UTXOs are like a coin you can split and join. The other option is accounts, fuck that. The database doesn't have to index everything. It should index in reverse and not store anything spent, and the older, the less optimized, as it's unlikely to ever be spent the further you go back.

Yes, but I think he was only talking about nodes. I don't think utxos can be gotten rid of, but maybe there's another option than holding the utxo set in ram. Idk, maybe he'll clarify That partial index sounds good. Are there any other options?

Just have the chainstate be queriable. Here, Eric Voskuil jumps in the thread to explain it pretty well the the Bitcoin Core Devs:

Delving Bitcoin

Libbitcoin for Core people

I don’t understand why inputs are being fetched from a cache at all. Only one block at a time is being validated, and the block had to have just ...

It's tough to wrap the mind around but memory maps are very fast to query and don't have to have a list of UTXOs. On a per block basis, "is the coin being double spent?" and "is the coin there(in the spending address)?" solves the issue of needing to know where all the coins are simultaneously.