Canvas

[josevalim]

We want to build a canvas where we can freely place outputs. This can be useful for data exploration (for example, the canvas can run in a separate monitor) and also for deployed apps (we can deploy the canvas). This is explorer in the B2 paper and some projects, such as Count, rely heavily on this idea.

[jannikbecher]

The canvas view mode could be visualizing a complete notebook, so there are two main modes: "notebook" and "canvas".
Notebook mode has its own subviews (Normal, Code Zen and Presentation).
Canvas has Normal and App-UI.
Each section is a saved view port where you can easily zoom into.
When declaring new variables in a cell, they are outputs and are connected to each cell where the variable is used.
When selecting App-UI the outputs of the cells are shown in the foreground and can be moved to the wanted location.

This are just some thoughts which came to my mind while on vacation. Is this something worth exploring?

[josevalim]

Maybe but I think we need to go with baby steps. The important is to have the canvas for visualization and app-UI. Maybe other needs will arise in the future, but we need to use the basic versions first, otherwise we may end up cluttering the UI for features that no one needs or use. So I would remain focused on the current proof of concept.

[jonatanklosko]

I wouldn't make the notebook itself non-linear. The linearity of cells gives an intuition about the evaluation order, and I think spreading cells across the screen would easily lead to confusion.

[josevalim]

Closing in favor of #2109.