Add Public Key Encryption

[auryn-macmillan]

This PR resolves the comments from unmerged PR #31.

I'll also make a PR back to the OPs fork so that these changes could be incorporated into the original PR.

[enricobottazzi]

On top of the in-line comments, can you please add further indication in the README on how to generate the parameters for the public key encryption case? I would also ask you to add a gtihub action for testing the new circuit.

As I run

python3 scripts/circuit_sk.py -n 4096 -qis '[                                      
    27424203952895201,
    27424203952895203
]' -t 65537

I first run into this error (you can also see it from the GitHub action)


  File "/Users/enricobottazzi/Developer/GitHub/greco/scripts/circuit_sk.py", line 478, in <module>
    main(args)
  File "/Users/enricobottazzi/Developer/GitHub/greco/scripts/circuit_sk.py", line 30, in main
    bfv_crt = BFVCrt(crt_moduli, n, t, discrete_gaussian)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/enricobottazzi/Developer/GitHub/greco/scripts/venv/lib/python3.11/site-packages/bfv/bfv.py", line 342, in __init__
    self.bfv_q = BFV(RLWE(n, crt_moduli.q, t, discrete_gauss))
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/enricobottazzi/Developer/GitHub/greco/scripts/venv/lib/python3.11/site-packages/bfv/bfv.py", line 32, in __init__
    assert np.gcd(q, t) == 1, "modulus q and t must be coprime"
           ^^^^^^^^^^^^
OverflowError: Python int too large to convert to C long

I think this might be caused by an updated version of np which is a dependency on my bfv-py repo. If you confirm that you are facing the same error, I’ll update the requirements.txt in bfv-py to pin to the correct (old) version of np.

In general it seems that there are some small errors in the naming of the variables or in the related comments. Some errors might be carried from the original PR.

The tests seem to be passing correctly based on your generated data. But I'm quite outside the loop with Greco, so I will leave it to @0xjei to provide a more thorough review on the PR and the adherence of the protocol to what has been described in the paper

[auryn-macmillan]

Yes, I can confirm I see the same numpy error. For the time being, I updated the requirements.txt in this repo. But it would be better for it to be updated in bfv.py.

I also had to make these changes in bfv.py

In polynomial.py, I had to add an implementation for len

    def __len__(self) -> int:
        return len(self.coefficients)

In bfv.py, I changed line 442 to this.

    ct_1 = pk1_u.coefficients + e1

[0xjei]

hey @auryn-macmillan and @fhedude, Thanks for the update. I'm still getting the same error as Enrico mentioned above.

I first run into this error (you can also see it from the GitHub action)


  File "/Users/enricobottazzi/Developer/GitHub/greco/scripts/circuit_sk.py", line 478, in <module>
    main(args)
  File "/Users/enricobottazzi/Developer/GitHub/greco/scripts/circuit_sk.py", line 30, in main
    bfv_crt = BFVCrt(crt_moduli, n, t, discrete_gaussian)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/enricobottazzi/Developer/GitHub/greco/scripts/venv/lib/python3.11/site-packages/bfv/bfv.py", line 342, in __init__
    self.bfv_q = BFV(RLWE(n, crt_moduli.q, t, discrete_gauss))
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/enricobottazzi/Developer/GitHub/greco/scripts/venv/lib/python3.11/site-packages/bfv/bfv.py", line 32, in __init__
    assert np.gcd(q, t) == 1, "modulus q and t must be coprime"
           ^^^^^^^^^^^^
OverflowError: Python int too large to convert to C long

I'm just checking in to see if you need any help or hints on this and how I can be of service. Thanks!

[auryn-macmillan]

@0xjei, do you still get that error after doing pip3 install -r requirements.txt in scripts/?

That error is caused by an incompatible version of Numpy.

There is also a subsequent error that requires the above changes to BFV.py.

I think we should perhaps have Greco depend on an instance of BFV.py in the PSE GitHub org, rather than the one in @enricobottazzi profile. Just so we're not dependent on Enrico to merge fixes.

[0xjei]

@0xjei, do you still get that error after doing pip3 install -r requirements.txt in scripts/?

yes, I'm still getting the same error after running the installation on a fresh venv.

I think we should perhaps have Greco depend on an instance of BFV.py in the PSE GitHub org, rather than the one in @enricobottazzi profile. Just so we're not dependent on Enrico to merge fixes.

sure, we should move whatever Greco is dependent on under the same organization. I will get back to you on this asap.

[auryn-macmillan]

@0xjei I've just updated the requirements.txt to point to an updated version of bfv.py in the Gnosis Guild github org. Give it a try again now, the errors should be resolved.

[0xjei]

@0xjei I've just updated the requirements.txt to point to an updated version of bfv.py in the Gnosis Guild github org. Give it a try again now, the errors should be resolved.

thank you @auryn-macmillan - I've got it working locally now, using cargo build && cargo test --release --nocapture. But I'm still getting errors on the CI. I've traced it back to the --feature bench in the cargo test command. Could you please make sure you get rid of these errors when running the benchmarks too? This should be the last 10% that's causing problems.

[Vishalkulkarni45]

Hi @auryn-macmillan and @0xjei, I've implemented the changes suggested by Enrico in my PR. You can check it out here. I hope this is helpful, and please let me know if there's anything else I can assist with

[0xjei]

Hi @auryn-macmillan and @0xjei, I've implemented the changes suggested by Enrico in my PR. You can check it out here. I hope this is helpful, and please let me know if there's anything else I can assist with

Hey @Vishalkulkarni45 thx - can you just push your new changes back on your PR about public key encryption to ease the review process? tysm

[Vishalkulkarni45]

Hi! @0xjei i had already made the PR of it #32 but enrico said it's not relevant because SK and PK encryption have almost same algorithm or may be its because of my bad python skills 😅 😁

[auryn-macmillan]

@Vishalkulkarni45 and @0xjei, we've shifted our efforts to supporting fhe.rs rather than bfv.py over in our fork of Greco.

Could you please make sure you get rid of these errors when running the benchmarks too? This should be the last 10% that's causing problems.

Will try to address this in our fork as well. cc @fhedude