⚡️ STL Optimizations

[Vectorized]

Description

Changes:

• Assembly reverts for ⚡️ Optimize STL further, do require in asm #157.
• Optimized the success checks.
• Removed redundant mstore(0x60, 0)s.

You can try copypasta the revert assembly into something like Remix and verify it.

Checklist

Ensure you completed all of the steps below before submitting your pull request:

• Ran forge snapshot?
• Ran npm run lint?
• Ran forge test?

Pull requests with an incomplete checklist will be thrown out.

[z0r0z]

I like it. If we can't get t11s to accept custom errors, let's use assemblored reverts.

[transmissions11]

I like it. If we can't get t11s to accept custom errors, let's use assemblored reverts.

lol i opened the issue to do reverts in asm #157

[z0r0z]

ok ok ok

[Vectorized]

The test fails for ERC1155Test::testFailFuzzSafeBatchTransferInsufficientBalance for certain rare fuzz cases.

You can try the following:

function testFailFuzzSafeBatchTransferInsufficientBalanceRareCase() public {
    uint256[] memory ids = new uint256[](2);
    ids[0] = 0;
    ids[1] = 0;
    uint256[] memory mintAmounts = new uint256[](2);
    mintAmounts[0] = type(uint256).max - 2;
    mintAmounts[1] = 2;
    uint256[] memory transferAmounts = new uint256[](2);
    transferAmounts[0] = type(uint256).max - 2;
    transferAmounts[1] = 2;
    testFailFuzzSafeBatchTransferInsufficientBalance(
        address(0xABCD),
        ids, 
        mintAmounts,
        transferAmounts, 
        bytes(""), 
        bytes("")
    );
}

The fix is to simply add a continue line in the fuzz test like so:

for (uint256 i = 0; i < minLength; i++) {
    uint256 id = ids[i];

    uint256 remainingMintAmountForId = type(uint256).max - userMintAmounts[from][id];

    if (mintAmounts[i] == remainingMintAmountForId) continue; // To mitigate the rare case.

    uint256 mintAmount = bound(mintAmounts[i], 0, remainingMintAmountForId);
    uint256 transferAmount = bound(transferAmounts[i], mintAmount + 1, type(uint256).max);

    normalizedIds[i] = id;
    normalizedMintAmounts[i] = mintAmount;
    normalizedTransferAmounts[i] = transferAmount;

    userMintAmounts[from][id] += mintAmount;
}

Lemme know if you want to add this line into the fuzz test, although it is not related to the PR.

[Vectorized]

Note that we need 4 mstores for the reverts, as the scratch space may be dirty from those special tricks.

[transmissions11]

wait how is it safe to remove the gt(returndatasize(), 31)) check

[Vectorized]

Before the call, we store the 4-byte selector at 0x00.

// Equivalent to `mstore(0x00, 0x0000000000000000000000000000000000000000000000000000000023b872dd)`
mstore(0x00, 0x23b872dd)

Since the selector 0x23b872dd does not end with byte 0x01
(e.g. 0x11223301)…

the only way for the memory slot at 0x00 to be equal to 1 is for the call to write 0x0000000000000000000000000000000000000000000000000000000000000001 to 0x00, which gives a returndatasize() that is at least 32.

This can be applied for the other selectors too (0xa9059cbb, 0x095ea7b3).

[transmissions11]

oh what the fuck this is genius

[sambacha]

This is unreadable