This seems like it would degrade service on networks like testnet with a low level of addresses / address traffic, where the local address would be broadcast far less frequently (versus current code).

A time based rotation (clearing) is not equivalent to a traffic based rotation.

@jgarzik The local address is broadcast to all outbound connected peers on connect.

Agree with @jgarzik. I'm not convinced that this is a good idea, specially rebroadcasting our own address once in a while is useful, I don't think we should rely on the implementation details of the rolling bloom filter to do this as a side-effect.

I guess I'll remove this part from #7125 as well then.

EDIT: #7125 only removes the wipe-the-filter logic, not the rebroadcast logic.

@sipa Well unless you think our reasoning is wrong? Is there some privacy compromise here?
(at some point we might replace the rolling bloom filters with something else again, and we may have forgotten the decision to remove this, hence I'm somewhat in favor of doing it explicitly)

I guess there are a few ways forward:

• Keep as-is, drop the entire filter every 24 hours + rebroadcast.
• Don't drop the filter, but still rebroadcast your own address periodically in a forceful way (bypassing the filter).
• Don't drop the filter, but still rebroadcast periodically (subject to the filter, which means no rebroadcast if there was not enough noise to hide the data in).
• This PR (no drop, no rebroadcast).

I think I prefer 3.

@sipa how about randomly settings bits in the filter to 0 instead of completely clearing it?