Detailed package versions in requirements

[DominikFilipiak]

Some packages are loosely defined in the requirements. For example, matplotlib does not have an upper boundary for its version. With 3.9.4 running panda.teach(panda.qz) from the README will open a window, but the sliders will be unusable. Downgrading Matplotlib to 3.8.6 solved this problem for me, but I expect that we will encounter more in the future. For some of my students the interactive three angles demo doesn't work and we're not sure why - probably some other version mismatches.

Therefore, can we ask for the detailed packages version? Event a dump of pip freeze will be helpful here, I can take care of tidying it up and making PR.

[petercorke]

Matplotlib is a great package but really annoying. Because my maintenance efforts have become very sporadic (am trying to improve) I didn’t want to prevent people from using the latest versions. For some packages like numpy I had no idea they were going to release a major breaking change, and that caused a lot of grief for users. In the first instance I will try to understand why 3.9.4 breaks sliders. I worry that the pip freeze data I send will contain old versions of very many packages, and may not be helpful to users in general. Should we just cap the ones we know are harmful?

…

On 25 Oct 2024, at 6:12 am, Dominik Filipiak ***@***.***> wrote: Some packages are loosely defined in the requirements. For example, matplotlib does not have an upper boundary for its version. With 3.9.4 running panda.teach(panda.qz) from the README will open a window, but the sliders will be unusable. Downgrading Matplotlib to 3.8.6 solved this problem for me, but I expect that we will encounter more in the future. For some of my students the interactive three angles demo doesn't work and we're not sure why - probably some other version mismatches. Therefore, can we ask for the detailed packages version? Event a dump of pip freeze will be helpful here, I can take care of tidying it up and making PR. — Reply to this email directly, view it on GitHub <#21>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC2BIUSPG47V5JL4TALWISTZ5FIDLAVCNFSM6AAAAABQR3H4PKVHI2DSMVQWIX3LMV43ASLTON2WKOZSGYYTENBXGUYDGMQ>. You are receiving this because you were assigned.

[DominikFilipiak]

I'd say we should cap all packages, as we can only expect more of such problems in the future. Providing exact package versions will prevent these.

This way users can make their environments with battle-tested packages on a specific python version and if an end user wants to have newer libraries that can be done in a separate virtual environment. After all, that's why they were invented. So I'd not fear submitting pip freeze. We can, however, explicitly suggest to install RVC tool in a separate venv.

PS. For Jupyter animations, I also needed to write my own function (eg. in chapter 2 notebook):

def tranimate2(T, **kwargs) -> str:
    dim = kwargs.pop("dims", None)
    ax = kwargs.pop("ax", None)
    anim = Animate(ax=ax, **kwargs)
    anim.trplot(T, **kwargs)
    return anim.run(**kwargs)

It was the dim/dims which yielded the error in the original code.

PPS. I really appreciate all the work you have done. It is a wonderful teaching material for introductory robotics course.

[petercorke]

Thank you for your interest in RVC3-python, and for the thoughtful discussion here. Digging back into this: none of rvc3python's dependencies (RTB, MVTB, bdsim) pin matplotlib either, so this isn't an RVC3-python-specific gap — it's an unpinned posture across the whole toolbox family. The specific breakage you and others hit traces to matplotlib ~3.9.0's Slider API change; that's several releases behind us now and appears fixed upstream (this repo's own test suite currently runs clean on matplotlib 3.11).

Given how sporadic active maintenance here is, we've decided against pinning: a static version range needs ongoing upkeep to stay accurate, and can't protect against whatever the next bad release turns out to be anyway. Recommended instead:

• If you hit visual/widget glitches, try a different matplotlib version (usually upgrading) — that's fixed this every time it's come up so far.
• Install into a dedicated conda/venv environment, so rvc3python's dependency resolution never touches a matplotlib version you rely on elsewhere.

We're adding this reasoning to the wiki's FAQ so it's easy to find next time this comes up. Closing this out, but thank you for pushing on it.