Simplifying the BrainGlobe (meta-)package structure for users and developers

[alessandrofelder]

Currently, the BrainGlobe tools are split into 35 different repositories (not all of which are to be used by users directly), which

• makes searching for what package to install hard for users
• makes CI and testing for interdependent packages hard

We'd like to address this by

• [x] brainglobe/brainglobe-meta#40, so users can have a one-line install conda install -c conda-forge brainglobe that includes all up-to-date tools
• [x] do the equivalent for pypi (if possible?), so users can do pip install brainglobe with an optional napari dependency if needed: pip install brainglobe[napari]

This has the additional advantage that we can hide changes in the repo structure from users. So once this is done, we'd like to

• [x] brainglobe/BrainGlobe#3
• [x] brainglobe/BrainGlobe#4
• [x] brainglobe/BrainGlobe#6

in order to make CI and testing between closely related repos easier.

Possibly useful resources around Python Packaging

• pyopensci package guide
• Creating an open source python project from scratch
• Conda forge docs to add packages
• SeptembRSE talk about Python wheels

[alessandrofelder]

@adamltyson the above is initial draft for a meta-issue for @willGraham01 to work on once he joins. Comments welcome.

[adamltyson]

All looks good. We have the brainglobe name on PyPI. This package was eventually renamed bg-atlasapi.

do the equivalent for pypi (if possible?)

I think it's possible, even it it's just:

• A pyproject.toml file that has all the other packages as requirements
• An empty __init__

This would also allow us to do things like: init.py

from cellfinder_core.main import main as cell_detector_3D
from brainreg.main import main as register_whole_brain

Usage

import brainglobe as bg

bg.cell_detector_3D()
bg.register_whole_brain()

Then in later versions, we can completely restructure the underlying packages.

[adamltyson]

Superceeded by brainglobe/BrainGlobe#33