Is there a straightforward way of executing marimo notebooks (cells or in whole) in other applications _with_ arguments?

[Senjai]

I've been trying to understand this a bit better.

The following both work

python run notebook.py -- --arg1 --arg2

and

from notebook import app
app.run()

What perplexes me to no end is I cannot see any way of passing arguments into the run function if I wanted a notebook that is also useable as "source code." One of the most alluring parts of Marimo is that you can have an algorithm run as a notebook, but also as a regular python file. I cannot easily however use the algorithm in other python files as python code.

I would expect this to work:

from notebook import a_cell
a_cell(variables, required)

but this just gives an html output. Apparently I need to use the app script runner or the like to get outputs but can only do so in undocumented ways:

There seems to be only two ways of passing data into a notebook so far that I can tell https://github.com/marimo-team/marimo/blob/main/marimo/_runtime/context/script_context.py#L77-L87.

CLI Args are hardcoded to read from ARGV, which I guess I can spoof if required. Query params, are read, well as query params when running the notebook in a browser environment. I haven't looked further into manually setting this on the app execution environment before running it as I'm already a bit deep into private API's here.

What I would like to do is simply be able to pass in global values into the app.run() method that the notebook could then use. I've made versions of this work terribly using private api's that don't feel fantastic by executing cells manually.

from notebook import app, a_cell
_html, namespace = app._run_cell_sync(a_cell, {"overrideable_ancestor": 243})
print(namespace["output_variable_i_care_about"])

This seems to work just fine. But again, it doesn't seem like this is meant to be a public API for interacting with Marimo so I'd like to avoid building on it. Am I missing something obvious here? Are folks using Marimo similarly to how I aim to here? (n.b. would it make sense to not generate html if we have no use for it in a script context?)

Naively, @app.cell decoration could perhaps cache the original function, and if the function is being called outside of a marimo application context, just run it as a plain python function.

[akshayka]

1. For running cells with arguments, you use the cell's run method. This is documented in the API reference: https://docs.marimo.io/api/cell/#marimo.Cell.run. You'd write code that resembles the following:

from notebook import a_cell

a_cell.run(x=2, y=4)

We might be able to simplify this to omit the run, as syntactic sugar. But this should do what you want.

You may also be interested in following this issue, for serializing pure functions at the top-level of the notebook: #2293.

2. For app.run():

What perplexes me to no end is I cannot see any way of passing arguments into the run function if I wanted a notebook that is also useable as "source code."

Passing arguments to app.run() is not yet supported. It's something we may consider, if there is a compelling use case (if so, feel free to file a feature request as an issue). However, perhaps Cell.run is sufficient for your use case.