Let me know if I'm missing something, but if there's some data we don't need in the backup, shouldn't we prevent it from being included, instead of clearing it after restoration?

In D11077#319547, @ashoat wrote:

Let me know if I'm missing something, but if there's some data we don't need in the backup, shouldn't we prevent it from being included, instead of clearing it after restoration?

The reason for that is probably due to my concerns that I outlined in D10780. I think that you are right - we should avoid backing those data up instead of removing them after restoration. The problem is that those data are "invisible" in C++ since for KeyserverStore we don't have a C++ db entity like for MessageStore but we just store a JSON string under some id key instead. We could do parsing with folly but it is inefficient, buggy and shows weakness of db design.
The best solution is to migrate db and have a C++ struct for KeyserverStore matching the JS type. With this we would be able remove device specific data from backup with simple SQL query. But the fastest solution to ship is to remove such data in JS after restoration.

Is there a codepath that will lead to populating data that we remove here with accurate data?

I'm not entirely opposed to parsing the JSON in C++ via Folly. Seems better than storing the data only to throw it away