[cloud_firestore][Windows] Invalid field names (empty or `__x__`) close the app instead of throwing a FirebaseException

Aperta
#18,928 1 commento 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

@SelaseKay ci sta già lavorando.

Dal 12/11/2026.

  • #18953 di @SelaseKay — aperta

Valutazione

Difficoltà
4/5
Tempo stimato
3-5 giorni
Idoneità per principianti
52/100
Tipo di issue
Bug
Chiarezza
Specificata chiaramente
Stato di attività
Attiva
Stack tecnologico
c, dart, flutter
Ambito
api, backend

Direzione di ricerca

Start in packages/cloud_firestore/cloud_firestore/windows/cloud_firestore_plugin.cpp and compare DocumentReferenceSet, DocumentReferenceUpdate, the transaction handlers, and WriteBatchCommit, which already catches native exceptions. Trace how Pigeon handlers return errors to Dart, then reproduce the invalid field write on Windows and check that set, update, and transaction writes report a catchable FirebaseException instead of terminating the process.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

Needs Attention platform: windows plugin: cloud_firestore reproduced type: bug type: crash
Is there an existing issue for this?
  • I have searched the existing issues.
Which plugins are affected?

cloud_firestore

Which platforms are affected?

Windows

Description

On Windows, writing a document whose data contains an empty field name ("") or a reserved one (__x__) closes the app. Dart never sees an error: try/catch, FlutterError.onError and runZonedGuarded don't run, and Dart-level crash reporting records nothing. On Android the same write fails with a normal FirebaseException.

What happens

The Firebase C++ SDK validates field names and throws std::invalid_argument for an empty segment or one that starts and ends with __ (user_data.cc L185-L197). That rejection is correct. The problem is that the Windows plugin calls Set and Update without catching it:

The exception leaves the Pigeon handler on the platform thread, and the process terminates (0xe06d7363, uncaught C++ exception).

WriteBatchCommit already does the right thing: it catches std::exception and returns a FlutterError (L1776-L1779). So the same invalid data fails cleanly in a batch but crashes the app in set(), update() and transactions. Android wraps documentReferenceSet in try/catch and sends the error to Dart (FlutterFirebaseFirestorePlugin.java L645-L675).

Evidence from production

Our Windows desktop app writes query results as maps keyed by column name. A SQL Server query SELECT 1 returns a column with an empty name, so the app wrote {'': 1}. Several customers' apps then closed about 20 seconds after every start, because the pending write retried on each launch. We had no error report from any of them; we only found the cause through Windows minidumps a customer sent us:

  • Exception 0xe06d7363, thrown type std::invalid_argument (catchable types invalid_argument, logic_error, exception), on the platform thread.
  • Throw site: the C++ SDK's UserDataConverter field-name check, the "Document fields must not be empty" branch.
  • Call chain: platform channel → cloud_firestore Pigeon handler → DocumentReference::Set → ParseData → field-name validation.
  • Seen on Windows 11 (build 26100) and Windows Server 2019 (build 17763).

We observed the empty-name case in production. The __x__ case goes through the same check, so we expect the same crash, but we haven't observed it.

The broader problem

The Windows plugin has no general rule that native failures become Dart errors. Each handler decides for itself, and several don't catch at all. Other examples:

  • ParseQuery catches exceptions but only prints to stderr and returns a default Query (L1607-L1611), so the real error message never reaches Dart.
  • #18883: transactions on Windows can use a destroyed Transaction object. That is a different root cause (object lifetime, not a thrown exception), but users see the same thing: the app closes with no Dart error.

For a desktop app, a native exit with no Dart error is the hardest failure to fix remotely. The developer gets no stack trace or message, and must ask the user for crash dumps. An invalid write should fail like it does on Android, with an exception the app can catch and report.

Reproducing the issue
// Windows desktop
try {
  await FirebaseFirestore.instance.collection('test').doc('a').set({'': 1});
  // Also expected to crash: {'__x__': 1}, {'row': {'': 1}}, update(),
  // and transaction.set() with the same data.
} catch (e) {
  print('caught: $e'); // never reached on Windows: the process exits
}

Expected (as on Android): a FirebaseException with code invalid-argument and the SDK's message, for example "Invalid data. Document fields must not be empty".

Actual on Windows: the process exits with 0xe06d7363.

We reconstructed this from the minidumps and the source above. We haven't run this exact snippet in an isolated Windows project.

Firebase Core version

firebase_core 3.15.2 (Firebase C++ SDK 12.7.0) in production, with cloud_firestore 5.6.12. The affected handlers are unchanged on current main (13420be, cloud_firestore 7.1.0, C++ SDK 14.1.0).

Flutter Version

3.35.3 stable (Windows desktop)

Additional context and comments

A possible fix:

  • Wrap every Pigeon handler and the transaction update function in try/catch (const std::exception&), like WriteBatchCommit. Return a FlutterError with code invalid-argument and e.what() as the message.
  • In ParseQuery, report the caught error to Dart instead of printing it and returning a default Query.
  • Optionally, validate field names in Dart before calling the platform, so every platform fails the same way.

Happy to open a PR for the first point if that approach works for you.

Lingua principale
Dart
Stelle
9.3k
Fork
4.1k
Merge medio
2g 10h
PR unite (30g)
78

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di firebase/flutterfire

Tutte le issue di firebase/flutterfire

Issue simili

Altre issue su Dart

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.