IssueTracker

There are a few more changes to DataStoreImpl's logic which makes a reevaluation necessary.

Now with Yigit's changes, datastore's cold flow always tries to do a dirty read by readState which calls readAndInitOrPropagateAndThrowFailure that invokes readDataOrHandleCorruption only in InitDataStore's first run. Any following read is triggered by IPC's listener which calls readDataAndUpdateCache which updates the cache to ReadException returned from attemptRead.

A ReadException is not automatically handled and stays in cache until a updateData call which revisits readAndInitOrPropagateAndThrowFailure, but it doesn't handle the corruption exception because it's already not the first run. In such case, the handleUpdate call proceeds to transformAndWrite which does not catch the exceptions thrown from readDataFromFileOrDefault.

can we create a test failure for this? I'm not fully sure what is going wrong here, it is a bit hard to track w/o a repro case.

can we create a test failure for this?

Sure I'm working on the code changes.

In the meantime I think this bug might makes the errors e.g. in b/309382591 harder to recover because it doesn't try to handle the corruption until the app restarts so the DS can readAndInit again (with corruption handler logic).