ark-r1cs-std interpolate_and_evaluate failing for n>11

[arnaucube]

In the r1cs-std lib, using the poly/evaluations/univariate:

Let $n$ be such that the evaluation domain is of size $2^n$. Then, when trying to use EvaluationsVar::interpolate_and_evaluate, it leads to stack overflow when $n>11$, while for $n<=11$ seems to work fine.
A temporary isolated sample to illustrate the issue can be found at: https://github.com/arnaucube/r1cs-interpolate-poly-test/blob/main/src/lib.rs
where different $n$ values can be tried, getting stack overflow for $n>11$.

This $n>11$ is needed for interpolating and evaluating at the challenge the polynomials from the $W_{i+1}.W,~ W_{i+1}.E$ vectors in the DeciderEthCircuit.

[winderica]

This PR should fix this issue :)

[CPerezz]

@winderica I think this issue can be closed right?

[winderica]

I guess no? Although arkworks-rs/r1cs-std#145 has already been merged, we need to depend on the latest r1cs-std to fix this issue, but this will result in some dependency hell... I tried to port to the latest arkworks in my fork several months ago, but at that time I wanted to wait for the release of arkworks 0.5 before submitting a PR.

I think for now we can cherry-pick the changes in arkworks-rs/r1cs-std#145 while preparing for the next release of arkworks at the same time.

[arnaucube]

I was just working on this, made a cherrypick of your commit fixing the poly eval, and porting it to the cherry-pick branch of your fork r1cs-std: winderica/r1cs-std#1
it also adapts some other stuff to make it work with the branch arkworks version (v0.4.0).

I think that with this you (@winderica ) should be able to uncomment the current poly-evals in the various decider's circuits and open a PR in Sonobe with the fix, which would then include the fix you did in arkworks and would make the poly evals work ^^

[CPerezz]

I was just working on this, made a cherrypick of your commit fixing the poly eval, and porting it to the cherry-pick branch of your fork r1cs-std: winderica/r1cs-std#1 it also adapts some other stuff to make it work with the branch arkworks version (v0.4.0).

I think that with this you (@winderica ) should be able to uncomment the current poly-evals in the various decider's circuits and open a PR in Sonobe with the fix, which would then include the fix you did in arkworks and would make the poly evals work ^^

Awesome! @winderica will you make a PR then tagging this such that we can close?

[winderica]

Sure! I am now working on this, but it takes more time than expected because I ran into another issue with r1cs-std. While everything should work correctly in proof generation (i.e., when the mode of cs is SynthesisMode::Prove), the interpolation will panic during trusted setup (i.e., when cs has mode SynthesisMode::Setup), because it internally unwraps .value(), which is None during setup.

[CPerezz]

GL with this madness @winderica <3