Acquiring a QtRO replica per call (acquireDynamic + waitForSource) is expensive:
under a tight loop — e.g. a proxy forwarding every method to its target, or a UI
backend driving a whole surface — it dominates and can even starve the nested
synchronous calls. Cache the LogosObject handle per object name in m_objectCache
and reuse it across calls (both the sync invokeRemoteMethod and the async
invokeRemoteMethodAsync paths); no per-call release(). A stale handle (source
went away — module unloaded / transport dropped) is detected via a new
LogosObject::isValid() (QtRO replica state == Valid) and transparently
re-acquired. The cache is released in clearObjectCache() from the destructor and
before reconnect().
- logos_object.h: add virtual bool isValid() (default true).
- qt_remote/remote_transport.{h,cpp}: RemoteLogosObject::isValid() (replica
Valid state) + a process-wide acquireCount() test hook.
- logos_api_consumer.{h,cpp}: m_objectCache + acquireCachedObject()/
clearObjectCache(); sync + async reuse the cached handle; async keeps the
QPointer guard and never releases the shared handle from its callback.
Test: RemoteEventTest.ConsumerReusesCachedHandleAcrossSyncAndAsyncCalls publishes
a provider over the qt_remote host, does 12 sync + 12 async echo calls, and
asserts every result is correct AND acquireCount() == 1 (one replica for all 24
calls). 164/164 green.
A `concurrency:"multi"` call's result comes back as a deferred completion
event (`__logos_call_complete__`), delivered by RemoteEventHelper::onEventResponse
— a slot fired by the replica's eventResponse signal. Cross-process, that slot
runs on QtRO's read stack (QRemoteObjectNodePrivate::onClientRead). Until now the
async user callback was invoked *inline* there, and that callback routinely (a)
emits a module event — which the host-side ModuleProxy serializes onto the QtRO
source — and (b) release()s the client object. Doing either while onClientRead is
still unwinding re-enters QtRO and corrupts the node: a SIGSEGV in onClientRead
(EXC_BAD_ACCESS, KERN_INVALID_ADDRESS at 0x80). This is the crash the EVM wallet
backend hit from refresh_balances, which fans balance reads out to eth_rpc via
call_async and then emits `balances_updated` from the gather completion.
Primary fix (remote_transport.cpp): deliver the async completion callback on the
next event-loop turn via QTimer::singleShot(0, m_helper, …) instead of inline, so
all user code (event emits, release(), further calls) runs after onClientRead has
fully unwound. m_helper is the context so the callback is dropped if the object is
torn down first.
Defense-in-depth for the same re-entrancy class:
- remote_transport.cpp release()/disconnectEvents()/dtor: deleteLater() the helper
(signal receiver) and replica (signal sender) and disconnect first, instead of
deleting them inline — deleting a QObject mid-emission corrupts the connection
list Qt is iterating.
- module_proxy.cpp: always queue the source eventResponse emit to the owning
thread (Qt::QueuedConnection), never emit inline, so a module that emits from
inside a same-thread dispatch can't re-enter QtRO's source serialization.
Tests (tests/protocol/test_remote_transport_events.cpp, newly wired): qt_remote
LocalSocket event delivery (direct + full provider chain) and a reentrant-release
regression that drives release() from inside a deferred-completion callback. The
hard crash only reproduces cross-process (in-process QtRO posts the event, so the
read stack has already unwound) — the cross-process guard is the wallet Anvil
integration doctest, where this fix is A/B-proven: the published backend crashes
on refresh_balances, the patched backend returns balances cleanly.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Dario Lipicar