Single flashloan encoding #3279

sunce86 · 2025-02-13T18:03:10Z

Description

Tackles #3218

Implements on flavor of the "repay debt with collateral" idea an a e2e test. The trader is a gnosis safe to make it easier to do all the necessary pre-interactions. The basic flow is like this:

make trader a safe
deposit 50K USDC
borrow 1 WETH
build appdata
- flashloan 1 WETH (flashloan hint)
- repay 1 WETH (1st pre-hook, tx pre-signed by safe)
- withdraw 50K USDC (2nd pre-hook, tx pre-signed by safe)
place order with that appdata like:

{
    "from": "<SAFE_ADDRESS>",
    "receiver": "<SETTLEMENT_ADDRESS>",
    "sell": "50_000 USDC",
    "buy": "1 WETH + flashloanFee",
    "kind": "buy",
}

assert that trade got executed, indexed and that changed balances make sense

This flavor of "repay debt with collateral" leaves trace amounts of debt in the AAVE pool. The alternative would be either over paying slightly or a bunch of very complex accounting logic using another helper contract. Both are more involved and not necessary to proof the concept.

Changes

some bug fixes / workarounds that were necessary to make it work:

introduce exceptions in filtering logic based on balances in autopilot and driver to prevent the order from getting filtered out
forked e2e test setup currently does some things that will not needed once the flashloan contracts are actually deployed:
- deploy contracts
- allow list them as solvers in the settlement contract
code to pipe the flashloan data around

Things not addressed by this PR

proper configuration options in the driver
more convenience for computing flashloan amounts that need fees
- currently the fee gets added on top of the flashloan amount which means the user needs to be aware that their buyAmount needs to be appdata.flashloan.amount + fee. Ideally the driver would reduce the flashloan amount such that loanedAmount + fee = totalSignedAmount.
error handling
supporting multiple flashloans in the same settlement

How to test

added new forked e2e test
adjusted unit test for solvable orders filtering exception for flashloan orders

crates/e2e/src/setup/deploy.rs

MartinquaXD · 2025-02-24T11:47:03Z

crates/driver/src/domain/competition/order/app_data.rs

+                            serde_json::from_str(&response.text().await?)
+                                .context("invalid app data document")?;
+                        match appdata.full_app_data == "{}" {
+                            true => None, // empty app data


I think to avoid fetching the underlying app data for 0x000...000 (i.e. "{}") over an over again we need to store a dummy value for that hash.

MartinquaXD · 2025-02-24T11:49:02Z

crates/driver/src/domain/competition/solution/encoding.rs

@@ -173,6 +173,51 @@ pub fn tx(
        min: None,
        prices: auction.prices().clone(),
    };
+
+    // Add all interactions needed to move flash loaned tokens around


Probably makes sense to move this and the other flashloan specific encoding logic into a separate function. Also technically all of this should happen at the very end to make sure no other interactions get added before or after the flashloan interactions.

MartinquaXD · 2025-02-24T13:18:42Z

crates/e2e/tests/e2e/flashloans.rs

+        // Receiver is flashloan wrapper, so that borrowed funds can be returned to the lender
+        receiver: Some(onchain.contracts().flashloan_wrapper.address()),


This is not how we probably end up using this feature. However, while thinking about how to correct this code I realized that we didn't think some of the details through enough. Fine for now to show that the wrapper contract works but ultimately we should probably have an e2e test that repays a debt with collateral since that is the primary use case for this feature AFAIK.

There was a discussion we had that led to this implementation:
cowprotocol/flash-loan-wrapper-solver#2 (comment)

I ended up having a discussion with Felix and Federico about the exact details for this. The decision was made that the underlying order should always have the settlement contract as the receiver. That allows the order to be oblivious of the flashloan helper contract that actually ends up getting used. The driver encoding would then have to take this into consideration and transfer the respective funds from the settlement contract to the flashloan helper contract itself.

Thanks.
Will adjust the code and tests.

MartinquaXD · 2025-02-26T11:13:56Z

crates/driver/src/domain/competition/solution/encoding.rs

+        .transfer_from(
+            contracts.settlement().address().into(),
+            flashloan_wrapper.address(),
+            flashloan.amount.0, // or order buy amount?


Our initial implementation assumes the values are always the same. But if they were not, using the flashloan amount would lead to better outcomes:
if flashloan.amount > order.buyAmount the solution would not simulate in the first place which encourages solvers to make sure the buy amount covers the flashloan amount. Unless the settlement contract has enough fees already buffered but then this would still count as negative slippage which is a punishment for the solver.
if flashloan.amount < order.buyAmount the difference would stay in the settlement contract instead of accumulate in the wrapper contract which is more convenient for us and would count as positive slippage the solver can claim when paying out rewards.

crates/e2e/tests/e2e/flashloans.rs

m-lord-renkse · 2025-03-05T12:40:46Z

crates/e2e/tests/e2e/flashloans.rs

+        }
+    };
+
+    // pay 9 bps of flashloan fee to the AAVE pool


where does this number come from?

I googled it. After double checking it turned out I got the wrong number. Updated it to be 5 bps. Still didn't link a reference in the source code in the driver encoding since this hopefully going to change before the release anyway.

m-lord-renkse · 2025-03-05T12:44:01Z

crates/e2e/tests/e2e/flashloans.rs

+        // Receiver is always the settlement contract, so driver will have to manually send funds to
+        // solver wrapper (flashloan borrower)


How does this happen in this test? is baseline doing this? or is it a TODO or not needed?

reworded the comment

m-lord-renkse · 2025-03-05T12:44:37Z

crates/e2e/tests/e2e/flashloans.rs

+        &order.data().hash_struct(),
+    )));
+
+    {


Please can you add comments to this section? 🙏

Instead of a comment I have a log explaining what is happening. Can move it above the block though to avoid confusion.

crates/e2e/tests/e2e/flashloans.rs

sunce86 · 2025-03-05T12:47:25Z

crates/e2e/tests/e2e/flashloans.rs

+    let current_safe_nonce = trader.nonce().await;
+
+    // Build appdata that does:
+    // 1. take out a 1 WETH flashloan for the trader


is not actually part of the appdata right, it's already done above?

Did you confuse flashloan with borrowing the WETH (which indeed happens above)?

The flashloan is not explicitly encoded in an interaction but it's still part of the appdata as the flashloan hint. Updated the comment to make this clearer.

crates/driver/src/domain/competition/solution/encoding.rs

mstrug · 2025-03-05T16:51:24Z

crates/driver/src/domain/eth/mod.rs

+#[derive(Debug, Clone)]
+pub struct Flashloan {
+    pub lender: ContractAddress,
+    pub borrower: ContractAddress,


The borrower shouldn't be an ordinary Address? Is there a requirement that it is a contract?

Good catch. Indeed any Address could be the receiver of a flashloan.

mstrug · 2025-03-05T17:07:19Z

crates/driver/src/infra/solver/dto/solution.rs

+                    solution
+                        .flashloans
+                        .into_iter()
+                        .map(|flashloan| eth::Flashloan {


Can this conversion be moved into From trait implementation in the dto Flashloan?
This function is quite long with lots of indentation, and moving all those conversions to implementation blocks will improve the code readability a bit.

From implementations turned out to be a double edged sword with the domain driven design that requires a ton of conversions because if you use an IDE to jump to the definition of the .into() call it will jump to the blanket implementation on the From trait which is not really helpful.

mstrug · 2025-03-05T18:06:52Z

crates/e2e/tests/e2e/flashloans.rs

+    .unwrap();
+
+    let trader_usdc = balance(&web3, trader.address(), usdc.address()).await;
+    assert!(trader_usdc > to_wei_with_exp(47_000, 6));


Why the trader is not getting back the whole collateral value?

~~Double that. From the logic above, I couldn't understand where 3000 USDC remain. A comment would be helpful.~~ Ah, it is related to 1 weth debt. Got it. Still, a comment would be helpful.

crates/driver/src/domain/competition/order/app_data.rs

squadgazzz · 2025-03-06T10:52:10Z

crates/e2e/tests/e2e/flashloans.rs

+    assert!(settlement_weth < 100_000_000u128.into());
+    tracing::info!("settlement contract only has dust amounts of WETH");
+
+    assert!(balance(&web3, trader.address(), ausdc).await < 10_000.into());


If the assertion above checks that the trader spent less than 3000 USDC on the loan, why do we have 10k aUSDC here? Should it be 3000 also?

aUSDC only tracks how much USDC you have currently deposited in AAVE (+ interest). Since the trader actually had to sell some USDC they wipe out their entire debt (close to 0 aUSDC) but don't get their entire USDC back (~3000 sold for 1 WETH).

Ok, but still not sure why 10k is used here. Should it be 3k due to the assertion above in the code?

assert!(trader_usdc > to_wei_with_exp(47_000, 6));

If this passes, and initially the trader had only 50k, then <3k of aUSDC remains, right? Or did I miss something?

Maybe you overlooked the different orders of magnitude. The user has ~3000 actual USDC less (i.e. 3_000_000_000 atoms). The remaining debt is only 10_000 atoms which is not even 1ct of debt.

Aaah, i knew I missed something, thank you!

Single flashloans

3972e2c

sunce86 self-assigned this Feb 13, 2025

sunce86 and others added 7 commits February 13, 2025 21:31

transfer tokens

48a784f

make tests pass

bd36c78

Make driver tests pass

cad6da8

Merge branch 'main' into flashloan-encoding

3c89f26

Move flashloan interactions around and update comments

e71438d

first part of e2e test

3684046

lint

2b95776

MartinquaXD reviewed Feb 18, 2025

View reviewed changes

crates/e2e/src/setup/deploy.rs Outdated Show resolved Hide resolved

sunce86 added 10 commits February 18, 2025 20:14

small fixes

e727be3

fixed test

d5cd792

lint

a47ddfa

Merge branch 'main' into flashloan-encoding

9297726

print flashloans

9afb861

fix appdata fetching

d61ef1a

amount serde fix

394625f

better detect empty app data

3f067ea

and fix driver test

44aa910

lint

748e5e4

MartinquaXD reviewed Feb 24, 2025

View reviewed changes

sunce86 added 8 commits February 25, 2025 11:24

Merge branch 'main' into flashloan-encoding

52963a2

use Abstract contract for encoding

69821c6

multiple lenders

af310fc

fix driver test

a54e1bd

Merge branch 'main' into flashloan-encoding

92430ec

driver tests

17baf85

order receiver is always settlement contract

ec30dc0

partially fillable false

83d1933

MartinquaXD reviewed Feb 26, 2025

View reviewed changes

MartinquaXD added 7 commits March 5, 2025 10:29

Remove duplicated deployment workaround

d11bbae

add TODO comment

701a260

Add nonce() function on Safe instead of exposing underlying contract

7b0ff5c

Merge remote-tracking branch 'origin/main' into flashloan-encoding

0f4ee0c

fixup merge

b282a15

Better name and comment for e2e test

311f7d6

relax aUSDC check since it causes issues in CI

b90064f

MartinquaXD marked this pull request as ready for review March 5, 2025 11:25

MartinquaXD requested a review from a team as a code owner March 5, 2025 11:25

m-lord-renkse reviewed Mar 5, 2025

View reviewed changes

crates/e2e/tests/e2e/flashloans.rs Show resolved Hide resolved

m-lord-renkse reviewed Mar 5, 2025

View reviewed changes

crates/e2e/tests/e2e/flashloans.rs Show resolved Hide resolved

sunce86 commented Mar 5, 2025

View reviewed changes

MartinquaXD added 3 commits March 5, 2025 13:11

Reduce fee to 5 BPS

6b16485

update comment

ba963c6

Comments

ced2e49

mstrug reviewed Mar 5, 2025

View reviewed changes

squadgazzz reviewed Mar 6, 2025

View reviewed changes

crates/driver/src/domain/competition/order/app_data.rs Outdated Show resolved Hide resolved

squadgazzz reviewed Mar 6, 2025

View reviewed changes

MartinquaXD added 4 commits March 7, 2025 12:20

add comment in test

b4af917

.then_some() instead of match

13a9da7

Address instead of ContractAddress

7fe4303

Merge branch 'main' into flashloan-encoding

2d13cb8

MartinquaXD changed the title ~~[DRAFT] Single flashloan encoding~~ Single flashloan encoding Mar 7, 2025

MartinquaXD approved these changes Mar 7, 2025

View reviewed changes

MartinquaXD merged commit 169cb46 into main Mar 7, 2025
11 checks passed

MartinquaXD deleted the flashloan-encoding branch March 7, 2025 12:31

github-actions bot locked and limited conversation to collaborators Mar 7, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Single flashloan encoding #3279

Single flashloan encoding #3279

sunce86 commented Feb 13, 2025 •

edited by MartinquaXD

Loading

MartinquaXD Feb 24, 2025

MartinquaXD Feb 24, 2025

MartinquaXD Feb 24, 2025

sunce86 Feb 24, 2025

MartinquaXD Feb 25, 2025

sunce86 Feb 25, 2025 •

edited

Loading

MartinquaXD Feb 26, 2025 •

edited

Loading

m-lord-renkse Mar 5, 2025

MartinquaXD Mar 5, 2025

m-lord-renkse Mar 5, 2025

MartinquaXD Mar 5, 2025

m-lord-renkse Mar 5, 2025

MartinquaXD Mar 5, 2025

sunce86 Mar 5, 2025

MartinquaXD Mar 5, 2025

mstrug Mar 5, 2025

MartinquaXD Mar 6, 2025

mstrug Mar 5, 2025

MartinquaXD Mar 6, 2025

mstrug Mar 5, 2025

squadgazzz Mar 6, 2025 •

edited

Loading

squadgazzz Mar 6, 2025

MartinquaXD Mar 7, 2025

squadgazzz Mar 7, 2025 •

edited

Loading

MartinquaXD Mar 7, 2025

squadgazzz Mar 7, 2025

		// Receiver is flashloan wrapper, so that borrowed funds can be returned to the lender
		receiver: Some(onchain.contracts().flashloan_wrapper.address()),

		// Receiver is always the settlement contract, so driver will have to manually send funds to
		// solver wrapper (flashloan borrower)

Single flashloan encoding #3279

Single flashloan encoding #3279

Conversation

sunce86 commented Feb 13, 2025 • edited by MartinquaXD Loading

Description

Changes

Things not addressed by this PR

How to test

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sunce86 Feb 25, 2025 • edited Loading

Choose a reason for hiding this comment

MartinquaXD Feb 26, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

squadgazzz Mar 6, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

squadgazzz Mar 7, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sunce86 commented Feb 13, 2025 •

edited by MartinquaXD

Loading

sunce86 Feb 25, 2025 •

edited

Loading

MartinquaXD Feb 26, 2025 •

edited

Loading

squadgazzz Mar 6, 2025 •

edited

Loading

squadgazzz Mar 7, 2025 •

edited

Loading