TOB-FUEL-14: Inconsistent encoding of vectors in ABI

[iqdecay]

Description

The encoding and decoding of vectors is inconsistent. The decoded values do not match the original values. The following test demonstrates this.

Figure 14.1: Test which demonstrates that encoding and decoding a vector results in two different vectors.

#[test]
fn inconsitent_decoding() -> Result<()> {
    let arg = Token::Vector(vec![Token::U64(42)]);
    println!("Tokens: {:?}", &arg);
    let encoded = ABIEncoder::encode(&vec![arg])?.resolve(0);
    println!("Encoded: {:?}", hex::encode(&encoded));
    let param_type = Vec::<u64>::param_type();
    println!("Type: {:?}", param_type);
    let out: Vec<Token> = ABIDecoder::decode(&[param_type], &encoded)?;
    println!("Decoded: {:?}", out);
    Ok(())
}

Figure 14.2: The output of executing the test form 14.1.

Tokens: Vector([U64(42)])
Encoded: "000000000000001800000000000000010000000000000001000000000000002a"
Type: Vector(U64)
Decoded: [Vector([U64(24), U64(1), U64(1), U64(42)])]

The reason for the difference is that the decode_vector function does not match the encode_vector function. While the encoding adds a pointer, capacity and length of the array, the decoding function interprets them as values.

Figure 14.1: Decoding function for vectors. (fuels-rs/packages/fuels-core/src/codec/abi_decoder.rs#92–100)

fn decode_vector(param_type: &ParamType, bytes: &[u8]) -> Result<DecodeResult> {
  let num_of_elements = ParamType::calculate_num_of_elements(param_type,
  bytes.len())?;
  let (tokens, bytes_read) = Self::decode_multiple(vec![param_type;
  num_of_elements], bytes)?;
      Ok(DecodeResult {
          token: Token::Vector(tokens),
          bytes_read,
      })
  }

Recommendations

Short term, change the decode_vector function such that it expects a pointer, length and capacity.
Long term, add tests which verify that the encoded value always decodes to its original value.

[iqdecay]

So it turns out that this is actually expected behavior.
Vectors can be returned from contracts only thanks to custom bytecode injection, as can be seen in

fuels-rs/packages/fuels-programs/src/call_utils.rs

Line 339 in df21541

// The instructions are different if you want to return data that was on the heap

. This bytecode injection is necessary because otherwise, what the VM returns is only the pointer to the data.
It is therefore necessary that the decoding is different from the encoding.