nomad / cola Goto Github PK

Hi! I'm taking a look at cola. Your g-tree data structure looks to be several times faster than the b-tree I have in diamond types - which is super impressive! I'm considering using it.

I just ran the test suite for cola and got a failure:

$ cargo test
...
     Running tests/insertions.rs (target/debug/deps/insertions-53609808c6069dab)

running 5 tests
test conflicting_insertions_2 ... ok
test conflicting_insertions ... ok
test woot_figure_2 ... ok
test random_edits ... FAILED
test random_insertions ... ok

failures:

---- random_edits stdout ----
seed: 2497429969382240968
thread 'random_edits' panicked at /home/seph/3rdparty/cola/src/gtree.rs:1266:9:
assertion `left == right` failed
  left: 547
 right: 549
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


failures:
    random_edits

test result: FAILED. 4 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out; finished in 16.21s

But I can't reproduce the failure. I assume each testing run is using rng with a random seed, and the seed is discarded when the test fails.

Just browsed through the code and saw something:
You are using usize to encode and decode the length of a byte array. The number of bytes differs on different architectures, so decoding would fail on a 32 bit system if it was sent from a 64 bit system.

Why don't you just let bincode encode and decode the whole tuple of fields? It saves code and should be safer. You also need to think about security here, as it is supposed to be sent over the wire.