Invalid witness generated on Android about witnesscalc HOT 11 CLOSED

@nasko25 We have applied bug fixes from circom 9 days ago. You can test if this fixes your issue. Also check if USE_ASM flag in cmake changes result.
In general we haven't tested executables on Android (but should work as you describe), we are using compiled libraries.
Also I've just merged feature/arm-asm branch into main.

from witnesscalc.

I tried the latest changes and setting the USE_ASM flag, but the witness is still not correct.
I also tried generating a witness from the host, and it is not valid either.

from witnesscalc.

@nasko25 we will update this witness calc to support cpp files generated by the latest circom in about a week, maybe it will fix your issue.

from witnesscalc.

Thanks, I will wait for the update. I already tried several previous releases of circom to generate a cpp file (most notably v2.0.7 and v2.0.5), and none of them produced a correct witness. Do you think there is a specific version of circom which might generate cpp files that should work?

from witnesscalc.

@OBrezhniev I tried the latest changes on an x86_64 host machine and the witness is still not correct. Is there anything else I can try?

from witnesscalc.

I tried the feature/update-authV2 branch with a cpp file from the latest circom, but the proof calculated with the generated witness file is still not valid. Circom's witness calculator still calculates a correct witness.

I also tried a different circuit with these versions of witnesscalc, circom, and snarkjs and it produced a valid proof. So something goes wrong only for the circuit I linked above.

from witnesscalc.

I also noticed that the logs in this circuit are not printed when generating a witness with witnesscalc. But if I add a log to the very beginning of this circuit (in the main component) it is printed. So something with the use of multiple components probably goes wrong.

from witnesscalc.

Hi @nasko25 , Did your issue resolve?

from witnesscalc.

Hi @sanketsaagar

Unfortunately, it is not yet resolved. Generating a witness with witnesscalc for the circuit linked above still results in an invalid proof.
However, I think I was able to isolate the issue further.
This circuit also generates an invalid witness:

pragma circom 2.0.0;

template Add2(n) {
    signal input a[n];
    signal input b[n];
    signal output out[n];

    log("add");

    for (var k=0; k<n; k++) {
        out[k] <== a[k] + b[k];
    }
}

template Main() {
    signal input a[2];
    signal input b[2];
    signal output out[2];

    component add = Add2(2);

    for (var i = 0; i < 2; i++) {
        add.a[i] <== a[i];
        add.b[i] <== b[i];
    }

    for (var j = 0; j < 2; j++) {
        log("out[", j, "] = ", add.out[j]);
        out[j] <== add.out[j];
    }

}

component main = Main();

And I believe the source of the issue is an implementation of assert(expr) somewhere withing witnesscalc that does not execute its parameter.

from witnesscalc.

@nasko25 try generating witnessccalc with this fix in circom compiler: https://github.com/iden3/circom/tree/fix-counter-decrease-in-assert
It should fix (at least for me it works with your code), but in general check that you don't have NDEBUG defined somewhere, which disables asserts, as asserts are used instead of exceptions in the code, which is not right way of exiting on failed condition checks.

from witnesscalc.

Thank you. This version of circom fixes the issue.

I did check whether NDEBUG was defined anywhere while I was debugging it, but I could not find a manual definition anywhere, so I assumed it was automatically enabled in the cmake/compiler's release mode.

from witnesscalc.