Suggestion menus: whitespace-boundary anchoring and no blank renders

[janglad]

What changed

• resolveSuggestionMenuTarget: with boundary: "whitespace", the match now anchors on the last trigger that has whitespace (or the start of the lookbehind) before it, instead of rejecting when the last trigger character does not. A later trigger character is part of the query, so @ada@example resolves as one query rather than closing the menu at the second @. boundary: "any" is unchanged.
• useSuggestionMenu:
  • A synchronous getItems result is applied in the same state update as the new target, with no intermediate loading render.
  • For an async getItems, a query refined on the same trigger keeps the previous items while status is loading. Stale responses are still dropped.
  • A getItems that throws synchronously now resolves to the error state instead of escaping the refresh.
• @input/pen-types: new isPromiseLike guard, used by the hook and by core's extension lifecycle in place of two inline thenable checks.

Why

A host that renders the menu only while it has items saw it close and reopen on every keystroke, and a whitespace-bounded trigger could not be used for queries that contain the trigger character, such as email addresses.

Breaking

Yes, for hosts using boundary: "whitespace" that relied on a second trigger character closing the menu (for example :smile:). Set closingChar to keep that behaviour.

Specs and changesets

• spec/rules/selection.md (N6) and spec/packages/rendering/react.md amended.
• Two changesets: @input/pen-core minor (breaking), and @input/pen-react / @input/pen-core / @input/pen-types patch.

Testing

• New cases in resolveSuggestionMenuTarget.n6.test.ts for the anchoring and the no-valid-trigger case.
• New cases in suggestionMenu.staleResults.test.tsx for the synchronous path, retained items during loading, and the synchronous throw.

---

Note

Medium Risk
Breaking change to whitespace-boundary trigger resolution affects mention/emoji hosts; menu refresh logic changes observable loading/error behavior in React suggestion UIs.

Overview
Suggestion menus get steadier UX and more predictable trigger matching.

resolveSuggestionMenuTarget with boundary: "whitespace" now anchors on the last @ (or trigger) that has whitespace or lookbehind start before it, so text like @ada@example is one query instead of reopening at the second trigger. Hosts that depended on a second trigger character to close the menu (e.g. :smile:) should set closingChar. boundary: "any" is unchanged.

useSuggestionMenu avoids blanking the list on each keystroke: synchronous getItems applies items in the same update (no flash of empty loading); async refreshes on the same trigger keep prior items while loading; synchronous throws surface as error. isPromiseLike is exported from @input/pen-types and shared by the hook and extension activate/deactivate lifecycle.

^{Reviewed by Cursor Bugbot for commit cc0b350. Bugbot is set up for automated code reviews on this repo. Configure here.}

[janglad]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit cc0b350. Configure here.}