spend: feerate using rounded-up vbytes can be lower than target value

[jp1ac4]

This issue came up recently in #1125 (comment) and also in the past.

When creating a new spend, we target a feerate in terms of sat/vb, but the bdk_coin_select crate we use for coin selection and determining the change amount uses sat/wu, which can lead to the feerate in sat/vb being lower than the target value, given that we round up when converting from wu to vb: vb = ceil(wu/4).

One option to fix this is to modify the bdk_coin_select crate so that, if a feerate is specified in sat/vb, it calculates the excess and change amount of a given selection based on the size in vb of the transaction rather than its weight.

[darosior]

Michael opened an issue about this upstream: bitcoindevkit/coin-select#26.

[darosior]

To be clear, and to make this appear in searches: this can in particular make a transaction feerate be below min fee (for 1sat/vb txs) thereby making the transaction not broadcastable.

[darosior]

Preventing the transaction from broadcasting is probably the worst case, and could probably be worked around short term for v7 (h/t @kloaec) by slightly increasing the feerate asked by the user when it's 1sat/vb. For instance making it 1.1 or 1.2 sat/vb. I guess this could work.

[jp1ac4]

I don't think this specific rounding issue can lead to a feerate below 1 sat/vb:

• If we enter 1 sat/vb, coin selection uses 0.25 sat/wu.
• The fee paid will be at least ceil(0.25 * weight) (need ceiling to ensure integer fee).
• Then vsize = ceil(0.25 * weight) and so feerate in sat/vb will be fee / vsize >= ceil(0.25 * weight) / ceil(0.25 * weight) = 1

[jp1ac4]

It's worth noting that both the coin selection crate and mempool.space are using vsize = weight/4, without any rounding. If we were to follow the same approach, then we wouldn't face this particular issue.

As per my previous comment (#1132 (comment)), I don't think using this approach could lead to a feerate below 1 even if rounding up.

Unless I'm missing something, I would favour this approach of not rounding up vbytes.

[jp1ac4]

Unless I'm missing something, I would favour this approach of not rounding up vbytes.

It seems that rounding up when calculating the feerate is consistent with Bitcoin Core (bitcoindevkit/coin-select#26 (comment)) so let's stick to that for now.

[jp1ac4]

The fix has now been merged upstream (bitcoindevkit/coin-select#29) so we just need to wait for the next release of the crate.