[bitcoin-dev] Forget dormant UTXOs without confiscating bitcoin

[jl2012 at xbt.hk]

It is a common practice in commercial banks that a dormant account might 
be confiscated. Confiscating or deleting dormant UTXOs might be too 
controversial, but allowing the UTXOs set growing without any limit 
might not be a sustainable option. People lose their private keys. 
People do stupid things like sending bitcoin to 1BitcoinEater. We 
shouldn’t be obliged to store everything permanently. This is my 
proposal:

Dormant UTXOs are those UTXOs with 420000 confirmations. In every block 
X after 420000, it will commit to a hash for all UTXOs generated in 
block X-420000. The UTXOs are first serialized into the form: 
txid|index|value|scriptPubKey, then a sorted Merkle hash is calculated. 
After some confirmations, nodes may safely delete the UTXO records of 
block X permanently.

If a user is trying to redeem a dormant UTXO, in addition the signature, 
they have to provide the scriptPubKey, height (X), and UTXO value as 
part of the witness. They also need to provide the Merkle path to the 
dormant UTXO commitment.

To confirm this tx, the miner will calculate a new Merkle hash for the 
block X, with the hash of the spent UTXO replaced by 1, and commit the 
hash to the current block. All full nodes will keep an index of latest 
dormant UTXO commitments so double spending is not possible. (a 
"meta-UTXO set")

If all dormant UTXOs under a Merkle branch are spent, hash of the branch 
will become 1. If all dormant UTXOs in a block are spent, the record for 
this block could be forgotten. Full nodes do not need to remember which 
particular UTXO is spent or not, since any person trying to redeem a 
dormant UTXO has to provide such information.

It becomes the responsibility of dormant coin holders to scan the 
blockchain for the current status of the UTXO commitment for their coin. 
They may also need to pay extra fee for the increased tx size.

This is a softfork if there is no hash collision but this is a 
fundamental assumption in Bitcoin anyway. The proposal also works 
without segregated witness, just by replacing "witness" with "scriptSig"