First of all, I’d note BIP37 user != SPV. You can think of another kind of implementation of BIP37, which filters, for example, 25% of all the addr space (not sure it’s practical currently w.r.t filter size), doing that over N=20 connections with different filters to receive all transactions overall.
Given how unrealistic that is today, we can probably implement Erlay support for that once we know about this scenario?
Secondly, let’s consider a regular small-filter (say, 1000 of addrs for redundancy) SPV node. If it’s connected to X serving nodes, it’d still get X times more tx announcements than needed. But considering how little is that bandwidth, reducing this redundancy might be useless, so yeah.
At the same time, I don’t think this justification is particularly strong, but yeah, in the end we could just suggest SPV nodes to not send SENDRECON if they don’t want. In that case, the outcome is that regular nodes would send SENDRECON
even though ‘fRelay=false’, and in some cases that would be useless, but that’s not a big deal.
Do you agree with this thinking?