Post-constantinople, we should use extcodehash instead of extcodesize

[chriseth]

Once constantinople is finalized, we should use extcodehash instead of extcodesize to check if the called account exists, because it is cheaper.

[anurag-git]

@chriseth , is it as simple as replacing extcodesize with extcodehash. Please help to produce some details so that I can work on it

[chriseth]

Yes, it is very simple, but I think we should wait until constantinople is finalized.

[axic]

In theory it can be implemented under the constantinople switch, but I'd feel safer committing to this around early October because then the EIPs included in constantinople should be final.

[Agusx1211]

Bump!

This may reduce all calls to external contracts by ~300 GAS, because currently after each call the address is checked using extcodesize, which uses 700 GAS, and extcodehash uses only 400 GAS.

@chriseth , is it as simple as replacing extcodesize with extcodehash. Please help to produce some details so that I can work on it

It should also be checked that the result of extcodehash is not c5d2460186f7233c927e7db2dcc703c0e500b653ca82273b7bfad8045d85a470 (account with no code).

[chriseth]

@Agusx1211 thanks for bringing this up again! Note that the constant c5d2460186f7233c927e7db2dcc703c0e500b653ca82273b7bfad8045d85a470 has a deploy-time cost of more than 8576 gas, so this only amortizes after 20 calls. This should be checked against the runs optimizer parameter.

[chriseth]

If we want to avoid providing the constant literally, we could use keccak256(0, 0), which is an extra 39 gas and 4-5 bytes of code (which costs 1000 gas to deploy).

[axic]

This may reduce all calls to external contracts by ~300 GAS, because currently after each call the address is checked using extcodesize, which uses 700 GAS, and extcodehash uses only 400 GAS.

This was an obvious pricing mistake and it is rectified by EIP-1884 in the upcoming Istanbul update.

[chriseth]

Oh wow, I thought extcodehash is cheaper because the size is not directly stored in the trie. So in the end, both opcodes have the same cost?

[axic]

Good point though that size is not stored, size should cost more :)

But the issue was balance and codehash should cost equal because both are stored in the account.

[chriseth]

So does that mean we can now close this?

[axic]

Unless extcodesize is repriced, it seems that added cost of checking for empty hash makes this change not very useful. I'd opt for closing now.

[thegostep]

Does solidity provide a global for determining if a contract exists or is this assembly technique still best approach?

uint256 codeSize;
assembly { codeSize := extcodesize(0x123...) }
bool exists = codeSize > 0;

[chriseth]

It is the best approach.