hevm `startPrank` overrides msg.sender at all call depth levels, making it inconsistent vs foundry and causing unintended operations

[GalloDaSballo]

Impact

I have written this repo which shows a full repro

https://github.com/Recon-Fuzz/prank-echidna-vs-foundry

Fundamentally, when using startPrank and calling a contract that performs a transferFrom, echidna (using hevm under the hood), will perform the transferFrom with the pranked address instead of the router, causing the transfer to fail

Mitigation

I believe that the prank should only be enforced at the current call level, a nested call should not longer have it's msg.sender altered

[msooseth]

Right, so overrideCaller is set from Nothing to the address when startPrank is called. This also sets resetCaller to False. Then, when stopPrank is called, overrideCaller is set to Nothing, and resetCaler is set to True.

So what you are saying would I think mean that when we do another call, overrideCaller should be reset to Nothing, along with resetCaller being reset to True. Let me make a PR like that, and maybe you can then check if that fixes it. Would that be OK?

[msooseth]

So I have a new branch, fix_prank that may fix this. I'm trying to check if it does now :)

[elopez]

@msooseth I don't think that would be enough, as when you return from the call, overrideCaller should be set back to the address and resetCaller should be back to True. I'm thinking those two options should be moved to the frame rather than being a global VM setting.

EDIT: I think you did just that on the branch, it'll probably work fine now 💯 thanks!

From what I understand it should be like this:

A.foo()
\__startPrank(X)
|__B.bar() /* msg.sender is X */
|    \__C.baz() /* msg.sender is B, as the override only applies on A.foo's frame */
|__B.baz() /* msg.sender is X */
|__stopPrank()
|__B.bar() /* msg.sender is A */
     \__C.baz() /* msg.sender is B */

[elopez]

@GalloDaSballo I opened a draft PR on Echidna that uses @msooseth's hevm branch, if you can try it out and see if it behaves as expected now, that'd be great: https://github.com/crytic/echidna/pull/1331

[GalloDaSballo]

Thank you all, will test it soon

[msooseth]

OK, seems to be fixed with c6c4b44af0137ccfdab8197a7dd1e4ae69114edb. The corresponding echidna is:

https://github.com/msooseth/echidna/pull/1

You can cone that, go into fix_prank branch, then nix shell, then git clone https://github.com/Recon-Fuzz/prank-echidna-vs-foundry then cd prank-echidna-vs-foundry and then echidna . --contract CryticTester --config echidna.yaml and it should all pass.

[GalloDaSballo]

Have tested this release candidate by @elopez : https://github.com/crytic/echidna/actions/runs/12081766889

I ran it for 1 Million tests with no reverts

In contrast see the previous version failing in less than 50k tests