Re: [bitcoin-dev] TLA+ specification for Succint Atomic Swap

[Dmitry Petukhov <dp@simplexum.com>]

В Thu, 14 May 2020 07:31:13 +0200
Ruben Somsen <rsomsen@gmail.com> wrote:

> Hi Dmitry,
> 
> >While refund_tx_1 is in the mempool, Bob gives success_tx to the
> >friendly miner
> 
> I see, so you're talking about prior to protocol completion, right
> after Alice sends Bob the success_tx. The reason this is not an issue
> is because Alice and Bob both had to misbehave in order for this to
> happen. Bob is misbehaving here because he should have published the
> success_tx before refund_tx_1 became valid, and Alice is misbehaving
> here because she should have sent the revoke_tx (which invalidates
> the success_tx) followed by refund_tx_2 (revealing her secret only
> AFTER Bob can no longer claim the BTC). In other words: yes, the
> protocol can fail if Alice and Bob together work towards that goal. A
> feature, not a bug. This won't happen if either of them doesn't want
> it to. I imagine this is difficult to model.

Right. But it should be noted that it is not enough that Bob publishes
success_tx before refund_tx_1 became valid. The success_tx needs to be
confirmed before refund_tx_1 became valid.

Only Bob can spend success_tx so this is unlikely to be the practical
problem, unless the original fee of success_tx is too small and Bob
epically screws up CPFP-ing it.

> >Bob will receive BTC, and the LTC can be locked forever, but Bob
> >doesn't  
> care, he got his BTC.
> 
> No, because diagram step 5 comes before step 6 -- Alice won't give
> her key until she learns secretBob.

I somehow missed it, and steps 5 and 6 in the diagram was not modelled
at all (on the other hand, it made the model simpler and I had
something working relatively quick). I now made the `signers_map` into
variable that can be changed to give Bob the ability to sign for Alice.

With that change, step 6 can be modelled, but this will add a bunch of
new txs to the model (each Alice&Bob spend will have 'Bob unilateral
override' case)