What

Adds async support (AsyncMockFirestore + async reference classes) to this fork, ported from upstream mdowds/python-mock-firestore#62 (credit: @anna-hope), which was never merged upstream.

Ported (additive only):

• 5 new modules: async_client.py, async_collection.py, async_document.py, async_query.py, async_transaction.py
• consume_async_iterable helper added to _helpers.py
• Async exports added to __init__.py (AsyncMockFirestore, AsyncDocumentReference, AsyncCollectionReference, AsyncQuery, AsyncTransaction)
• 5 async test files (203 → 207 tests; aiounittest added to requirements-dev-minimal.txt)

What was deliberately NOT taken from PR mdowds#62

PR mdowds#62 also refactored the sync classes (client.py, collection.py, query.py, transaction.py, _helpers.py) and black-reformatted the whole package. Both were skipped: this fork has diverged on exactly those files with semantic fixes (DELETE_FIELD-doesn't-fail, don't-drop-empty-dicts, set(merge=True), collection.get generator fix, query iteration) that a wholesale merge would clobber.

Adaptations needed to fit this fork's sync API

The async layer was written against PR mdowds#62's refactored sync classes, so four surgical adaptations were required:

1. async_query.py called Query._process_field_filters / _process_pagination, which only exist in PR Add async support and format with black mdowds/python-mock-firestore#62's refactor. Instead, the filtering/ordering/cursor/limit body of this fork's Query.stream was extracted into Query._process_snapshots (pure extract-method; sync behavior unchanged) and the async query reuses it.
2. async_collection.py::get inherited list(self.stream()), which breaks when stream is an async generator — now await consume_async_iterable(self.stream()). Its where() also gained this fork's filter=FieldFilter keyword support to match the sync signature.
3. async_document.py::set re-implemented PR Add async support and format with black mdowds/python-mock-firestore#62-era merge logic, which would bypass this fork's set(merge=True) fix and break with unawaited coroutines (the sync set calls self.update, which dispatches to the async override). It now mirrors this fork's merge branch (flatten_for_merge + __name__ stamping) with proper awaits.
4. client.py::_ensure_path used exact type() in (...) checks, which misroute AsyncMockFirestore/AsyncDocumentReference when resolving multi-segment paths. Changed to isinstance — behavior-identical for the sync classes (CollectionReference is not a subclass of either type).

AsyncMockFirestore shares the sync in-memory store

Yes — verified empirically. AsyncMockFirestore subclasses MockFirestore; all reads/writes (sync and async reference classes alike) operate on the same plain-dict _data store. Consequences for consumers:

• A sync MockFirestore view pointed at the same store lets test fixtures keep seeding synchronously while production code under test reads via await:

  afs = AsyncMockFirestore()
  sync_view = MockFirestore()
  sync_view._data = afs._data
  sync_view.collection('users').document('u1').set({'name': 'Ada'})   # sync seed
  await afs.collection('users').document('u1').get()                   # sees it

• Async writes are equally visible to sync reads (verified both directions). No shared-store shim is needed for the future editor/scraper async migration.

Test results

Sync test files are untouched, and the sync suite result is identical to master — the port is additive.

🤖 Generated with Claude Code