Silent crash on GitHub CI that I don't know how to start debugging

[maartenflippo]

I help maintain a constraint solver implemented in Rust with a Python interface through PyO3. I wanted to make the python interface stateful, but our solver does not implement Send. Hence, I annotated our pyclass with unsendable. Since that change, our CI fails on Windows, but pytest contains no output (see the action here: https://github.com/ConSol-Lab/Pumpkin/actions/runs/20173204097/job/57914338591?pr=315). On my local windows installation I cannot reproduce the failure, and clearly it is no problem for Linux or Mac.

My question now is, is there anybody with an idea how to start debugging this problem? At the moment I do not even know where to start with the limited amount of information I have. Presumably, it has something to do with unsendable, which I know may crash under certain circumstances, but I would expect some output at least.

[davidhewitt]

Hmm, I am unsure what the solution is here. It's surprising that it crashes in just 2s, that seems extremely fast, like an error during import?

If you don't use PyO3's unsendable but instead make your own wrapper that guarantees thread-safety (e.g. panic on access from the wrong thread), do you get different results?

[maartenflippo]

After long debugging it seems to be a out-of-memory crash. The tests eagerly created instances of the Rust data type (which is extremely large). I refactored the tests to use pytest fixtures for delayed instantiation which fixed the problem.

I noticed this by enabling tests step-by-step, but I still do not know if there is another way for me to have figured out it was a memory crash.

Anyways, thank you for the great work on this amazing crate!