w&b error correction interpolation algorithm implemented

[changxu2]

Update with current miniviff
Not ready for merge

[amiller]

This shouldn't be necessary if strip_trailing_zeros was called earlier

[amiller]

Slight terminology correct - this is Reed Solomon not secret share encoding. So k is not the number of shared secrets, but it could be called the number of code symbols.
Polinomial -> Polynomial

[changxu2]

Fixed!

[codecov-io]

Codecov Report

Merging #4 into miniviff will increase coverage by 4.95641%.
The diff coverage is 84.33735%.

@@                Coverage Diff                 @@
##           miniviff          #4         +/-   ##
==================================================
+ Coverage   58.7886%   63.74502%   +4.95641%     
==================================================
  Files            11          13          +2     
  Lines           842        1004        +162     
  Branches        131         182         +51     
==================================================
+ Hits            495         640        +145     
- Misses          330         336          +6     
- Partials         17          28         +11

[sbellem]

We do not need to explicitly invoke the test function (test_decoding) as pytest takes care of finding and executing the test functions and classes.

So, when you run

pytest test_wb_interpolate.py

all functions starting with test_ and classes starting with Test will be executed.

More information about this behavior can be found under https://docs.pytest.org/en/latest/goodpractices.html#conventions-for-python-test-discovery.

[amiller]

We should import this simple_router from somewhere rather than copy-pasting it.

[sbellem]

We could probably put it in a pytest fixture.

Right now it is defined in test-asyncio-simplerouter.py I think. We cannot import things from it I think because the module name contains "dashes/hyphens". A workaround is to use something like:

from importlib import  import_module                                                     
simple_router = getattr(
    import_module('honeybadgermpc.test-asyncio-simplerouter'), 'simple_router')

or in two steps, if you need more things from it

from importlib import  import_module                                                     
async_simple_router_test_module = import_module('honeybadgermpc.test-asyncio-simplerouter')
simple_router = async_simple_router_test_module.simple_router
Share = async_simple_router_test_module.Share
# etc

also note that the above will work only if the Runtime class is fixed, e.g.:

diff --git a/honeybadgermpc/test-asyncio-simplerouter.py b/honeybadgermpc/test-asyncio-simplerouter.py
index 90e64d5..14c8dee 100644
--- a/honeybadgermpc/test-asyncio-simplerouter.py
+++ b/honeybadgermpc/test-asyncio-simplerouter.py
@@ -51,7 +51,7 @@ class Runtime():
 
     async def _run(self):
         while True:
-            await   # noqa TODO fix: await ?; e.g.: await asyncio.sleep(1)
+            await asyncio.sleep(1)  # noqa TODO fix: await ?; e.g.: await asyncio.sleep(1)
 
     def createshare(self, val):
         s = Share(self)

In any case, I think we should spend the time to put such things in fixtures.

[sbellem]

I would suggest the following:

1. Move the test module test_batch_reconstruction.py under the tests dir.
2. Add a simple_router fixture in tests/conftest.py, e.g.:

# tests/conftest.py
import asyncio
import random

from pytest import fixture


@fixture
def simple_router():

    def _simple_router(N):
        """
        Builds a set of connected channels
        @return (receives, sends)
        """
        # Create a mailbox for each party
        mbox = [asyncio.Queue() for _ in range(N)]

        def makeSend(i):
            def _send(j, o):
                # print('SEND %8s [%2d -> %2d]' % (o[0], i, j))
                # random delay
                asyncio.get_event_loop().call_later(
                    random.random()*1, mbox[j].put_nowait, (i, o))
            return _send

        def makeRecv(j):
            async def _recv():
                (i, o) = await mbox[j].get()
                # print('RECV %8s [%2d -> %2d]' % (o[0], i, j))
                return (i, o)
            return _recv

        sends = {}
        receives = {}
        for i in range(N):
            sends[i] = makeSend(i)
            receives[i] = makeRecv(i)
        return (sends, receives)

    return _simple_router

3. Remove the simple_router function definition in tests/test_batch_reconstruction.py and use the simple_router fixture (defined in conftest.py) in the test like so:

def test(simple_router):
    N = 4
    p = 73
    t = 1

    async def _test():
        # Test with simple case: n = 4, t =1
        # After AVSS, poly1 = x + 2, poly2 = 3x + 4, secret1 = 2, secret2 = 4
        # Hard code the shared secret value as input into batch_reconstruction function
        # The final constructed polynomial should be p = 4x + 2
        sends, recvs = simple_router(N)
        towait = []
        for i in range(N):
            # ...

[sbellem]

The above is demonstrated in 19737f7 (and 0768e06).

[amiller]

@changxu2 what is left to do before this can be merged? I think you need to add some integration with an MPC program, maybe online_active.py that uses the batch opening operations and that we can run with custom programs

[amiller]

@changxu2 One thing missing from batch_reconstruction is that if a fault is detected, we should return (along with the solution) a list of the player ids corresponding to any faults. Then we can ignore subsequent messages from them and effectively remove them from future comparisons

[changxu2]

1. For the simple_router issue: Thank you very much @sbellem ! I indeed encoutered the issue with module name containing "dashes", thank you for pointing out the direction for refactoring!
2. Things to do before merging (that I can think of now): a) I will implement the evil nodes number in return b) some clean up work c) fft version of the algorithm d) active_online.py. For a) , b) and d) , I am planing on finishing before merging. For d) @amiller , do you think it would be better to merge first or implement first and then merge altogether?

[amiller]

@changxu2 I think you mean you'll leave c) the fft computation for after merging? If so yes that's fine

[amiller]

What is the status of this @changxu2 it looks like a test is failing due to code style but that is it?