Ariel Kwiatkowski

My perspective is just that by allowing `truncated==terminated==True` we get strictly more information. If the terminal state happens to be reached at the same time as truncation would have occurred,...

> so it's still time to keep previous behavior and prevent confusion ;) Can we be adults and skip the passive aggressive comments? This decision was discussed a bunch before...

I don't know what exactly happened, but it's likely something to do with the new version of `setuptools`. Consider the code in [this colab](https://colab.research.google.com/drive/1lhEhaXYCyBoFb4wdicJsq-vCOOMGim_n?usp=sharing): ```python !pip install --upgrade setuptools==66 !pip...

Helpful response

So overall my opinion is that before 1.0, we should keep things as backwards-compatible as possible, then make a complete, sane design for everything mentioned here, and do breaking changes...

Yea, that's the ballpark of what I'm thinking. I would keep it more general than just making it for jax jitting, but it would definitely be one of the possibilities....

@pseudo-rnd-thoughts I'm not sure I understand. If it doesn't expose the state, then it's still very different from what I'm proposing. As I understand, the actual `step` that the user...

This actually makes me even think that it might be worthwhile to add a consistent behind-the-scenes functional API for the upcoming new envs, and potentially only in the future we...

Ultimately I think it's a mix of both things. It's possible to express any computation in an object-oriented way (in fact, it's a thing I pitched a few times as...

> EpisodeData should be immutable A dataclass can be [frozen](https://docs.python.org/3/library/dataclasses.html#module-contents). > How would you define a better representation for that? Is it important? Something more concise with the important characteristics...