Connecting test harness

[hoytech]

Hi! I got the harness working. Unfortunately the test suite isn't passing yet. I'm trying to track it down. Here's my work so far: master...hoytech:rust-negentropy:harness

I added the harness and started comparing the behaviour of the C++ versus the Rust impl. The first thing that was causing me problems was that you used a vector for pending_output, but I used a deque. I just made the Rust one use a deque instead, so I can compare more directly for now. We can roll that back later if not necessary.

The first difference I noticed was in the splitRange function. The C++ function has an iterator curr whereas the rust one doesn't. I added some tracing lines (visible in the above commit) and this is the output I get from rust:

TRACE A 0
TRACE A 1677970534
TRACE A 1677970534
TRACE A 1677970534
TRACE A 1677970534
TRACE A 1677970534
TRACE A 1677970534
TRACE A 1677970534
TRACE A 1677970534
TRACE A 1677970534
TRACE A 1677970534
TRACE A 1677970534
TRACE A 1677970534
TRACE A 1677970534
TRACE A 1677970534
TRACE A 1677970534
TRACE B 18446744073709551615

But this is what I got from the C++:

TRACE A 1677970550
TRACE A 1677970563
TRACE A 1677970577
TRACE A 1677970595
TRACE A 1677970609
TRACE A 1677970622
TRACE A 1677970634
TRACE A 1677970648
TRACE A 1677970657
TRACE A 1677970668
TRACE A 1677970677
TRACE A 1677970687
TRACE A 1677970691
TRACE A 1677970713
TRACE A 1677970725
TRACE A 18446744073709551615
TRACE B 18446744073709551615

So something isn't advancing through the ranges properly in the rust version. You can reproduce this by checking out the two repos side-by-side (make sure to pull latest master in negentropy repo) and running the following command inside the negentropy/test/ directory:

DEBUG=1 RUST_BACKTRACE=1 RECS=200 P1=1 P2=0 P3=0 perl fuzz.pl rust rust

[yukibtc]

Thanks! I replaced the VecDeque with a classical Vec because it's faster. I adapted the rest of the code to use the it instead of the VecDeque but probably something goes wrong. I'll try to revert it and search any other issues.

[hoytech]

Got it working! See my PR here: #2

I'm sure some of the code is not idiomatic Rust -- sorry about that. But the test suite passes using all language combinations: Rust/C++, C++/Rust, Rust/JS, etc. I confirmed the wire output of the Rust version is identical to the C++.

You might be able to convert pending_outputs back to a Vec. If you try that, let me know and I'll make sure it passes the test suite again. If so, I might make a similar change to the C++ version.

[yukibtc]

Thanks for the fix!

Days ago I added a unit test that check if the output of the reconciliation was the same that I received from C++ (and effectively it was the same), but using the harness test you found an issue...
Would be awesome to have another automated test to check if the C++ and Rust version match. Have you any idea how to do that? Or how to improve the current unit test?

https://github.com/yukibtc/rust-negentropy/blob/master/negentropy/src/lib.rs#L767

[yukibtc]

The code seems fast as before also with the VecDeque: ~100 ms per 1 million ids, ~40 ms without the hex encoding.

[hoytech]

Glad to hear! I agree it would be good to incorporate the reference test suite into the rust library, but it's quite a lot of extra stuff that would need to get pulled into your repo. And you'd need to compile the C++ code too. Maybe there is some more minimal integration we can do, I will think about that.