GridTester has no way to deselect a row, and no Grid editor support

[totally-not-ai]

Problem. GridTester has select(int), selectAll() and getSelected(),
but no way to undo a selection. In the browser a user deselects by
ctrl-clicking a selected row in multi-select mode, by clicking the row's
checkbox again, by unchecking the select-all header checkbox, or by clicking
the selected row in single-select mode.

var grid = new GridTester<>(component);
grid.select(0);
// no deselect(0), no deselectAll()

The Kotlin layer already has the pieces — Grid._select calls
deselectAll() internally for single-select before
selectionModel.selectFromClient(item).

While in there, two related gaps:

• Grid editor. Grid.getEditor() — the user double-clicks a row, edits the
  buffered fields and saves or cancels. There is no tester support for any of
  it. GridProTester.setValue covers GridPro's inline editing, but not the
  plain Grid editor.
• selectAll() doesn't consider
  setSelectAllCheckboxVisibility(HIDDEN), even though the Kotlin
  Grid._selectAll Javadoc claims it "fails if ... the select all checkbox is
  hidden". If the checkbox isn't rendered there is no way for the user to
  trigger it. (Worth deciding deliberately — this is the feat: give every value tester a uniform clear() #168 question in
  reverse.)

Ask. Add deselect(int row) and deselectAll(), routed through the
selection model's from-client path like select(int) is. Decide on the
select-all-checkbox gate. Grid editor support can be split off if it's too big
for one PR.

Found while auditing the project against #153 / #168.

[totally-not-ai]

Valid on all three points, but they are three different sizes of change. I implemented the first one and left the other two for a human decision — details and evidence below.

1. deselect(int) / deselectAll() — real gap, straightforward, implemented

The plumbing already exists in Flow and just isn't exposed. GridSelectionModel declares deselectFromClient(T) right next to the selectFromClient(T) that GridKt._select uses, and AbstractGridMultiSelectionModel has a clientDeselectAll() that mirrors the clientSelectAll() that _selectAll reaches by reflection. So this is purely a missing tester method, not missing behaviour.

Implemented as GridTester.deselect(int row) and GridTester.deselectAll(), both routed through the from-client path so app-side selection listeners see isFromClient() == true, exactly like select(int).

Two deliberate choices worth reviewing:

• deselect(row) delegates straight to deselectFromClient(item), which ignores the call when the row is not selected, when the item is not selectable (Grid.setItemSelectableProvider), or when the grid is single-select with setDeselectAllowed(false). In the browser the user's click is ignored in those cases too, and select(row) behaves the same way today, so no extra gate was added.
• deselectAll() deselects the current selection item by item rather than calling clientDeselectAll(). That keeps it working in single-select mode as well (where there is no select-all checkbox at all), and it avoids inheriting the hidden-checkbox no-op described in point 2. It does throw IllegalStateException for SelectionMode.NONE, where there is no selection for the user to clear.

2. selectAll() vs SelectAllCheckboxVisibility.HIDDEN — real bug, but the fix direction is a policy call

Confirmed, and it is worse than the issue suggests: today it is a silent no-op, not a failure. AbstractGridMultiSelectionModel.clientSelectAll() opens with

// ignore call if the checkbox is hidden
if (!isSelectAllCheckboxVisible()) {
    return;
}

and _selectAll invokes exactly that method by reflection. Probe on a 2-row multi-select grid with setSelectAllCheckboxVisibility(HIDDEN): selectAll() returns normally and getSelected().size() is 0. So the Grid._selectAll Javadoc claim that it "fails if ... the select all checkbox is hidden" is simply not true — the test passes and selects nothing. clientDeselectAll() carries the same guard.

Options, for someone to pick:

• A — throw IllegalStateException when the checkbox is hidden (my recommendation). A silent no-op cannot be what any test wants, and it matches the framework's "fail if the user could not have done this" stance. feat: give every value tester a uniform clear() #168's reasoning does not transfer: a hidden clear button still leaves the user a way to empty a field (select all, press Delete), whereas a hidden select-all checkbox leaves no way at all to select every row in one gesture. Behavioural change, so it needs to be called out in release notes.
• B — keep the no-op and fix the Javadoc to say it silently does nothing. Zero risk, but keeps a footgun that makes a test look green while asserting nothing.
• C — make selectAll() select every item through per-item selectFromClient, so it works regardless of checkbox visibility. Makes the method always succeed, but it stops modelling the select-all gesture and starts modelling "the user checked every row one by one", which is a different user action with a different event stream.

I did not touch selectAll() in the PR, so whichever way this goes it stays a separate, clearly-labelled change.

3. Grid editor support — agreed, should be split off

There is genuinely nothing today, and the one existing piece is dead: Editor<T>._editItem in Grid.kt has no caller anywhere in the repo, so it is not reachable from any Java tester. Nothing exists for save, cancel, editor-open state, or the editor's bound fields.

Worth noting for whoever picks it up: plain Grid does not open the editor on double-click by itself — the application wires addItemDoubleClickListener to editor.editItem(item). So doubleClickRow(0) already opens the editor in apps that do that; what is missing is everything after the editor is open. A plausible shape is editRow(int), isEditorOpen(), getEditorComponent(int|String column), saveEditor(), cancelEditor(), with the buffered/unbuffered distinction and Editor.getBinder() validation errors as the parts that need real thought. That is its own issue.

What the PR contains

• GridTester.deselect(int) and GridTester.deselectAll(), plus Grid._deselect / Grid._deselectAll in the Kotlin layer.
• Tests in BasicGridTesterTest: single-select deselect (including the unselected-row no-op and deselectAll() on a single selection), from-client selection event assertion, setDeselectAllowed(false) keeps the selection, multi-select deselect removes only the targeted row followed by deselectAll(), deselectAll() with the select-all checkbox HIDDEN, SelectionMode.NONE throws, and disabled grid throws.
• One test in LocatorApiTest proving both methods surface on the generated GridLocator — LocatorProcessor only walks the tester chain, so a method that failed to surface there would be invisible from the locator API.

API delta against main:

// com.vaadin.flow.component.grid.GridTester
// Added
public void deselect(int row)
public void deselectAll()

Full build is green (mvn verify, all 8 modules).

[mcollovati]

Deselect issue fixed by #187

[mcollovati]

@totally-not-ai create a separated issue for Grid editor support. Also make a PR to fix selectAll() with option A

[totally-not-ai]

@mcollovati Grid editor support is now #262 — it carries the API sketch plus the points that need deciding first (per-item setEditorComponent(SerializableFunction), buffered vs unbuffered save, and how a Binder-refused save should surface).

On selectAll(): option A is already on main. It went in with #187, in commit e5ae43e, so there is nothing left to open a PR for. What landed:

• _selectAll and _deselectAll share one guard, so selectAll() now throws IllegalStateException instead of silently selecting nothing whenever isSelectAllCheckboxVisible() is false — SelectAllCheckboxVisibility.HIDDEN, an item-selectable provider (which forces it to false), a non-in-memory data provider, and VISIBLE with undefined size.
• BasicGridTesterTest.basicGrid_hiddenSelectAllCheckbox_selectAllAndDeselectAllThrow pins both directions and asserts the selection is left untouched. I re-ran BasicGridTesterTest against current main: 33/33 green.

If what you had in mind was something beyond that guard — say applying the same rule to other testers, or a release note for the behavioural change, since selectAll() used to pass quietly — say the word and I will open a PR for it.

[mcollovati]

Closing: grid editor support has its own ticket and GridTester issues have been fixed.