Skip to content
Permalink

Comparing changes

Choose two branches to see what’s changed or to start a new pull request. If you need to, you can also or learn more about diff comparisons.

Open a pull request

Create a new pull request by comparing changes across two branches. If you need to, you can also . Learn more about diff comparisons here.
base repository: vaadin/flow
Failed to load repositories. Confirm that selected base ref is valid, then try again.
Loading
base: 25.2.13
Choose a base ref
...
head repository: vaadin/flow
Failed to load repositories. Confirm that selected head ref is valid, then try again.
Loading
compare: 25.2.14
Choose a head ref
  • 13 commits
  • 31 files changed
  • 9 contributors

Commits on Oct 12, 2026

  1. chore: Update frontend dependencies (25.2) (#26327)

    ## Automated Frontend Dependency Updates
    
    This PR contains automated updates to frontend dependencies.
    
    ### Update Strategy
    - Target Branch: 25.2
    - Updates: Minor and patch releases (no major)
    
    ### Changes
    The dependencies have been updated using `npm-check-updates`.
    
    ---
    *This PR was automatically created by the update-frontend-dependencies
    workflow.*
    
    ---------
    
    Co-authored-by: github-merge-queue <118344674+github-merge-queue@users.noreply.github.com>
    Co-authored-by: Mikael Grankvist <mgrankvi@vaadin.com>
    3 people authored Oct 12, 2026
    Configuration menu
    Copy the full SHA
    633806e View commit details
    Browse the repository at this point in the history
  2. Configuration menu
    Copy the full SHA
    3023482 View commit details
    Browse the repository at this point in the history
  3. fix: throw from removeAttribute for bound attribute with null value (#…

    …26264) (CP: 25.2) (#26350)
    
    This PR cherry-picks changes from the original PR #26264 to branch 25.2.
    
    ---
    
    > Related to #26237
    > 
    > **behavior deviation** · flow-server · apps that bind attributes to
    signals
    > 
    > Removing an attribute that is bound to a signal throws an exception,
    except when the signal's current value is `null`: then the call silently
    does nothing. It now throws in both cases, so the rule does not depend
    on the current signal value.
    > 
    > **Risks:**
    > - ⚠️ Behaviour change: `removeAttribute` on an attribute bound to a
    signal whose value is `null` now throws `BindingActiveException` instead
    of doing nothing. The old behaviour was inconsistent with the non-null
    case and with `setAttribute`.
    > - ✅ No public API, security, memory, serialization, threading,
    performance or migration concerns.
    > 
    > **Context.** `removeAttribute` only acts when the attribute is
    present, and a bound attribute with a `null` value is reported as
    absent, so the binding check in the attribute map was never reached.
    > 
    > - `Element.removeAttribute` now also forwards the call when the
    attribute is absent but has an active binding, so the existing binding
    check throws.
    > - Extended the binding test to cover a signal with a `null` value.
    >
    
    Co-authored-by: totally-not-ai[bot] <290682512+totally-not-ai[bot]@users.noreply.github.com>
    Co-authored-by: Artur Signell <artur@vaadin.com>
    Co-authored-by: Anton Platonov <anton@vaadin.com>
    4 people authored Oct 12, 2026
    Configuration menu
    Copy the full SHA
    2d57ec3 View commit details
    Browse the repository at this point in the history

Commits on Oct 13, 2026

  1. fix(flow-push): use pagehide instead of unload in push client (#26348)…

    … (CP: 25.2) (#26405)
    
    This PR cherry-picks changes from the original PR #26348 to branch 25.2.
    
    ---
    
    > Fixes #26318
    > 
    > **Review risk:** two-way door — build-time patch of the bundled
    Atmosphere client in `flow-push`; affects only when push cleanup runs on
    page leave.
    > 
    > **Behavior deviation** · flow-push · apps using `@Push` in
    Chromium-based browsers
    > 
    > **Background — unload permissions policy.** Chrome and Edge are
    deprecating the `unload` event by defaulting the `unload` permissions
    policy to `()`. On those pages, `unload` listeners never fire and
    registering one logs a policy violation. `pagehide` is the recommended
    replacement.
    > 
    > ## Summary
    > 
    > With `@Push`, the browser console logs "Permissions policy violation:
    unload is not allowed in this document." when the push client loads.
    Atmosphere's unsubscribe-on-leave also never runs for long-polling and
    streaming. The push client now listens to `pagehide`, so there is no
    violation and cleanup runs again. Applications don't need a workaround
    header anymore.
    > 
    > **Risks:**
    > - ⚠️ Behaviour change: Atmosphere now unsubscribes on `pagehide`,
    which also fires when a page enters the back/forward cache. Flow already
    reloads the page on `pageshow`, so a restored page reconnects as before.
    > - ✅ No public API, security, memory, serialization, threading,
    performance or migration impact.
    > 
    > **Context.** The client is `org.atmosphere.client:javascript:4.0.1`,
    inlined into `vaadinPush.js` at build time. Upstream fixed this on
    `master`, but that fix was never released, and the project is no longer
    maintained, so a version bump isn't possible.
    > 
    > ## What changed
    > 
    > - `flow-push` rewrites the `unload` listener registration and removal
    in `atmosphere.js` to `pagehide` while inlining it into `vaadinPush.js`.
    > - The build fails if the inlined client still contains an `"unload"`
    listener, so a changed upstream source can't silently skip the patch.
    > - Added an integration test that serves a push view with the `unload`
    policy blocked and asserts the browser log has no policy violation.
    > 
    > ## Test summary
    > 
    > - New integration test: a long polling push view served with
    `Permissions-Policy: unload=()`; the test fails if the browser log has a
    permissions policy violation after the push client loads. Fails on
    `main`, passes with this change.
    > 
    > - Built `flow-push`. The generated `vaadinPush.js` and
    `vaadinPush-min.js` register `pagehide` and contain no `"unload"`
    listener.
    > - Broke the replacement string on purpose and checked that the build
    fails with the guard message.
    > - Loaded the old and the new client in headless Chromium with
    `Permissions-Policy: unload=()`. The old one logs the violation and the
    new one doesn't, and `vaadinPush.atmosphere` still initializes.
    > 
    >
    
    Co-authored-by: totally-not-ai[bot] <290682512+totally-not-ai[bot]@users.noreply.github.com>
    Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
    Co-authored-by: Anton Platonov <anton@vaadin.com>
    4 people authored Oct 13, 2026
    Configuration menu
    Copy the full SHA
    345b313 View commit details
    Browse the repository at this point in the history

Commits on Oct 14, 2026

  1. chore(deps): bump atmosphere javascript client to 4.0.3 (#26419) (CP:…

    … 25.2) (#26429)
    
    This PR cherry-picks changes from the original PR #26419 to branch 25.2.
    
    ---
    
    > **Review risk:** two-way door — dependency bump of the Atmosphere
    JavaScript client bundled in `flow-push`; affects push connection setup
    and cleanup on page leave.
    > 
    > Follow-up to #26348, Related to #26318
    > 
    > **Behavior deviation** · flow-push · apps using `@Push`
    > 
    > The push client is updated to Atmosphere JavaScript 4.0.3, which
    includes the upstream fix for the `unload` permissions policy violation.
    Because of this, the build-time patch that rewrote the `unload` listener
    to `pagehide` is removed again.
    > 
    > **Risks:**
    > - ⚠️ Behaviour change: 4.0.3 also brings the other upstream changes
    since 4.0.1. On page leave the unsubscribe request goes out with
    `sendBeacon` or `fetch` with `keepalive`. When a page enters the
    back/forward cache, Atmosphere now closes the connection and reopens it
    on `pageshow`, instead of unsubscribing. Flow still reloads the page on
    `pageshow`, so restored pages behave as before.
    > - ✅ No public API, security, memory, serialization, threading,
    performance or migration impact.
    > 
    > **Context.** 4.0.1 registered an `unload` listener, so #26348 patched
    the source while inlining it into `vaadinPush.js`, because no fixed
    release was available then. 4.0.3 registers `pagehide`/`pageshow` by
    default. It only falls back to `beforeunload` or `unload` when an app
    turns off `usePageHide`, and Flow doesn't do that.
    > 
    > - Bumped `atmosphere.client.version` from 4.0.1 to 4.0.3.
    > - Removed the `unload` → `pagehide` replacement and its guard checks
    from the `flow-push` build. `atmosphere.js` is inlined unchanged again.
    > - Kept the integration test from #26348 (push view with
    `Permissions-Policy: unload=()`, no policy violation in the browser log)
    as the regression guard.
    > 
    > ## Test summary
    > 
    > - Built `flow-push`: the generated `vaadinPush.js` /
    `vaadinPush-min.js` are version 4.0.3 and register `pagehide`.
    > - The `unload` permissions policy integration test passes without the
    patch.
    > - Basic push, reconnect, toggle push and push configuration
    integration tests pass for WebSocket, long polling and SSE.
    > 
    > 🤖 Generated with [Claude Code](https://claude.com/claude-code)
    >
    
    Co-authored-by: totally-not-ai[bot] <290682512+totally-not-ai[bot]@users.noreply.github.com>
    Co-authored-by: Artur Signell <artur@vaadin.com>
    3 people authored Oct 14, 2026
    Configuration menu
    Copy the full SHA
    b978b72 View commit details
    Browse the repository at this point in the history
  2. chore: update default Node.js version to v24.22.0 (25.2) (#26434)

    **Review risk:** two-way door — only affects builds where Flow installs
    Node.js itself
    
    ## Summary
    Flow now downloads Node.js v24.22.0 instead of v24.21.0 when no suitable
    Node.js is installed. v24.22.0 is the newest patch release in the
    Node.js 24 LTS line this branch (25.2) already tracks.
    
    ## What changed
    - `FrontendTools.DEFAULT_NODE_VERSION` changed from `v24.21.0` to
    `v24.22.0`
    - `FrontendTools.DEFAULT_NPM_VERSION` changed from `11.19.0` to
    `11.19.1`, the npm version that ships with Node.js v24.22.0
    - Behaviour change: the next build downloads Node.js v24.22.0. A
    user-installed Node.js or a `vaadin.nodeVersion` pin is not affected
    
    ## Test summary
    - Checked against nodejs.org's release index that v24.22.0 is published
    with linux-x64 binaries and ships npm 11.19.1
    - `flow-build-tools` compiles, and the Node.js install tests pass
    
    🤖 Generated with [Claude Code](https://claude.com/claude-code)
    
    Co-authored-by: totally-not-ai[bot] <290682512+totally-not-ai[bot]@users.noreply.github.com>
    Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
    totally-not-ai[bot] and claude authored Oct 14, 2026
    Configuration menu
    Copy the full SHA
    ebc7d91 View commit details
    Browse the repository at this point in the history

Commits on Oct 15, 2026

  1. chore: update license-checker to 3.1.4 on 25.2 (#26444)

    **Review risk:** two-way door — one dependency version on 25.2; the
    update is additive at the API level, but offline key validation behaves
    differently in the cases listed below. Reverting means changing one
    version back.
    
    ## Summary
    Updates the managed `com.vaadin:license-checker` version in the root
    `pom.xml` from 3.1.3 to 3.1.4.
    
    3.1.4 adds support for distribution keys
    (vaadin/license-checker-vaadin10#248,
    vaadin/license-checker-vaadin10#260): a license key that a vendor embeds
    in an application they ship to their own customers, so the application
    can pass a license check when it runs without each installation needing
    its own license file.
    
    ## What changed
    New in license-checker 3.1.4:
    - A new `BuildType.RUNTIME` check for running an application, as opposed
    to building one. It passes with a key that allows runtime or production,
    so an ordinary server key also covers it.
    - A distribution key read from the application's classpath at
    `META-INF/vaadin-distribution-key`. It counts only for runtime checks
    and is never used for development or production builds.
    
    Behavior changes that apply even without a distribution key:
    - Offline key lookup now also checks the application classpath for an
    embedded key. If it finds one, it adds it as the lowest-priority key and
    logs that at info level.
    - A key whose `exp` claim cannot be read, or whose `build_types` claim
    is not a list of known build types, is now rejected as malformed.
    Before, the `build_types` case granted everything. Keys without these
    claims work as before, and a key without `exp` no longer throws a
    `NullPointerException`.
    - An unsigned key, or one signed with an unknown key id, is now reported
    as an invalid license (`LicenseException`). Before, it threw
    `ClassCastException` / `NullPointerException`.
    
    This update only makes the new API available. Nothing on this branch
    calls `BuildType.RUNTIME` yet.
    
    ### Compatibility
    Going from 3.1.3 to 3.1.4 only adds API. Comparing the released jars
    with `javap` shows these new members and no removed or changed ones:
    `BuildType.RUNTIME`, `OfflineKeySource.DISTRIBUTION`,
    `LocalOfflineKey.DISTRIBUTION_KEY_LOCATION`,
    `OfflineKey.isRuntimeAllowed()` and `OfflineKey.isRuntimeOnly()`.
    Nothing on this branch switches over `BuildType`, so the new enum
    constant does not affect existing code. The artifact's own dependencies
    are the same as in 3.1.3.
    
    ## Test summary
    - Compared the public and protected API of the 3.1.3 and 3.1.4 jars with
    `javap`: the changes are additive only.
    - Compared the dependency lists in the published POMs: unchanged.
    - Compiled the Flow modules that call license-checker
    (`flow-plugin-base`, `vaadin-dev-server`, `flow-polymer-template`)
    against the new version on one branch of each line (24.4 with 2.3.4,
    main with 3.1.4), and confirmed that `flow-server` resolves the new
    version.
    
    🤖 Generated with [Claude Code](https://claude.com/claude-code)
    
    Co-authored-by: totally-not-ai[bot] <290682512+totally-not-ai[bot]@users.noreply.github.com>
    Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
    totally-not-ai[bot] and claude authored Oct 15, 2026
    Configuration menu
    Copy the full SHA
    65c88db View commit details
    Browse the repository at this point in the history

Commits on Oct 16, 2026

  1. fix: escape < in initial UIDL script in index.html (#26455) (CP: 25.2) (

    #26473)
    
    This PR cherry-picks changes from the original PR #26455 to branch 25.2.
    
    ---
    
    > Server-rendered content containing "</script>"
    > or "<!--" ended the inline initial UIDL script
    > early and broke the page. Escape "<" as "\x3C"
    > the same way BootstrapHandler does.
    >
    
    Co-authored-by: caalador <mikael.grankvist@vaadin.com>
    vaadin-bot and caalador authored Oct 16, 2026
    Configuration menu
    Copy the full SHA
    d2db77c View commit details
    Browse the repository at this point in the history
  2. fix: track visual modality so dialogs are refreshed on hotswap (#25417)…

    … (CP: 25.2) (#26487)
    
    This PR cherry-picks changes from the original PR #25417 to branch 25.2.
    
    > [!WARNING]
    > The cherry-pick had merge conflicts that were resolved automatically
    by Claude Code. Please review the changes carefully.
    
    ---
    
    > Dialogs are visually modal by default since 25.0, but
    `ModalityMode.VISUAL` was mapped to `setChildModeless`, so a component
    with visual modality was not tracked on the server side at all.
    > 
    > Both code paths that deal with overlays attached to the UI instead of
    the route were guarded by `UI.hasModalComponent()`, so with an open
    dialog they became no-ops:
    > 
    > - `AbstractNavigationStateRenderer` no longer detached the dialog on a
    route refresh, leaving it visible with outdated contents.
    > - `Hotswapper` no longer requested a refresh of the whole route chain.
    > 
    > ## What changed
    > 
    > - Components set to `ModalityMode.VISUAL` are now tracked in
    `UIInternals` without making the UI inert. The meaning of
    `hasModalComponent()` and the inert semantics are unchanged; the new
    `hasModalityComponent()` / `getModalityComponents()` cover both strict
    and visual modality, and both call sites take the new information into
    account.
    > - All components with modality are kept in a **single recency-ordered
    stack** of (component, mode, detach registration) entries, replacing the
    previous strict-only modal component stack. The active modal component
    is the most recent `STRICT` entry, and the inert state of the UI and its
    children is updated in one place by comparing the active modal component
    before and after each modality change. A single stack keeps
    `getModalityComponents()` in the order the components were set and
    cannot report a component twice when it switches between modes.
    > - The detach listener that stops tracking a component is registered
    once per component and kept in its stack entry, so repeatedly changing
    the modality of a component does not pile up listeners on its element;
    it is removed when the component is set modeless.
    > - Calling `setChildModal` again on the already active modal component
    now moves it to the top of the stack (above any visually modal component
    set after it) instead of being a no-op.
    > - `UI.setChildComponentModal(Component, boolean)` still maps `false`
    to `MODELESS`, which now also clears visual modality; its Javadoc points
    visually modal components to the `ModalityMode` overload. `Dialog` needs
    to report its actual mode through that overload for the fix to cover all
    paths (re-attach, `setVisible`, `setModality` on an open dialog,
    add-then-open) — companion change: vaadin/flow-components#10440.
    > - `cleanModalComponents()` in `AbstractNavigationStateRenderer` now
    removes all components with modality on `REFRESH_ROUTE`, instead of
    draining only the strict-modal stack. Modeless components are left
    untouched.
    > 
    > Fixes #25415
    > 
    > ## API Changes
    > 
    > ### com.vaadin.flow.component.internal.UIInternals
    > 
    > ```java
    > // Added
    > public boolean hasModalityComponent() // true if a child component
    with STRICT or VISUAL modality is set; unlike hasModalComponent() it
    also counts components that only show a modality curtain
    > public List<Component> getModalityComponents() // child components
    with STRICT or VISUAL modality, most recently set first, never null
    > public void setChildModal(Component child, ModalityMode mode) //
    overload of setChildModal(Component) that sets any modality mode; VISUAL
    tracks the child as having modality without making the UI and its other
    components inert
    > ```
    > 
    > 
    >
    
    Co-authored-by: totally-not-ai[bot] <290682512+totally-not-ai[bot]@users.noreply.github.com>
    vaadin-bot and totally-not-ai[bot] authored Oct 16, 2026
    Configuration menu
    Copy the full SHA
    633dbf2 View commit details
    Browse the repository at this point in the history
  3. fix: nest overlays inside visually modal components (#26425) (CP: 25.…

    …2) (#26497)
    
    This PR cherry-picks changes from the original PR #26425 to branch 25.2.
    
    ---
    
    > **Review risk:** two-way door — changes where overlays such as
    notifications, context menus and nested dialogs are attached while a
    visually modal dialog is open; easy to revert.
    > 
    > Fixes #25136
    > Builds on #25417 (merged)
    > 
    > **Behavior deviation** · flow-server · apps that open an overlay while
    a default (visually modal) `Dialog` is open
    > 
    > ## Summary
    > 
    > An overlay opened while a visually modal dialog is showing was
    attached to the UI, outside the dialog. The dialog's modality curtain
    blocks pointer events and traps focus outside of it, so the overlay
    could not be clicked or reached with the keyboard. Overlays are now
    nested inside the most recently opened strictly or visually modal
    component.
    > 
    > **Risks:**
    > - ⚠️ Behavior change: an overlay opened while a visually modal
    component is showing now becomes a child of that component, so it is
    removed together with it. This is the same as with strictly modal
    components. Before, it was attached to the UI, where the user could not
    reach it on the client.
    > - ⚠️ Public API: one method added to the internal `UIInternals`.
    > - ✅ No security, memory, serialization, threading, performance or
    migration impact.
    > 
    > ## What changed
    > 
    > - Added `UIInternals.getActiveModalityComponent()`, which returns the
    most recently set component with `STRICT` or `VISUAL` modality, based on
    the modality stack from #25417.
    > - Made `UI.addToModalComponent()` add to that component instead of
    only to the active strictly modal one.
    > - Left `hasModalComponent()`, `getActiveModalComponent()` and all
    inert handling unchanged, so the UI is still only inert for `STRICT`,
    and shortcut routing through the inert-escaping component works as
    before.
    > 
    > ## Test summary
    > 
    > - An overlay opened while only a visually modal component is showing
    is nested into it (this flips the test from #25417 that pinned the old
    attach-to-UI behavior).
    > - An overlay is nested into a visually modal component opened on top
    of a strictly modal one.
    > - A strictly modal component opened on top of a visually modal one is
    nested in it, its overlays are nested in it and stay interactive, and
    nesting goes back to the visually modal component once it is removed.
    > - Existing UI, shortcut and route refresh unit tests pass.
    > 
    >
    
    Co-authored-by: totally-not-ai[bot] <290682512+totally-not-ai[bot]@users.noreply.github.com>
    Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
    3 people authored Oct 16, 2026
    Configuration menu
    Copy the full SHA
    f9b13db View commit details
    Browse the repository at this point in the history

Commits on Oct 19, 2026

  1. chore: Update frontend dependencies (25.2) (#26516)

    ## Automated Frontend Dependency Updates
    
    This PR contains automated updates to frontend dependencies.
    
    ### Update Strategy
    - Target Branch: 25.2
    - Updates: Minor and patch releases (no major)
    
    ### Changes
    The dependencies have been updated using `npm-check-updates`.
    
    ---
    *This PR was automatically created by the update-frontend-dependencies
    workflow.*
    
    ---------
    
    Co-authored-by: github-merge-queue <118344674+github-merge-queue@users.noreply.github.com>
    Co-authored-by: Mikael Grankvist <mgrankvi@vaadin.com>
    3 people authored Oct 19, 2026
    Configuration menu
    Copy the full SHA
    576a40a View commit details
    Browse the repository at this point in the history
  2. fix: give add-on frontend sources in jar-resources their own tsconfig…

    ….json (#25241) (CP: 25.2) (#26518)
    
    This PR cherry-picks changes from the original PR #25241 to branch 25.2.
    
    ---
    
    > ## Problem
    > 
    > Frontend sources of add-ons referenced by `@JsModule` are copied into
    `Frontend/generated/jar-resources`, which the generated project
    `tsconfig.json` **excludes** so that the type checking rules of the
    project are not enforced on third party sources.
    > 
    > Bundlers apply `compilerOptions` only from a configuration that *owns*
    the file, and a file excluded from the project configuration is owned by
    no configuration at all. Add-on sources were therefore transformed with
    no compiler options whatsoever:
    > 
    > - their TypeScript decorators were left as raw syntax that browsers
    cannot parse, so the chunk fails to load with a `SyntaxError`;
    > - they lost `useDefineForClassFields`, so a native class field shadows
    the accessor Lit installs for a reactive property and the element stops
    rendering updates.
    > 
    > Fixes #24982
    > 
    > ## What changed
    > 
    > **A generated configuration for the copied sources.** A new internal
    `TaskGenerateJarResourcesTsConfig` writes a `tsconfig.json` into the
    jar-resources folder from the new `jar-resources-tsconfig.json`
    template. It takes ownership of the folder through a local `include` and
    an empty `exclude`, and `extends` the project configuration — inheriting
    its compiler options while turning the project type checking rules
    (`strict`, `noImplicitAny`, `noImplicitThis`, `noImplicitReturns`,
    `noFallthroughCasesInSwitch`) off for add-on sources. The `extends` path
    is resolved relative to the jar-resources folder, since that folder is
    not at a fixed depth below the project folder.
    > 
    > **Ordering, so the file survives the build.** Both copying add-on
    resources and cleaning npm files empty the folder the add-on sources are
    copied to, and both run after the project `tsconfig.json` is generated.
    Generating this configuration alongside the project one lost it again in
    the same run, right before the bundle was built. It is now a task of its
    own, registered in `NodeTasks#commandOrder` after the tasks that empty
    the folder.
    > 
    > **Lifecycle.** The configuration cannot outlive the project
    `tsconfig.json` it extends — a configuration whose `extends` does not
    resolve makes bundlers and the TypeScript compiler fail outright:
    > 
    > - `TaskCopyFrontendFiles` removes files in the target folder that it
    did not copy itself, so the generated configuration is now kept out of
    that cleanup;
    > - `TaskCleanFrontendFiles` removes it whenever it removes the project
    `tsconfig.json`. The file is resolved from the options by the generating
    task, which also covers the production build, whose cleanup task is
    created from options that only know the frontend directory. The frontend
    directory is resolved the same way the rest of the build resolves it, so
    projects still keeping their frontend files in the legacy location are
    handled too.
    > 
    > **Configurations shipped by add-ons are discarded.** A bundler
    resolves the configuration of a file from the closest folder that has
    one, so a `tsconfig.json` copied in along with add-on sources takes
    those sources back from the generated one — whether it lands next to it
    or in a sub folder — and its `extends` points at the project it came
    from, which no longer resolves.
    `TaskGenerateJarResourcesTsConfig#discardCopiedTsConfigs` drops them at
    any depth while copying, and logs a warning naming the discarded files.
    This is applied both in `TaskCopyFrontendFiles` and in
    `TaskCopyLocalFrontendFiles`, since local frontend resources are copied
    into the same folder. Previously this only went unnoticed because the
    generating task usually ran afterwards and overwrote the file.
    > 
    > ## Testing
    > 
    > - `JarResourcesTsConfigTest` covers the generated configuration, the
    discarding of copied ones, and the clean up.
    > - A new add-on TypeScript source using experimental decorators
    (`DecoratorElement.ts` in `test-express-build/frontend-add-on`) is
    copied into the jar-resources folder like the sources of a real add-on.
    `AddonDecoratorIT` asserts against the bundle the build actually
    produces that its decorators are transpiled to the `__decorate` helper
    and that the initializer of the reactive property ends up in the
    constructor. The unit tests around the configuration would have passed
    just as well on the broken setup, since the regression is in what the
    bundler does with those sources rather than in the configuration itself.
    The browser behaviour is already covered by the add-on component of
    `TemplateIT`, so this test pins down the transformation, letting a
    regression name its cause instead of surfacing as an application that
    does not start.
    > 
    > The assertions hold with the currently pinned Vite, and all of them
    fail on a Vite version whose bundler scopes compiler options to the
    files a configuration owns, so the test gets its teeth with that bump.
    
    ---------
    
    Co-authored-by: totally-not-ai[bot] <290682512+totally-not-ai[bot]@users.noreply.github.com>
    Co-authored-by: Artur Signell <artur@vaadin.com>
    Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
    4 people authored Oct 19, 2026
    Configuration menu
    Copy the full SHA
    0faeecc View commit details
    Browse the repository at this point in the history

Commits on Oct 20, 2026

  1. fix: reject misplaced dots in EmailValidator local part (#26533) (CP:…

    … 25.2) (#26536)
    
    This PR cherry-picks changes from the original PR #26533 to branch 25.2.
    
    ---
    
    > **Review risk:** two-way door — only affects which strings
    `EmailValidator` (and anything reusing `EmailValidator.PATTERN`)
    accepts; reverting restores the old pattern.
    > 
    > ## Summary
    > `EmailValidator` accepted local parts with leading, trailing or
    consecutive dots (`.foo@example.com`, `foo.@example.com`,
    `foo..bar@example.com`), which `vaadin-email-field` rejects since
    vaadin/web-components#7179. The local part of the pattern now uses the
    same dot-separated form as the web component. The domain part is
    unchanged, so `foo@example..com` is still rejected (see #17353).
    > 
    > ## What changed
    > - `EmailValidator.PATTERN` local part changed from
    `([a-zA-Z0-9_\.\-+])+` to `[a-zA-Z0-9_\-+]+(?:\.[a-zA-Z0-9_\-+]+)*`.
    > - Behavioral change: addresses with misplaced dots in the local part
    now fail validation. Code that reuses the public `PATTERN` constant
    becomes stricter in the same way.
    > 
    > ## Test summary
    > - Added a test that leading, trailing and consecutive dots in the
    local part are rejected (it fails without the fix).
    > - Existing email validator tests, including the
    consecutive-dots-in-domain case, still pass.
    > 
    > Fixes #18887
    > 
    > 🤖 Generated with [Claude Code](https://claude.com/claude-code)
    >
    
    Co-authored-by: totally-not-ai[bot] <290682512+totally-not-ai[bot]@users.noreply.github.com>
    Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
    3 people authored Oct 20, 2026
    Configuration menu
    Copy the full SHA
    74a1e70 View commit details
    Browse the repository at this point in the history
Loading