search

vojtamolda / Plotly.swift

Interactive data visualization library for Swift
https://vojtamolda.github.io/Plotly.swift/
MIT License
82 stars 8 forks source link

Support for Jupyter causes unresolved symbols in projects using PythonKit #22

Open vojtamolda opened 2 years ago

vojtamolda commented 2 years ago

When Plotly.swift is used in a larger project, that also depends on PythonKit, build fails with unresolved symbols.

Currently, the support for Jupyter notebooks is implement in a slightly hack-ish way

#if canImport(PythonKit)

The #if guard is placed around a Figure extension that implements the .display() method. This works auto-magically without any user intervention in Google Colab/Jupyter Notebooks, because they rely on Python to communicate with the Jupyter kernel and make it available for import by default.

However, when Plotly.swift is used in a larger project, that also happens to use PythonKit, the package fails to build. Since PythonKit is available for import the #if guard evaluates to true. The SwitPM build, though, doesn't see any formally declared dependency on PythonKit in Package.swift manifest and therefore doesn't link against it. This causes the unresolved symbols linking error.

vojtamolda commented 2 years ago

I have a few ideas how to fix the problem but none of them seems to be ideal:

  1. Add PythonKit as a regular dependency. This makes the project a little bit heavier for all users. Everyone would need to download and build it although they don't need the notebook functionality nor Python.

  2. Introduce a new preprocessor variable JUPYTER_SUPPORT (suggestions for a better name are welcome ;) and replace the guard with

    #if JUPYTER_SUPPORT

    This breaks the auto-magically working support for Colab/Jupyter notebooks because users would now have to pass a -D JUPYTER_SUPORT flag.

  3. Detect Jupyter via an environmental variable (ProcessInfo.processInfo.environment dict) at runtime. I'm not 100% sure if this is possible so it needs proper investigation. It isn't clear what to do if the variable isn't defined. The Figure class would have .show() and .display() methods available everywhere without a clear distinction to a new user.

  4. Drop the support for Colab/Jupyter altogether. There seems to be some effort to revive S4TF orchestrated by Philip Turner. So maybe there are better days ahead...