Declare @types/react and @types/lodash as optional peers of slate-react

[unrevised6419]

Problem

slate-react's shipped type declarations reference typings packages that package.json never declares:

• react — 33 references across dist/**/*.d.ts, plus nine /// <reference types="react" /> directives (use-slate-static.d.ts, use-focused.d.ts, use-element.d.ts, …)
• lodash — dist/hooks/android-input-manager/android-input-manager.d.ts imports DebouncedFunc in a type position

@types/react and @types/lodash are present only in devDependencies, so they are not part of the published contract.

Under a hoisted node_modules layout this resolves by accident, which is why it survives normal testing (and skipLibCheck: true hides the resulting errors inside the .d.ts). It does not resolve under pnpm's isolated linker with a global virtual store: each package is linked from a store path outside the consuming project, and TypeScript resolves a package's imports from where that package physically lives. react then resolves to the untyped react/index.js, the editor component props degrade to any, and consumers building with noImplicitAny get errors like:

error TS7016: Could not find a declaration file for module 'react'.
'.../react/index.js' implicitly has an 'any' type.

Change

@types/react and @types/lodash are added as optional peer dependencies of slate-react. Optional matters: JavaScript-only consumers must not get an unmet-peer warning (verified with strictPeerDependencies: true and no @types/* installed).

The ranges track majors and are deliberately open-ended at the top: @types/react majors follow React majors, and the meaningful constraint is already carried by the existing react peer, so pinning an upper bound would only produce false unmet-peer failures the day @types/react 20 ships. @types/lodash tracks lodash 4.x, matching the lodash dependency range. The shipped declarations were typechecked against @types/react 17, 18 and 19 and are clean in all three.

"peerDependencies": {
  "@types/lodash": "^4.14.0",
  "@types/react": ">=18.0.0",
  "react": ">=18.2.0",
  "react-dom": ">=18.2.0",
  "slate": ">=0.121.0",
  "slate-dom": ">=0.119.1"
},
"peerDependenciesMeta": {
  "@types/lodash": { "optional": true },
  "@types/react": { "optional": true }
}

The existing react, react-dom, slate and slate-dom ranges are unchanged. A patch changeset is included.

No @types/react-dom entry: react-dom is imported only for values (unstable_batchedUpdates, flushSync) and does not appear in any type position in the shipped declarations.

@types/lodash is only reachable through a deep import today (slate-react/dist/hooks/android-input-manager/android-input-manager), since that module is not re-exported from index.d.ts. It is still a specifier appearing in a type position in a shipped .d.ts, and the package has no exports map, so deep imports are possible. An alternative worth considering separately is dropping the DebouncedFunc reference from that declaration in favour of a structural type, which would remove the requirement entirely.

Sibling packages

The other packages in the monorepo were swept with the same rule — every specifier appearing in a type position in a shipped .d.ts must be declared — and all are already correct:

Package	External specifiers in dist/**/*.d.ts	Status
slate	none	OK
slate-dom	slate	already a peer
slate-history	slate	already a peer
slate-hyperscript	slate	already a peer (no React or JSX types)
slate-react	react, lodash, slate, slate-dom	slate/slate-dom already peers; the two @types/* entries are this PR

Verification

All five packages were built and packed, and both a patched and an unpatched baseline slate-react tarball were installed into scratch consumers.

1. pnpm 11.26 isolated linker with enableGlobalVirtualStore: true and a virtualStoreDir outside the consumer, typechecking with skipLibCheck: false and noImplicitAny:
   • unpatched: TS7016 for react in editable.d.ts, element.d.ts, leaf.d.ts, text.d.ts and others
   • patched: clean
2. The assertion is written as a negative, because a positive one keeps passing when types widen to any:

   // @ts-expect-error — collapses to `any` if @types/react is unreachable
   <Editable nonexistentProp={1} />

   It still errors as expected with the patch applied, so the props are real types rather than any.
3. tsc --traceResolution from slate-react/dist/components/editable.d.ts: Module name 'react' was successfully resolved to '.../@types/react/index.d.ts' — a .d.ts, not react/index.js.
4. npm with --install-strategy=nested typechecks cleanly both before and after, confirming that a hoisted layout forgives the missing declaration.
5. attw --pack reports "No problems found" for all five tarballs. publint reports no errors; the warnings it does emit are pre-existing and identical across all five packages (repository uses the git:// shorthand, dist/index.es.js is ESM interpreted as CJS, no exports map) and are untouched here.

Note on side findings

Two unrelated packaging observations surfaced during the sweep. Neither is addressed in this PR, since both are dependency removals with their own compatibility risk for anyone relying on the transitive install, and they are worth deciding on separately:

• slate-dom declares five dependencies that its bundle never requires: @juggle/resize-observer, direction, lodash, scroll-into-view-if-needed and is-plain-object. The built bundle only requires is-hotkey and slate.
• slate-react declares is-hotkey and tiny-invariant, which its bundle likewise never requires.

The publint warnings listed above are a third such finding: they affect every package equally and predate this change.

🤖 Generated with Claude Code

[changeset-bot]

🦋 Changeset detected

Latest commit: 9b1f0ae

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
slate-react	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR