Update arkworks to 0.4.2 for all crates

[chiro-hiro]

This PR just migrate all crates to ark-ec@0.4.2 and ark-ff@0.4.2. I will upgrade the code first then fix all the issue regarding to test and CI/CD.

[chiro-hiro]

Upgrade the test suite and rewrite test cases.

ark-test-curves = "0.4.2"
ark-algebra-test-templates = "0.4.2"
ark-serialize="0.4.2"

[dannywillems]

Hey, thanks for the contribution!
I try to build on my machine but I have errors regarding the trait F in the Poseidon. Can you have a look?
I opened #1358 for the CI. We will fix the run here later.

[dannywillems]

It also requires to update ark-ff and ark-ec in the other libraries, like poseidon, visu, etc. Otherwise, we have conflicting versions (0.3.0 and 0.4.2, see Cargo.lock).

[chiro-hiro]

Hi @dannywillems,

I think update the whole proof-system to 0.4.2 would be a big task. So I tried to break down it to multiple PRs. If you think this PR good enough I will open a new PR that update 0.4.2 for all crates.

[dannywillems]

Hi @chiro-hiro. Unfortunately, I cannot test this PR without having the whole repository bumped to 0.4.2. However, by having a manual look, it looks good!
That would be very helpful if you can bump up to 0.4.2 for the others. Thanks!

[chiro-hiro]

Hi @dannywillems,

I'm working on it, going to upgrade all other crates to make sure that everything will work fine with arkworks 0.4.2

[chiro-hiro]

Hi @dannywillems,

I have a problem with the migration. In the latest update we have poly-commitment/src/pairing_proof.rs, there are some test on curve bn254 this curve still using arkworks 0.4.0. I'm stuck here and still need to figure out what should we do? here are some possible options:

• Implement bn254 in proof-systems/curves with arkworks 0.4.2
• Waiting for arkworks update their curves
• PR to arkworks

Which one do you prefer?

[chiro-hiro]

Hi @dannywillems, any update on this?

[mrmr1993]

I'm concerned that this upgrade seems to significantly reduce the performance of the proof system. I think we should hold off merging until we know what's causing the performance regression, and how to fix it.

[chiro-hiro]

Hi @mrmr1993, thanks for your reply. We definitely need to inspect the root cause of this performance issue. Could you share with me your opinion on bn254 issue and what should we do next?.

Btw, I will keep maintain this branch to minimize the conflict with the upstream.

[mrmr1993]

Could you share with me your opinion on bn254 issue and what should we do next?

No opinion yet, we'll want to figure out the performance issues before thinking about how/whether to continue with the upgrade.

[volhovm]

An important PR, but converting it to draft before we resolve the performance issue/are ready to proceed.

[chiro-hiro]

Catch up with upstream:master one last remaining issue is trait bound of PolyComm<G>.

    Blocking waiting for file lock on build directory
   Compiling kimchi v0.1.0 (/home/chiro/Git/proof-systems/kimchi)
error[E0369]: cannot subtract `&PolyComm<G>` from `&PolyComm<G>`
   --> kimchi/src/verifier.rs:927:25
    |
927 |         &chunked_f_comm - &chunked_t_comm.scale(zeta_to_domain_size - G::ScalarField::one())
    |         --------------- ^ ------------------------------------------------------------------ &PolyComm<G>
    |         |
    |         &PolyComm<G>
    |
help: consider further restricting this bound
    |
765 |     G: KimchiCurve + ark_ec::Group,
    |                    +++++++++++++++

For more information about this error, try `rustc --explain E0369`.
error: could not compile `kimchi` (lib) due to previous error

[volhovm]

Closed in favour of:

• Upgrade to arkworks 0.4.2 (mina compatible / p-s develop) #2474
• Update arkworks to 0.4.2 (master) #2588