily landed,
causing intermittent CI failures. Switch to the retrying cy.url().should()
form already used elsewhere in the file.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>