Vala async retry pattern: re-invoking a GError-throwing method without duplicating writes

Answer

Yes – the standard way to retry a Vala async method that may throw a GError is to wrap each yield in a try/catch block and, on catching a transient error, simply yield the same async method again. Each call creates a fresh GLib.Task, so no state is reused and the continuation is scheduled sequentially on the main context.

Confirmed facts

• A completed GLib.Task cannot be restarted; therefore each retry must invoke the async method anew.
• The GError only supplies domain, code and message; there is no built‑in guarantee about whether the underlying write happened, was partial, or never started.
• Retrying from within the same async caller does not create re‑entrancy hazards; the caller is suspended at each yield and the main loop runs other sources before resuming.

Likely explanation / needed steps

1. Define a bounded attempt counter (e.g. max_attempts = 3).
2. In an async wrapper method, loop while attempt < max_attempts:
3. Wrap yield obj.write_async(...) in try/catch (GLib.Error e).
4. If the catch block runs, inspect e.domain and e.code. Retry only for error codes that your API documents as transient (e.g. network timeouts, temporary lock failures).
5. If the error is not transient, re‑throw or handle as permanent.
6. On a successful yield, break the loop.
7. Optionally insert an asynchronous back‑off between attempts: yield GLib.timeout_add_seconds_async(backoff_seconds, () => { return false; }) (or use a timeout source via GLib.MainContext).
8. To avoid duplicate writes, make the operation idempotent (unique key, sequence number, or pre‑write check) or design the callee to fail before any write when possible.

Missing diagnostic detail

To decide whether a caught GError is safe to retry, you need to know which specific error domains/codes your write operation treats as transient. If you can share those codes (or confirm that all errors from this method are transient), the recommendation can be finalized; otherwise you must filter retries to the known transient set.