| Previous | Next |
| RO_E_CLOSED | RO_E_CHANGE_NOTIFICATION_IN_PROGRESS |
RO_E_EXCLUSIVE_WRITE
Concurrent writer violated exclusive-write requirements
RO_E_EXCLUSIVE_WRITE is HRESULT 2147483668 (0x80000014) from winerror.h. AllStat describes it as “Only one thread may access the object during a write operation.” In the Windows Runtime metadata, object lifetime, asynchronous operation, activation, and apartment model, the code identifies a specific failure boundary and should not be replaced by a generic COM exception.
Telemetry should preserve this result as a separate outcome rather than merging it with unrelated COM or WinRT failures.
Decisive interpretation boundary
Before choosing recovery, verify that the active writer owner, critical section, transaction, and operation generation are known before another writer is started. That boundary prevents this result from being misclassified as corruption, network failure, or a reason for an unsafe automatic retry.
Also confirm that all observed objects, tokens, buffers, proxies, metadata files, or ACLs belong to the current operation generation and were not retained from an earlier attempt.
Where it is encountered
- This result can be returned during C++/WinRT, WRL,.NET projections, or native WinRT ABI calls. Record the interface, method, component version, thread, apartment, process, and correlation ID.
- This result can be returned during runtime-class activation, WinMD reflection, asynchronous objects, observable state, or view lifecycle.
- It can be returned during cross-thread and cross-apartment handoff between UI, ASTA, STA, MTA, broker, and background work.
The immediate focus for it is a WinRT object or storage surface that permits only one writer during a mutation or transaction. Keep it attached to that operation; the same numeric severity outside the owning API does not supply enough context.
Correct handling and recovery
Serialize writes, await the existing commit or rollback, and release writer ownership on every exception path. Do not solve the conflict with blind retries.
An unchanged retry loop for it can duplicate effects, deepen reentrancy, or hide an invalid lifetime transition.
Difference from nearby HRESULTs
It reports competing write ownership; E_CHANGED_STATE reports a state version invalidated by interleaving work.
Merging it with the neighboring HRESULT would choose the wrong retry, recreation, authorization, or cleanup path.
Lifetime, retry, and cleanup rules
After it, determine whether the current object, interface pointer, call context, token, stream, metadata reader, asynchronous operation, or access-control instance remains valid. Release only resources owned by the failing attempt, cancel callbacks through their documented mechanism, and avoid double close, double commit, repeated activation, or replay of a non-idempotent remote method.
The retry policy for it should state the trigger, maximum attempts, cancellation owner, and reconciliation step for effects that may have completed elsewhere.
Practical scenario
Two autosave tasks write the same runtime buffer. A per-document async mutex ensures the second waits for the first commit and then recalculates its delta.
A regression test should reproduce it, assert the raw HRESULT and all relevant outputs, then correct only the decisive condition and verify the intended success or neighboring failure result.
References
- Microsoft: C++/WinRT error handling
- Microsoft: concurrency and asynchronous operations
- Microsoft: WinMD files
- Microsoft: IAgileObject
- Microsoft: COM apartments
- Microsoft: HRESULT values
Looking for a different code? Search another status or error code.