There were a couple transactions before, each sending 10 BTC to the wallet itself. So the wallet should still have a balance of 50 – fees BTC, and presumably 3 UTXOs of 10 BTC, 10 BTC, and 30 - fees BTC. I‘m not sure why APS (avoid partial spend) would make a difference here. APS just means that we’ll force use of all coins associated with the same scriptpubkey at once rather than a single UTXO if there is address reuse, but it seems to me that funds are always sent to new addresses here.
It seems to me that this test works because BnB is disabled due subtractfeefromamount being active, Knapsack finding the single input solution and Knapsack being run before SRD and therefore Knapsack’s result being preferred, but it’s not obvious why this transaction would generally end up not having any change if e.g. we change how we pick the input set from the various coin selection algorithm results. It might be more future proof to get the current balance, and then send the entire balance while setting subtractfeefromamount.
Usually I would tend to suggest sendall to make a changeless transaction, but sendall doesn’t use coinselection which is what we are actually trying to test here.
So maybe, these two test-cases could just be folded into one and the distinction for APS could be dropped, or the first should perhaps also spend the entire balance.
