Response errors; unexpected block execution error OutputAlreadyExists When Running Multiple Tests about fuels-rs HOT 10 CLOSED

Update: managed to replicate, have a few ideas of what might be causing it. Will post an update soon!

from fuels-rs.

Alright, this was a tricky one to debug, and it highlights some shortcomings in the fuel-core and the SDK. Many thanks to @Voxelot for helping me

The issue begins at this fundamental truth: every call we make at the SDK/client level generates a transaction.

That means we don't have a concept of read-only .call()-like method. Like this for instance.

That said... Sushi Sway is doing a couple of things:

1. deposit (write)
2. check balance ("read")
3. withdraw (write)
4. check the balance again ("read")

The issue here is that (2) and (4), the two checks, look like this:

let balance_of_check0 = mycontract_mod::BalanceOfInput {
    address: [10; 32],
    asset_id: [0; 32],
};

let initial_bal = contract_instance
    .balance_of(balance_of_check0) // <--- first call to balance_of
    .call()
    .await
    .unwrap();

and

let balance_of_check1 = mycontract_mod::BalanceOfInput {
    address: [10; 32],
    asset_id: [0; 32],
};

let final_bal = contract_instance
    .balance_of(balance_of_check1) // <--- second call to balance_of
    .call()
    .await
    .unwrap();

Notice that, because balance_of_check1 and balance_of_check0 are the same (same address and asset_id), the transactions generated by balance_of will be the same. Meaning the same transaction id, meaning OutputAlreadyExists.

The shortcoming on our end is that a method that's supposed to be read-only is actually generating a transaction when it shouldn't. But we don't have an API for read-only yet.

Short-term very weird workaround: make BalanceOfInput have an unused RNG-based field so that balance_of_check0 and balance_of_check1 are different and will then generate different transactions, avoiding the collision.

The actual solution, as suggested in a conversation with @Voxelot, to split the generated methods (by the abigen) between .send() (or something) that will actually create a transaction and a .call(), that underneath will use fuel-core's dry-run to not create a transaction, meaning these read-only functions can be called many times and not cause tx id collision.

cc @adlerjohn once we agree on the solution, the semantics, and the naming of .call() and .send() I'll create an issue so that we can tackle it asap.

from fuels-rs.

Yeah, makes sense to separate call and send in the same way as Ethereum, and using dry run. Good work debugging!

from fuels-rs.

Should this issue be transferred to fuels-rs @digorithm ?

from fuels-rs.

Note that we also need a solution for non-view-only calls that happen to modify state rather than being inherently view-only. This could be accomplished by inserting a random coin input. Or better, a change output to a random address!

from fuels-rs.

Does this happen when running tests sequentially? My first guess is that if you run tests more than once, it'll try to deploy the same contracts again. Since they already exist in the state, you'll get that error. If can be fixed by clearning ~/.fuel and restarting your node. Or, running the node with an in-memory db: --fuel-core --db-type in-memory.

from fuels-rs.

It still seems to be happening even if I comment out all the other tests, rm ~/.fuel and restart my computer, not sure what is causing it then. Can you replicate?

from fuels-rs.

Guessing this is probably an SDK issue then. cc @digorithm to triage.

from fuels-rs.

That's odd. I always get this error when deploying a contract already deployed, as expected. And then removing the fuel dir does it for me.

Weird that removing the dir isn't resolving the issue in your case, I'll do some triaging today!

from fuels-rs.

Tracking this here #90

from fuels-rs.