v5: remove deprecated and dead exports

首先阅读 src/index.ts、src/useObservable.ts 和 src/auth.tsx 中的 v5 调用点，并在处理被阻塞的 Group 1 之前检查 #740。对于 Group 3，检查 test/auth.test.tsx 及其对 AuthCheckWrapper 的使用；完成工作还需要执行 npm run docs:fork，并在 v5 upgrade-guide 中添加相应替换条目。

Background

reactfire ships several exports that are deprecated or dead but kept because removing them breaks consumers at load time, not just at compile time. v5 is the release that can drop them. Scope widened 2026-08-03 from the two checkOptions helpers to every deprecated export, so the breaking-removals work is tracked in one place.

Corrected 2026-08-05. An earlier revision of this issue described the work as a checklist of independent deletions. Call sites were checked against v5 at e7b18c2 and that is not accurate: one group is blocked on another PR, one is not dead code at all, and one requires test changes. The three groups below are sorted by what they actually cost, and they are not independent. Line references are as of e7b18c2.

---

Group 1: checkOptions and checkinitialData (BLOCKED, do not start)

• Remove checkOptions from src/index.ts (:34)
• Remove checkinitialData from src/index.ts (:43)

⚠️ These are not dead on v5 today. checkIdField (src/index.ts:47) still calls checkOptions, and so does checkinitialData. The earlier claim that they became unused "as of #740" is true of #740's branch, which rewrites checkIdField to read options?.idField directly, but #740 targets main and has not merged, so v5 has not received that rewrite.

Doing this group now would mean duplicating #740's change on v5, which then conflicts when main is forward-integrated. Blocked until #740 merges and is forward-integrated. checkIdField itself stays either way; it is still used by the Firestore and Database data hooks.

Group 2: startWithValue (NOT a deletion, it is a behavior change)

• Remove startWithValue from ReactFireOptions (src/index.ts:30), marked @deprecated use initialData instead

⚠️ This is not dead code. It has three live call sites on v5:

• src/useObservable.ts (:113): config?.initialData ?? config?.startWithValue, the actual fallback that makes the option work
• src/useObservable.ts (:78): the hasInitialData check that decides whether to skip loading
• src/auth.tsx (:34): useUser will not seed initialData from auth.currentUser if the caller passed startWithValue

So removing it changes runtime behavior for anyone still passing it, rather than only failing their typecheck. It needs the fallback logic removed alongside the field, and an upgrade-guide entry pointing at initialData. Independent of Group 1.

Group 3: ClaimsCheck and AuthCheck (self-contained, but touches tests)

• Remove ClaimsCheck (src/auth.tsx:215) and its exported ClaimsCheckProps (:60)
• Remove AuthCheck (src/auth.tsx:259) and its exported AuthCheckProps (:54)

Both are @deprecated Use useSigninCheck instead, both only function in experimental Suspense mode and console.warn otherwise. They must go together: AuthCheck renders ClaimsCheck internally (src/auth.tsx:271).

⚠️ test/auth.test.tsx needs more than deletion. It defines an AuthCheckWrapper and reuses it at :285-300 and :369 inside a useUser test that is not about AuthCheck at all. That test needs rewriting to use a plain provider, not removing.

Do not also remove ClaimCheckErrors (:67), despite it sitting between the two interfaces. It is part of the SigninCheckResult shape (:81, :96) and stays.

Then, once the groups above land

• Regenerate reference docs (npm run docs:fork, not npm run docs) to drop the corresponding pages
• Record each removal in the v5 upgrade guide with its replacement (initialData for startWithValue, useSigninCheck for both components)

Notes

• Each of these is a runtime break for plain JS importers (an ESM import error at load), not only a type error. That is why they were deferred rather than done in a patch.
• After #740's tightening, checkinitialData's inferred return type becomes unknown (was any). Harmless while unused, and another reason to retire it.
• Deliberately not folded into #740. That PR tightens ReactFireOptions generics, its squash body becomes the changelog, and a removal buried under a fix: title is how a break gets missed. Same release, separate PR.
• Groups 2 and 3 can proceed while Group 1 is blocked. Splitting this into more than one PR is reasonable; a single PR mixing a behavior change with two component removals makes the changelog harder to read.

Context: #740, #741.