https://issues.dlang.org/show_bug.cgi?id=23630

Dennis <dkorpel@live.nl> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dkorpel@live.nl
                 OS|Windows                     |All

--- Comment #1 from Dennis <dkorpel@live.nl> ---
Trying to reproduce it locally, it seems a segfault happens during certain gc allocations when I run `dub test`. However, when I `dub build --force --config=unittest && ./golem-test-unittest` is succeeds, which is really weird.

I'm trying to bisect it.

--

https://issues.dlang.org/show_bug.cgi?id=23630

--- Comment #2 from Dennis <dkorpel@live.nl> ---
When using digger bisect, it fails on older dmd versions with an assertion
error on golem.models.linear:165
```
assert(model.predict(dataset[5][0])[0] > 0.5);
```

--

https://issues.dlang.org/show_bug.cgi?id=23630

--- Comment #3 from lempiji@gmail.com ---
Thank you for investigating this issue quickly.

The failure of assert that was pointed out may be due to the use of random
numbers.
As far as I can tell from running it multiple times on my end, there is no
problem with the data, and it completed normally with version 2.100.1 (ldc2
1.30).
It does not seem to be related to the segfault that occurs with version
2.101.0.

Could you please try running it a few more times to confirm?

--

https://issues.dlang.org/show_bug.cgi?id=23630

--- Comment #4 from lempiji@gmail.com ---
Hello.

Is it possible to solve this problem?

Earlier you mentioned that you used git bisect for troubleshooting. Is it possible to investigate and fix it locally for me in the same way? I would like to know the procedure if possible.

Any other additional information or assistance I can provide?

--

https://issues.dlang.org/show_bug.cgi?id=23630

--- Comment #5 from Dennis <dkorpel@live.nl> ---
I'm dissecting with Digger: https://github.com/CyberShadow/Digger

Weirdly, `dub test` succeeds now with current dmd master for me.

--