Recursion PoC

[Brechtpd]

Currently implements recursion like this for n blocks:

• n x SuperCircuits -> n x root circuits -> 1 x aggregation circuit (EVM verifiable)

The best version is bench_root_super_circuit_prover_sdk. bench_root_super_circuit_prover is the older version I made without the sdk and should not be used anymore.

TODO:

• Make a TaikoSuperCircuit (current PR just changes the PSE super circuit, let's make a new circuit and leave the original as is)
• Let's add our own publlic input circuit as the only sub-circuit to TaikoSuperCircuit for now (can leave the rest commented so easy to enable more circuits in the future).
• Currently uses a slightly modified snark verifier at https://github.com/brechtpd/snark-verifier. This is just until it's merged into in the PSE snark verifier.
• I haven't gotten gen_pk to work because of trait issues, can just wait until this gets fixed upstream if hard to fix.
• Make this usable for the next testnet.

[johntaiko]

changes:

• pi_circuit2 lookups the block_table instead of assigns it
  • lookup block_table with pi's block_hash field
  • lookup block_table with pi's parent_hash field
• Taiko witness is merged into Block witness
• TaikoSuperCircuit uses Challenges instead of mock randomness

[Brechtpd]

Looks like the last commit or 2 makes the code not compile anymore with some generics errors (so hard to figure out what exactly is wrong).

[johntaiko]

Looks like the last commit or 2 makes the code not compile anymore with some generics errors (so hard to figure out what exactly is wrong).

Ok, let me get it done

[Brechtpd]

Looks good to me!

[smtmfft]

Looks like the last commit or 2 makes the code not compile anymore with some generics errors (so hard to figure out what exactly is wrong).

Let me see if reference brakes.

[johntaiko]

Looks like a lot of tests are still failing because of the signature problem (half fixed here: #119). If I merge in the master containing that "fix" for the signature problem there's a test failing:

failures:

---- pi_circuit2::pi_circuit_test::test_verify stdout ----
thread 'pi_circuit2::pi_circuit_test::test_verify' panicked at 'assertion failed: `(left == right)`

Diff < left / right > :
<Err(
<    [
<        Lookup in block table(index: 1) is not satisfied in Region 4 ('region 0') at offset 159,
<        Lookup in block table(index: 1) is not satisfied in Region 4 ('region 0') at offset 191,
<    ],
>Ok(
>    (),
 )

', zkevm-circuits/src/pi_circuit2.rs:799:9

fixed

[johntaiko]

Only some minor feedback for the PR itself. But let's make make test-all work for all tests (except the bytecode test failures, those are from the PSE side)! I temporarily ignored some PSE public input tests in #119 (files), but that was a workaround to get things working on the master again. Please find out why they are broken so we can enable them again (even though we won't use that circuit, we should keep all tests working unless there's a good reason for them to stop working).

According to the test intent, these errors in bytecode-circuit are as expected.

zkevm-circuits/zkevm-circuits/src/bytecode_circuit/test.rs

Lines 42 to 58 in c41e6ba

	pub fn test_bytecode_circuit_unrolled<F: Field>(
	k: u32,
	bytecodes: Vec<UnrolledBytecode<F>>,
	success: bool,
	) {
	let circuit = BytecodeCircuit::<F>::new(bytecodes, 2usize.pow(k));
	let prover = MockProver::<F>::run(k, &circuit, Vec::new()).unwrap();
	let result = prover.verify_par();
	if let Err(failures) = &result {
	for failure in failures.iter() {
	error!("{}", failure);
	}
	}
	let error_msg = if success { "valid" } else { "invalid" };
	assert_eq!(result.is_ok(), success, "proof must be {}", error_msg);
	}

If success == false, the test will show the ERROR message.

[2023-06-29T02:06:55Z ERROR zkevm_circuits::bytecode_circuit::test] Constraint 0 ('cur.index == 0') in gate 4 ('Header row') is not satisfied in Region 2 ('assign bytecode') at offset 0
    - Column('Advice', 0 - tag)@0 = 0
    - Column('Advice', 1 - index)@0 = 1
    - Column('Fixed', 0 - BYTECODE_q_enable)@0 = 1
    - Column('Fixed', 2 - BYTECODE_q_last)@0 = 0

[Brechtpd]

I re-enabled the PSE PI tests so fixing those isn't forgotten. I also fixed a small issue with the taiko 2tx_2max_tx test because of a wrong max_txs value.

[johntaiko]

I re-enabled the PSE PI tests so fixing those isn't forgotten. I also fixed a small issue with the taiko 2tx_2max_tx test because of a wrong max_txs value

Yes, finally passed all the tests 🥹