Fix masking of irrelevant bits in address groups.

[morcos]

If I'm reading this right, I believe the intent is to keep the high order nBits and mask away the others with 1's. But before this fix it was keeping the high order (8 - nBits). I don't think there was any effect, because the only time this was used was in the case of he.net /36 IPv6 groups so nBits was 4.

[sipa]

Nice catch. Changing the | into an & seems simpler, though.

[dcousens]

Agreed, the & would be much clearer here and is more typical for a masking operation of this type.

[morcos]

I thought about & but I wasn't confident it was ok to change the previous behavior.
I also couldn't see how to make it as short?
vchRet.push_back(GetByte(15 - nStartByte) & (((1 << nBits) - 1) << (8 - nBits)));
or
vchRet.push_back(GetByte(15 - nStartByte) & ~((1 << (8-nBits)) - 1));

[sipa]

Good point, we shouldn't change the existing behaviour.

ACK

[jonasschnelli]

utACK.
Would be nice if there would be test coverage in netbase_tests.cpp for this.

[laanwj]

utACK.
There was a similar error in the netmask parsing.
Agree @jonasschnelli that unit tests for this function would be nice.

[morcos]

Ok writing unit test. I have a question about IsLocal(), since those addresses are also !IsRoutable() they will be lumped in the NET_UNROUTABLE group instead of having their own group. The check for IsLocal() on line 932 is meaningless. Is that expected? Changing that would obviously change existing behavior.

[laanwj]

Yes, that check seems to be redundant. But let's leave removing it for another pull.

Thanks for adding tests.