[joker-777]

Fixes the JSI (C++) path of SQLiteAdapter (jsi: true), which the Kenhub companion now uses on Android, so setup failures reject the way the legacy bridges do after #5 instead of killing the process. Related upstream issue: Nozbe#1945.

• Schema setup, migration and unsafeResetDatabase() failures now reject the promise and reach onSetUpError, instead of std::abort()
• After a failed setup or reset the connection stays open, so the app can recover with database.unsafeResetDatabase(). Other calls on it reject until a reset succeeds, instead of failing an assertion
• A failed database.unsafeResetDatabase() restores the adapter and clears record caches, so the reset can be retried through the public API (this applies to every adapter)
• Every adapter clears its cached record IDs at the start of a reset, so a reset that fails early does not leave the adapter and JS record caches out of sync
• A database file that cannot be opened is reported through the rejected setup promise, instead of a thrown pointer that no handler catches terminating the process
• Calling a native method with the wrong number of arguments throws a JS error on Android too
• Raw SQL the bridge cannot run (empty or comment-only, or a record query not selecting id first) is rejected instead of failing a native assertion
• The prepared statement cache is bounded to 256 statements with least-recently-used eviction. Statements held by a caller are pinned and never evicted mid-use

Left as is: the std::abort() in configureJNI (DatabasePlatformAndroid.cpp) when GetJavaVM fails. It runs while the JSI binding is installed, before any adapter or promise exists, and a failing GetJavaVM means the JNI environment itself is broken.

Verification:

• Android CI builds the C++ and runs the shared adapter suite on an emulator in async and JSI mode. The shared tests now also check recovery after a failed migration and after a failed reset on JSI, and they pass there.
• iOS CI fails before building: xcodebuild: Unable to find a device matching the provided destination specifier (iPhone 16 / iOS 18.1). This happens on every CI run in the fork, so the cause is the workflow's simulator destination on the current macOS runner, not this change. The C++ was therefore not built for iOS.
• Local: the C++ was not compiled with the NDK or Xcode here. It was syntax-checked with clang++ -fsyntax-only against the JSI and simdjson headers, and a standalone copy of the LRU/pinning logic ran against sqlite3 under AddressSanitizer. yarn eslint, yarn flow, tslint, and yarn jest all pass. Jest ran with the sqlite-node suite (src/adapters/sqlite/test.js) excluded, because better-sqlite3 does not build on the host Node, so CI's JavaScript jobs cover that suite.

🤖 Generated with Claude Code

[Copilot]

🟡 Changes recommended

Database recovery reads an uninitialized flag and leaves successful resets uninitialized, while failed SQLite opens can leak handles.

2 open issues

• Recovery reset asserts on uninitialized state and leaves DB uninitialized · New
• Close SQLite handle before throwing on open failure · New

What changed in this PR

This PR brings JSI SQLite error handling closer to the legacy bridges by reporting native setup and reset failures to JavaScript rather than terminating the app.

Changes:

• Routes open, setup, reset, and argument-count errors through JavaScript error handling.
• Adds least-recently-used statement caching with pinning to protect statements in use.
• Adds mocked dispatcher regression tests and documents the behavior changes.

File	Description
src/​adapters/​sqlite/​makeDispatcher/​test.js	Tests error propagation and reset retries using mocked native adapters.
src/​adapters/​sqlite/​makeDispatcher/​index.native.js	Captures adapter-open errors and rejects subsequent calls.
native/​shared/​Sqlite.h	Defines movable statement guards with cache pinning.
native/​shared/​Sqlite.cpp	Throws open errors by value and implements statement pin tracking.
native/​shared/​JSIHelpers.h	Routes argument-count errors through the platform error wrapper.
native/​shared/​DatabaseBridge.cpp	Replaces setup and reset aborts with JavaScript errors.
native/​shared/​Database.h	Defines the statement cache structure and eviction interface.
native/​shared/​Database.cpp	Updates cache cleanup and clears reset mode after vacuum failure.
native/​shared/​Database-turboSync.cpp	Uses pinned statement guards during sync loading.
native/​shared/​Database-sqlite.cpp	Implements cache eviction and guarded statement preparation.
native/​shared/​Database-batch.cpp	Uses pinned statement guards for batch operations.
CHANGELOG-Unreleased.md	Documents JSI error handling and statement caching changes.

_{Review effort: Balanced}

[Copilot]

🟡 Changes recommended

A failed reset leaves the public Database API using ErrorAdapter and blocks retries, while the added tests do not exercise real native recovery or the public reset wrapper.

2 open issues

• Restore adapter after reset failure to allow public API retries · New
• Add JSI integration coverage for reset recovery and retry · New

2 resolved since last review

• Recovery reset asserts on uninitialized state and leaves DB uninitialized
• Close SQLite handle before throwing on open failure

_{Review effort: Balanced}

[Copilot]

🟡 Changes recommended

A failed reset can clear JS models while preserving adapter-side cached IDs, leaving the caches inconsistent and causing subsequent reads of cached records to throw.

3 open issues

• Restore adapter after reset failure to allow public API retries
• Failed reset can desynchronize JS and adapter record caches · New
• Add JSI integration coverage for reset recovery and retry

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

Human assessment is needed for cross-layer cache consistency after reset failures and potential native process aborts on empty or comment-only SQL.

3 open issues

• Restore adapter after reset failure to allow public API retries
• Failed reset can desynchronize JS and adapter record caches
• Add JSI integration coverage for reset recovery and retry

_{Review effort: Balanced}

[joker-777]

Re the review summary on e8762d0 ("potential native process aborts on empty or comment-only SQL"): fixed in d1b4c62, together with the other asserts that SQL from Q.unsafeSqlQuery can trigger. In prepareQuery, a statement that compiles to nothing (empty or comment-only SQL) now throws a JS error instead of hitting assert(statement != nullptr). In query, queryAsArray and queryIds, a first column other than id throws a JS error instead of failing an assertion, and release builds no longer read the wrong column as the id. count only requires a result column instead of asserting exactly one, matching the other bridges. The remaining asserts guard internal invariants or fixed SQL. New shared test "rejects raw SQL queries it cannot run (JSI)" runs on the Android device suite in JSI mode. CI on e8762d0: the JavaScript jobs (18 to 23) and Android pass. iOS fails at the simulator destination as on every fork run.

[Copilot]

🔵 Needs a closer look

Human assessment is needed because a schema-encoding failure before reset dispatch can leave JavaScript and driver caches inconsistent, causing subsequent record fetches to fail.

No open issues 🎉

3 resolved since last review

• Restore adapter after reset failure to allow public API retries
• Failed reset can desynchronize JS and adapter record caches
• Add JSI integration coverage for reset recovery and retry

_{Review effort: Balanced}

[joker-777]

Re the review summary on d1b4c62 ("a schema-encoding failure before reset dispatch can leave JavaScript and driver caches inconsistent"): no code change. SQLiteAdapter.unsafeResetDatabase() encodes the app schema with encodeSchema(), and its only throw is in migration-step encoding (Unsupported migration step), which this path does not use. The schema has already passed appSchema() validation, and a schema that could not be encoded would fail first in setUpWithSchema, which encodes the same schema. Every failure that can happen once the reset reaches an adapter is covered by clearing the adapter's cached record IDs first (e8762d0), for all five adapters.