tokencard / contracts Goto Github PK

Loop through and check if token exists by looping through the array first, to save some gas ...

Usually because a string needs to be cast into bytes. It works, but i am just mentioning it..

At the moment we pass multiple token labels to the addTokenBatch function and then use the strings library to parse the individual labels. We should improve this.

function addTokenBatch (address[] tokenIDs, string labels, uint8[] decimals) public onlyController 

In a couple of places the same value is being calculated multiple times instead of storing the result in a variable and re-using it. For example:

validIDs[queryId]

When the addToWhitelist() contains the address of the owner, addToWhitelistConfirm will skip adding the owner to the whitelist, but will fire event WhitelistAddition containing the owner address

Both submitWhitelistAddition and submitWhitelistRemoval have the following check:

require(!submittedWhitelistAddition && !submittedWhitelistRemoval, "whitelist operation has already been submitted");

This should be put into a new modifier, e.g. noWhitelistOperationInProgress for better maintainability and code re-use.

Currently the Oracle returns a rate of 0 if the rate of the specified token is not set. How would we go about handling the transfer of a token that doesn't have its rate set in the Oracle? Do we need this functionality?

At the moment we store a map of supported tokens and also an array of supported tokens:

 mapping (address => Erc20Token) public tokens;
 address[] private _contractAddresses;

This is quite redundant and we can most likely find a better solution.

Where is the "count" set to 0, I can't see it ...

setSpendLimitCancel sets it to 0, setSpendLimitConfirm does not

Currently the oracle doesn't handle uint overflow when calculating the rate, for most regular values this doesn't matter but it will be a problem in certain cases.

e.g. oracle.token.eth which resolves to the oracle address so that we can upgrade/change the oracle contract in the future.

In the Wallet we use external a lot more than public, my understand is that there are gas effeciencies with External, need help here ...

function spendAvailable() external view returns (uint) {
        if (now > _spendLimitDay + 24 hours) {
            return spendLimit;
        } else {
            return _spendAvailable;
        }
    }

spendLimit might not actually reflect _spendAvailable.
_spendAvailable should always be returned.

Subsequently (and perhaps unclear to me),

    /// @dev Confirm pending set daily limit operation.
    function confirmSpendLimit() external onlyController {
        // Require that the operation has been submitted.
        require(submittedSpendLimit, "spend limit has not been submitted");
        // Modify spend limit based on the pending value.
        modifySpendLimit(pendingSpendLimit);
        // Emit the set limit event.
        emit SetSpendLimit(msg.sender, pendingSpendLimit);
        // Reset the submission flag.
        submittedSpendLimit = false;
        // Reset pending daily limit.
        pendingSpendLimit = 0;
    }

    /// @dev Update available spend limit based on the daily reset.
    function updateSpendAvailable() internal {
        if (now > _spendLimitDay + 24 hours) {
            // Advance the current day by how many days have passed.
            uint extraDays = (now - _spendLimitDay) / 24 hours;
            _spendLimitDay += extraDays * 24 hours;
            // Set the available limit to the current spend limit.
            _spendAvailable = spendLimit;
        }
    }

    /// @dev Modify the spend limit and spend available based on the provided value.
    /// @dev _amount is the daily limit amount in wei.
    function modifySpendLimit(uint _amount) private {
        // Account for the spend limit daily reset.
        updateSpendAvailable();
        // Set the daily limit to the provided amount.
        spendLimit = _amount;
        // Lower the available limit if it's higher than the new daily limit.
        if (_spendAvailable > spendLimit) {
            _spendAvailable = spendLimit;
        }
    }

There isn't a 24 hour wait (or remainder of hours to wait before resetting the limit) to stop confirmSpendLimit() from updating the limit and it does not set _spendLimitDay to now

Last controller can remove himself using removeController() leaving contract without any controllers and any means to add a new controller.

In the wallet contract I used the convention of using a leading underscore _foo for:

1. Private variables
2. Private methods
3. Function parameters

It would be good to use the same approach in the oracle, for example instead of:

function addTokenBatch (address[] tokenIDs, string labels, uint8[] decimals) public onlyController

use:

function addTokenBatch (address[] _tokenIDs, string _labels, uint8[] _decimals) public onlyController 

Currently the proof code is commented out / incomplete, we should either finish implementing it or consider ignoring the proof if it doesn't work.

in acceptOwnership() there's require(_newOwner != address(0), "owner cannot be set to 0x0000000000000000000000000000000000000000");. Shouldn't this line also exist in transferOwnership() for address _account?

Would it be a good idea to add a freeze() function that can be called only by the owner which stops all wallet functionality, and a corresponding unfreeze() function which can be called only by the controller and in turn resumes the wallet functionality?

This would be akin to the freeze operation that popular challenger banks offer. Where the user can just have a panic button and then they have to contact us to unfreeze the wallet. Or some other variation of this.

@mischat

We have no cross-checks of which addresses are in the _submittedRemoval and _submittedAddition. So if the same address is in the both list, depending on the order in which the controller accepts, the address might end up being whitelisted or not.
Probably we should not allow adding of addresses while removal is in process and vice versa.

/// @dev Returns the amount of an asset owned by the contract.
    /// @param _asset address of an ERC20 token or 0x0 for ether.
    /// @return balance associated with the wallet address in wei.
    function balance(address _asset) external view returns (uint) {
        if (_asset != 0x0) {
            return ERC20(_asset).balanceOf(this);
        } else {
            return address(this).balance;
        }
    }

what happens if this is called with an address asset that's not ETH or ERC20? What would return ERC20(_asset).balanceOf(this); do? I'd assume it would just fall over?

is there a nicer way to detect if an asset is an ERC20 token? if possible, it would be nice if have this in a if, else if, else fail kind of statement

Blockchain logs are mostly used by other software, logging long sentences in the oracle contract should most likely be changed to something more useful.

emit LogNewOraclizeQuery("Oraclize query was sent, standing by for the answer...");

We should also have a require somewhere ...

The _topupLimitDay member field should get set to now in the constructor - at the moment it is 0, which means the top up day starts at midnight, unlike the spend limit which will start at the time the contract is deployed. These two "limit days" should be in sync.

Currently when the constructor receives a 0x0 address in the constructor it sets the address of the resolver to a hardcoded value:

        if (_oraclizeAddrResolver == 0x0) {
          _oraclizeAddrResolver = 0x6f485C8BF6fc43eA212E93BBF8ce046C7f1cb475;
        }

I think that this is perhaps unnecessary and we can remove it.

transaction returns false and thus reverted in the callback

Why doesn’t the Controllable contract extend the controller interface? Since it’s designed to implement it.

Also, interface could be renamed to IController or something similar to make it clear what it is and to make the above implementation possible without a name clash.

We could repeat this process for other parts of the code where we are using an interface, e.g. Resolver

names of public views pendingRemoval and pendingAddition do not reflect what is pending. Probably should be named pendingWhitelistRemoval, pendingWhitelistAddition

should probably do a length check of _encoded before

if (keccak256(_encoded[length - 2]) == keccak256("=")) {
            length -= 2;
        } else if (keccak256(_encoded[length - 1]) == keccak256("=")) {
            length -= 1;
        }

There are two modifiers that check for the same / opposite value:

    modifier tokenSupported(address tokenID) {
        require(tokens[tokenID].supported);
        _;
    }

    modifier tokenNotSupported(address tokenID) {
        require(!tokens[tokenID].supported);
        _;
    }

We can probably merge it into a single modifier.

This looks strange to me ...

It looks like we are deleting the property supported in the ERC20Token Struct ... surely this doesn't make any sense :

/**
    * @dev remove a token from the list of supported ones
    * @param tokenID token contract addresses
    */
    function removeToken(address tokenID) public onlyController tokenSupported(tokenID) {
        delete tokens[tokenID].supported;

        // Check if the address matches up to one token before the last one
        // the tokenSupported() modifier ensures that the token address actually exists.
        // If no match is found in the loop, it means that the last address was the desired one, simply reduce the size by one in any case.
        uint contractAddressesLength = _contractAddresses.length - 1;
        for (uint i=0; i<contractAddressesLength; i++)
            if (_contractAddresses[i] == tokenID) {
                _contractAddresses[i] = _contractAddresses[contractAddressesLength];
                break;
            }
        _contractAddresses.length--;

        emit TokenRemoval(tokenID);
    }```

Some comments are inconsistent or missing, should standardize the comment style and use it everywhere.

Also we should remove the commented out code from the contract:

    function __callback(bytes32 queryId, string result, bytes proof) public onlyOraclize {

        require(tokens[validIDs[queryId]].supported);//must be a valid token.

        /* if ((proof.length > 0) && (nativeProof_verify(result, proof, cc_pubkey))) {
          TKNETH = parseInt(result, 6);
          delete validIDs[queryId];
          last_update_timestamp = now;
        } */

        uint rate = parseInt(result, 18);

        tokens[validIDs[queryId]].rate = rate; //transform rate(string) to uint (wei precision)
        emit RateUpdated(validIDs[queryId], rate);
        delete validIDs[queryId]; //remove mapping
    }

Currently the controller can't find out about new whitelist request being submitted by listening on the blockchain events.

And can we get CI to error if there are linting errors ?

And also, can we write up how to get the linter into our IDEs so that we all have the same linter ?

When testing the convert function, it seems to return values with too few zeros:

e.g. When expecting 1650000000000 it returns 1650000000

Currently we don't emit any events in the addTokenBatch method, we should re-use the same event that the addToken method uses:

    function addTokenBatch (address[] tokenIDs, string labels, uint8[] decimals) public onlyController {
        require(tokenIDs.length == decimals.length);

        // Convert strings into the library's 'slice' format.
        strings.slice memory labelSlice = labels.toSlice();
        strings.slice memory delim = ".".toSlice();

        uint numTokenLabels = labelSlice.count(delim) + 1; //the number of labels is +1 of thenumber of '.' ["t1.t2.t3"] string expected
        require(numTokenLabels == decimals.length);

        for (uint i = 0; i < numTokenLabels; i++) {
            if(!tokens[tokenIDs[i]].supported){
                _contractAddresses.push(tokenIDs[i]); //push token to the array
                tokens[tokenIDs[i]].label = labelSlice.split(delim).toString();//split the string with a '.' delimiter
                tokens[tokenIDs[i]].decimals = decimals[i];
                tokens[tokenIDs[i]].rate = 0; //to be updated later
                tokens[tokenIDs[i]].supported = true;
            }
        }
    }

In order to have a audit log of what happened, setSpendLimitCancel should probably emit an event.

currently initializeWhitelist() does not check if owner is among the whitelisted addresses - this behaviour is different to addToWhitelist()/addToWhitelistConfirm()

At the moment submitWhitelistAddition has the following the code inside it:

   ....
        // Flag operation as initialized if not initialized already.
        if (!initializedWhitelist) {
            initializedWhitelist = true;
        }
        // Emit the submission event.
        emit SubmitWhitelistAddition(_addresses);

This is bad since, if the submission doesn't get confirmed then we've incorrectly flipped this flag to true.

Instead, submitWhitelistAddition and submitWhitelistRemoval should both check to see if initializedWhitelist is true. If not then these methods should throw.

This ensures that we will always have to call initializeWhitelist first before doing any further operations to the whitelist.

currently, setSpendLimit() does not emit any events. It would be useful for the controller to be able to pick up those events in order to know what needs to be confirmed/rejected

Should we support or not 0 decimal tokens?

https://github.com/tokencard/contracts/blob/oracle%2Bissues/contracts/oracle.sol#L240

Make sure that people don’t add 0x0 to whitelist ever ...

You can call initializeWhitelist without any addresses. This prevents further initialization of the whitelist.

Conceptually I think this could be considered a bug. It also makes it hard to know whether a whitelist has been initialized: BIS will always return an empty array.

I propose we fail the transaction when this function is called without addresses. Then we'll treat an empty array returned from BIS as if the whitelist was not initialized.

I'm not sure if there's anything that can be done about this but attempting to transfer a non-ERC20 token to a non-whitelisted address would burn the users spend available but also create a Token{} in the Oracle contract.

The attacker/user is probably an idiot for doing this... but i hope this would also open up the possibilities of bugs using non-ERC20 tokens.

Why doesn’t the Controllable contract extend the controller interface? Since it’s designed to implement it.

Also, interface could be renamed to IController or something similar to make it clear what it is and to make the above implementation possible without a name clash.

initialization is disabled only after setSpendLimitConfirm is called by the controller. Is this intended?
IMHO _initializedSpendLimit should be set to true already in setSpendLimit

We should be using “Control” to have a set of addresses for the Controller …

whilst the whitelist equivalents arent...