Node wallet's fee estimation can be off by 2 bytes? #19021

I'm writing functional tests that ensure that my code sending transactions using sendtoaddress is sent with the right transaction fee. These tests are failing, even though the calculation checks out.

Then I had a look at the functional tests and found this: https://github.com/bitcoin/bitcoin/blob/362f9c60a54e673bb3daa8996f86d4bc7547eb13/test/functional/test_framework/util.py#L40

The node's fee calculation can be off by 2 bytes? What is the reason for this and can it be fixed?

To reproduce

Rpc settxfee: 0.00100 (100 sat/b) Send transaction (sendtoaddress) (2segwit inputs, 1 wrapped segwit output - Not an important detail)

The transaction reports this: Fee: 0.00020680 vSize: 201 calculate sat/b from this and you get ~102

Expected behavior

The fee/b as set in settxfee should be used exactly (100)

Actual behavior

Effective sat/b fee of ~ 102

System information

Bitcoin core 0.18.1

What is the reason for this

Coin selection will add dust and otherwise negligible amounts to the fee instead of creating unspendable change outputs.

can it be fixed?

No. It is not a bug.

Coin selection will add dust and otherwise negligible amounts to the fee instead of creating unspendable change outputs.

Thanks for the explanation. That makes much more sense now. It seemed my wallet had some well split utxos hence was producing the 1 output and no change. Though i did note that the fee calculated was quite divergent in these cases (not really an issue as you suggest)

not really an issue as you suggest

Ok, closing this issue for now