Eric Wieser

> These are one in the same. Unless `registered_impls[0] == nullptr`, in which case they're not and you get a segfault. Could that be happening?

I'm struggling to see how `registered_impls` could be null, nothing seems to clear it.

Can you print `registered_impls.size()`? Perhaps we're indexing beyond the array somehow.

> Because of the nature of floating point arithmetic, almost-correct > results are common. For example, (on my system) 8e-9*1e12 is not 8e3, but > 8000.000000000001, Arguably a better way...

I think the motivation here is > I want to round, but only if I'm already close to a rounded value; so asking for 1.9 simulator steps is "close enough"...

Another approach would be a special object: ```python class _DriverHelper: def __getitem__(self, sig): return sig.value def __setitem__(self, sig, value): sig.value = value value_of = _DriverHelper() ``` Used as: ```python value_of[clk]...

A note on wierdness: Our use of `__le__` means that we inadvertently support ```python 3 >= dut.some_sig ``` as a spelling of ```python dut.some_sig

Doesn't the name of the test in the xml file contain the option information?

I'm not sure, but pytest generates much better test names of the form `test_foo[bar-baz]`. Perhaps just doing that is enough to solve the problem, by changing this last line to...

We ran into this all the time at my old employer, but never came close to finding the cause.