MissingNonWitnessUtxo even before signing

[stevenroose]

When you provide witness_utxo but not non_witness_utxo, previously BDK would complain at sign-time. The solution was in the SignOptions:

			let opts = SignOptions {
				trust_witness_utxo: true,
				..Default::default()
			};

But now I am already getting this error at tx creation time when calling TxBuilder::finish.

I see that TxBuilder has a only_witness_utxo option now, but only affects TxBuilder's internal behavior and doesn't resolve the error. I am getting this error when not setting only_witness_utxo and by adding foreign inputs.

[ValuedMammal]

Would it be possible to come up with a test or code snippet that you believe should work if the bug was resolved?

[stevenroose]

Here: https://github.com/stevenroose/bdk/tree/repro-foreign-utxo

So basically, removing that "only_witness_utxo" makes the build step fail. This might be intended behavior? But it's weird that if you don't indicate anything it will complain that non-witness utxo data is missing and you fix it by saying that you will not provide non-witness utxo data?

Before, only the signer would error when there was no non-witness utxo data and you would make the signer not error by adding the SignOptions field to make it trust the witness utxo data.

[stevenroose]

Like, in the very least, the "only_witness_utxo" method should be re-documented to say "do not require non-witness utxo data" instead of "do not provide non-witness utxo data". Because for foreign utxos, without that method, the builder also only provides witness utxo data but then somehow it is a problem while if you specifically tell you to do the same thing, it is not a problem.

I hope you can see what I'm saying. It might just be ok like this I guess, but it was a regression that wasn't like this before.

Also, you might want to provide non-witness utxo data for non-foreign inputs for example. Some signers might want/need them only for certain inputs so you can ignore them on the other ones. We use the foreign inputs for fee anchors and their value is negligible and their satisfaction is just the OP_TRUE witness script, so they might be treated very differently by signers.

[ValuedMammal]

As a followup question do you think this logic should be changed to allow both taproot and segwitv0 inputs to not have a non_witness_utxo?

bdk/crates/wallet/src/wallet/mod.rs

Lines 2170 to 2180 in 7969898

	let is_taproot = foreign_psbt_input
	.witness_utxo
	.as_ref()
	.map(|txout| txout.script_pubkey.is_p2tr())
	.unwrap_or(false);
	if !is_taproot
	&& !params.only_witness_utxo
	&& foreign_psbt_input.non_witness_utxo.is_none()
	{
	return Err(CreateTxError::MissingNonWitnessUtxo(outpoint));
	}