Bug: `runtime_sync()` uses `Path(__file__)` from inside mobuild package instead of caller's notebook

[fr1ll]

Summary

When mobuild.runtime_sync() runs in a notebook that is outside of mobuild, it attempts to _write_file where the file parameter (i.e. input file) resolves to the path of mobuild/init.py rather than to the current notebook.

This causes a crash because mobuild's ._app is a Typer app rather than a Marimo notebook.

Reproduction

Expected behavior

The notebook's exported cells (those marked with ## EXPORT) are written to src/package_name.

Actual behavior

File ".../mobuild/__init__.py", line 132, in runtime_sync
    _write_file(Path(__file__), output_path)
File ".../mobuild/__init__.py", line 75, in _write_file
    order = internal.execution_order
File ".../marimo/_ast/app.py", line 971, in execution_order
    self._app._maybe_initialize()
AttributeError: 'Typer' object has no attribute '_maybe_initialize'

Root cause

In mobuild/__init__.py:132, Path(__file__) resolves to mobuild's own __init__.py (because __file__ in any function refers to the module that defines the function, not the caller). _write_file then:

1. Imports mobuild/__init__.py as a "notebook".
2. Grabs its module-level app — which is typer.Typer(no_args_is_help=True) from line 12.
3. Wraps it in InternalApp and accesses .execution_order, which calls _maybe_initialize() on the Typer object → AttributeError.

Suggested fix

runtime_sync needs to discover the caller's file, e.g. via inspect.stack()[1].filename or by accepting the notebook path as an argument, then pass that to _write_file.

Attribution

Diagnosis and reproducible example by Claude Opus 4.7 with minimal edits to the explanation by me.

[fr1ll]

Thanks @koaning for creating mobuild -- I am using it via CLI mobuild export and it's very useful.

I particularly like the choice to just use ## EXPORT to flag cells for export.

I've been an nbdev user for some time, but increasingly I use marimo. I was considering writing something like this myself when I saw your video mentioning it! 😄

I ran into the issue documented here the first time I tried to do runtime_sync. Credit to Claude for the reproducer and the diagnosis. I've shared the diagnosis here and the reproducer on molab after minimal manual edits.