Open
Description
There is a problem with the example of solidity-compatible hashes. The way it's currently done is that it calls keccak_u256s_be_inputs to hash a span composed of u256 words - however this function expects only full u256 words!
Consider the following:
use debug::PrintTrait;
use keccak::keccak_u256s_be_inputs;
#[test]
fn test_full_world() {
let input_data: Span<u256> = array![
0xffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff
]
.span();
let hashed = keccak_u256s_be_inputs(input_data);
// Split the hashed value into two 128-bit segments
let low: u128 = hashed.low;
let high: u128 = hashed.high;
// Reverse each 128-bit segment
let reversed_low = integer::u128_byte_reverse(low);
let reversed_high = integer::u128_byte_reverse(high);
// Reverse merge the reversed segments back into a u256 value
let compatible_hash = u256 { low: reversed_high, high: reversed_low };
assert(
compatible_hash == 0xa9c584056064687e149968cbab758a3376d22aedc6a55823d1b3ecbee81b8fb9,
'wrong hash'
)
}
#[test]
fn test_partial_word() {
let input_data: Span<u256> = array![0xAA].span();
let hashed = keccak_u256s_be_inputs(input_data);
// Split the hashed value into two 128-bit segments
let low: u128 = hashed.low;
let high: u128 = hashed.high;
// Reverse each 128-bit segment
let reversed_low = integer::u128_byte_reverse(low);
let reversed_high = integer::u128_byte_reverse(high);
// Reverse merge the reversed segments back into a u256 value
let compatible_hash = u256 { low: reversed_high, high: reversed_low };
assert(
compatible_hash == 0xdb81b4d58595fbbbb592d3661a34cdca14d7ab379441400cbfa1b78bc447c365,
'wrong hash'
)
}If you run the tests, you'll notice that it works for the full u256 word (with all bits set) but not for the word that is not full.
This is a bit tricky, but basically what you should do instead is:
- Split the u256 words into u64 chunks
- If you're hashing multiple words, you need to pack the partially-filled last u64 word of the input
nwith the first u64 word of the inputn+1 - This requires counting how many bytes each word actually uses.
See how it's implemented in Alexandria for bytes inputs, or Herodotus for u64-words inputs.
Metadata
Metadata
Assignees
Labels
No labels