Anthony Towns

> This is insufficient for preventing transactions that will never be mined from staying in the mempool, because `TrimToSize` would leave transactions in the mempool if a parent and multiple...

> @ajtowns I think that gets tricky due to sibling pays for parent situations; in fact I think you could evict something under that criteria in situations where no RBF...

I guess I don't really understand how this PR on its own helps anything -- we don't do package relay yet, so except for a regtest only rpc, these changes...

> Right, the changes for package relay only (i.e. no v3, no package RBF) would be this, persist packages across restart, and allow any ancestor package instead of just child-with-parents....

ACK 6fb01d26c57f6af7205721e528bbafee8fa41027

> I'm not really sure where to draw the line between useful and not useful I think "submitpackage" would count as obviously useful if it was available on mainnet --...

Not an objection, but this feels backwards to me: presumably the user has manually listed those peers as addnode/connect because it's important that their node be connected to those peers,...

This seems like a non-starter to me? It's not even able to detect obvious double locks at compile-time: ```c++ ThreadSafePtr m; m->emplace(5, 25); { auto m_locked = *m; m_locked->emplace(6, 36);...

> I assume it can be annotated just like any other lock. Agree implementation should fix this, though. I think you'd have to mark the `Synced` object as being lock...

> I compiled with #28217, which sets default value of `permitbaremultisig` to false in `policy.h` and tried to run this test case with `'allowed': False` for expected result but the...