regression: assertion left == right failed for floats

[BoxyUwU]

• https://crater-reports.s3.amazonaws.com/beta-1.81-2/beta-2024-07-26/gh/TheLastBytebender.rust-workshop/log.txt
• https://crater-reports.s3.amazonaws.com/beta-1.81-2/beta-2024-07-26/gh/YogirajT.rustnet/log.txt
• https://crater-reports.s3.amazonaws.com/beta-1.81-2/beta-2024-07-26/reg/lightbird-0.1.10/log.txt

[INFO] [stdout] thread 'tests::test_get_ordebook_skew_by_range_orderbook' panicked at tests/test_orderbook.rs:645:9:
[INFO] [stdout] assertion `left == right` failed
[INFO] [stdout]   left: 0.8855105548523078
[INFO] [stdout]  right: 0.885510554852308

[tgross35]

The first crate is doing exact comparison on ln from floating point numbers, so the results are pretty order of operations sensitive (result is a difference in rounding). I haven't dug into what exactly could have caused the difference, I don't think the cmath implementation should have changed in this rev.

        let anw = sum_bid.ln() - sum_ask.ln();
        let result = orderbook.get_ordebook_skew_by_range(&range_diff);

        assert_eq!(anw, result);

and https://github.com/TheLastBytebender/rust-workshop/blob/66bb073d11487d3ad4ffa9716676e4f03666afad/src/trading/orderbook.rs#L126

[tgross35]

The second one is more surprising, those numbers are pretty different

[INFO] [stdout] thread 'tests::test_back_prop' panicked at tests/matrix.rs:280:9:
[INFO] [stdout] assertion `left == right` failed
[INFO] [stdout]   left: 2.9802323e-9
[INFO] [stdout]  right: 5.9604646e-9

Third one

[INFO] [stdout] thread 'processor::processor_test' panicked at src/processor.rs:173:9:
[INFO] [stdout] assertion `left != right` failed
[INFO] [stdout]   left: 0.0
[INFO] [stdout]  right: 0.0

[zachs18]

The lightbird test failure (third one) is not a regression AFAICT; I can reproduce it locally with 1.73.0, 1.76.0, and 1.80.1, after running it a few times.

I'm not entirely sure I understand the test in question, but it appears to be asserting that no processor had 0% usage over the past second, which is not necessarily the case, especially on multi-core systems and especially when running only cargo test and not much else.

[workingjubilee]

Hmm. Can we exclude a crate from running its tests only but still check if it builds?

[workingjubilee]

This was a float issue that got classed as a sorting issue (but all the sorting issues seem to actually be hashmap issues): https://crater-reports.s3.amazonaws.com/beta-1.81-2/beta-2024-07-26/gh/tommarek.MPCHomeControl/log.txt

[apiraino]

This is now a stable regression, but I'm not sure what we can do (or want to do)

[jieyouxu]

P Triage: status unclear, are these regressions still problematic?

[apiraino]

I guess we'll keep this issue around in case we can reference it from other reports (Zulip thread)

@rustbot label -I-prioritize