Extension with external dependencies

[bgailleton]

Hi @wkearn,

Let's say I'd like to add a plugin/extension that's exclusive for pytopotoolbox - for example a mapcomposer plotting engine to create sophisticated 2D/3D figures with matplotlib, pyplot and pyvista; or a cli interface that automate some tasks without having to write a script (e.g. pre-fill a DEM, ...).

What would you prefer in term of structure:

1. a separated package, pros: cleaner, cons: more complicated
2. a submodule (file or full directory with __init__.py) within pytopotoolbox

Cheers,

[wkearn]

I think it is important to keep such extensions separate from pytopotoolbox, so that a user does not need to install the extension or, more importantly, any dependencies, just to use pytopotoolbox.

I imagine that means multiple packages with their own pyproject.toml files, but I could be convinced that there is a better option. Are there examples of similar projects that we might want to base things on? I know landlab keeps certain components as submodules, but you can always create a component outside of that directory: it just makes importing your component a little non-standard. I think it does make adding component-specific dependencies a little tricky.

I'm not opposed, however, to hosting such a package within the pytopotoolbox repository, though that would need some reorganization of our existing code.

Probably the best approach is to start out with a separate repository, and we can figure out whether we want to merge it and what that might look like once we know what you have. I think @Teschl has been working on some QGIS integrations that look like this as well.

[bgailleton]

I really don't have strong opinion about any of the two options. I can really see the pros and cons and I am happy with separate packages. Maybe at some point it could be worth having a template github repository with pre-filled pyproject.toml + all the linting options and other bits and bobs.

One could draw the line at "does it need a new dependencies? then separate package" ?

Any thanks for the answer, That's a design choice in the end and being consistent is what matters

[Teschl]

@wkearn

I think @Teschl has been working on some QGIS integrations that look like this as well.

I have found a way to manage Python dependencies for QGIS plugins, so it's not really required to implement the pytopotoolbox functions with gdal instead of using rasterio. qpip is still in early development, but is able to install and remove needed packages for different plugins in QGIS (TopoToolbox/QGISTopoToolbox#1).

[Teschl]

I just noticed that i meant to comment this under Will's answer