Brian Ward

If you’d like to not be the reviewer, you will still need to remove your existing “request changes”. We can then ask @rok-cesnovar to review?

I think a best-effort basis is what we've done for some similar hackery on Windows. It's almost impossible to test this in CI. Is there no way to do this...

Yeah, unsure how to handle this. I think a simple check that says if the arch of the compiler doesn't match the arch of stansummary, fail with a nice error...

I guess the concern is if they installed cmdstan with something set for CXX that is arm64-based, and then later are trying to use a CXX which is x86, won't...

It seems like an easy thing to detect and (at least) issue a warning about, though. Would you mind adding that?

I'm seeing the definition of `cpp_options` as `Dict[str,Any]` - where are you seeing that other type? It is wrong if it is in the code somewhere. I think we need...

Yeah that is incorrect. I think the semantics of mypy’s typing mean that it doesn’t notice that because elsewhere it is declared to be Any

I’m not sure if we have a cmdstanpy dev with an M1/2 machine, so happy to wait. We can push this out of the 1.0.2 milestone

> I'm wondering what the thresholds are for logging the non-convergence warning and whether they're too loose? These were introduced in #577. The current threshold is any of them being...

We could set a configurable threshold for the treedepth and convergences if we didn't mind adding even more arguments to `sample()`. I believe cmdstanr has sample take a sequence of...