[Package Request] tornado (Required by jupyter-kernel-client / google-colab-cli)

[Nwokike]

Hello Flet team!

I am building a Colab client app for Android using Flet. I rely on the google-colab-cli package for session management and code execution, which depends on jupyter-kernel-client (by Datalayer).

jupyter-kernel-client requires jupyter-client>=8.6.3, which in turn requires tornado>=6.4.1. Because tornado contains C extensions and has no Android wheel on pypi.flet.dev, pip's resolver silently backtracks and installs an old, unrelated version of jupyter-kernel-client (0.1.4 by Wh1isper) whose dependencies can be satisfied — but this version is missing all the APIs my app needs (e.g. JupyterSubprotocol, KernelClient).

Dependency chain:

google-colab-cli → jupyter-kernel-client (0.9.0) → jupyter-client → tornado ❌

Package Details:

• Name: tornado
• PyPI: https://pypi.org/project/tornado/
• Latest version: 6.5.1
• Use Case: Async networking library required by jupyter-client, which is the standard Jupyter messaging/session layer used by jupyter-kernel-client for WebSocket kernel communication.

Could we get tornado added to the mobile forge so jupyter-kernel-client (and by extension google-colab-cli) can resolve correctly on Flet mobile?

Thank you!

[ndonkoHenri]

I can work on tornado (looks pretty trivial), but it doenst seem to be the only blocker for your request/usecase. The next one is pyarrow (jupyter-kernel-client → jupyter-mimetypes). Thats a harder guy i've been working on in the past days (for #3275). Was able to build it for Android, but supporting it on iOS is a little more complex.

[ndonkoHenri]

tornado 6.5.7 and pyarrow 24.0.0 are now supported on Android and iOS.

Retry building with your app with the google-colab-cli dependency and let know if you face any issues.

[Nwokike]

@ndonkoHenri Thank you so much! 🎉

Just rebuilt and tested. The app works perfectly on Android now. jupyter-kernel-client resolves to 0.9.0, and code execution on Colab sessions works from my phone. The fact that you proactively shipped pyarrow too before I even asked is next-level.

If you get a chance, I also have an open request for llama-cpp-python (#6627) for running local LLMs on device. Would love to know if that one's feasible, and is there anyway I can help?

Thank you very much for this. 🙏

[ndonkoHenri]

It's a pleasure.

Thanks for letting know it works. I would be interested to know a little more about your usecase - sounds interesting. 🙂