Kaleb Barrett

And there in lies the true solution, don't reinvent the wheel.

Yeah we need to better think about how to split these up. Maybe we have a separate `analysis_args` and `elaboration_args` since those map well to a lot of simulators, including...

In the above example, the failure happens, and the outcome is stored. What happens if the forked coro is never yielded on? Where does the exception go?

I don't agree. `fork`ed coroutines throwing exceptions should causing `TestError`. Many uses of `fork` are meant to start up 'threads' (yes, not great terminology but it mirrors `fork`) that run...

One potential way to support both behaviors is to make `scheduler.add` work as listed in the OP, and make `fork` wrap that with a function that yields on the coroutine...

Worth looking at, but I probably won't have time until after release. I'm sure it has something to do with how the custom test is written and not a bug...

I wonder if something has changed to the order of discovery which returns `vpiRefObj` which is not currently mappable, before whatever was previously returned, or if `vpiRefObj` was just mapped...

This was removed in 2.0: #3360. We need code to handle this case correctly. We can't assume that everything behind a `vpiRefObj` is a logic vector.

I'm not entirely sure here. Have you tried using a released version of Verilator? v5.024 works for me (HEAD is currently broken because of changes in Verilator). Make sure you...

Ah. Fun. Conda gets to be special. I think the ultimate solution is to stop setting PYTHONHOME as I mentioned [here](https://github.com/cocotb/cocotb/issues/3643#issuecomment-1879758728), but a PR with that fix would be welcome.