Python package

[jacobwhall]

I want to use PyO3 to create a Python package for skyway. This issue tracks progress to this end, and design ideas for the Python API

[jacobwhall]

conversion = skyway.create_conversion(input_file="data.osm")
conversion.add_filter("myfilter.skyfilter")
conversion.run()

Some thoughts:

• Might want to keep the "builder pattern" from Rust, so that conversion settings can be re-used.
• Convenience functions like (hey just convert this thing into this other thing) are my priority, but it would be nice to expose skyway internals to let Python users add their own filter logic or something.
• Allow Python users to pass buffers into skyway? Or just filenames?
• Shapely / Geopandas support is a must (though maybe an optional dependency)

[jacobwhall]

from typing import Dict, List
from skyway import convert, Element, Reader

input_reader = Reader(input_file="data.xml")

elements: List[Element] = input_reader.read()

first_element_tags: Dict[str, str] = elements[0].tags

[jacobwhall]

There seems to be two general use-cases for Python bindings:

1. Pass relevant file paths, filters to skyway and let it do a conversion on its own
2. Translate element objects into Python using skyway where the user can do something with them

The first is relatively easy. It's basically just requires recreating the functionality of src/main.rs in PyO3

from skyway import convert

# This could look a lot like the existing skyway CLI
# `input` and `output` would take pathlike objects
# this function could also take filter strings and/or pathlikes
skyway.convert(input="data.ext", from="xml", output="out.json")

The second is probably more important. It requires Python-native data types for skyway types, and a way to pass those types across the language boundary.

from typing import List
from geopandas import GeoDataFrame
from shapely import Geometry
from skyway import Element, read_elements, read_geometries, write_elements

input_reader = Reader(input="data.osm")

elements: List[Element] = input_reader.read_elements()

first_element_geom: Geometry = elements[0].as_geometry()

gdf: GeoDataFrame = input_reader.read_geometries()

# User could manipulate a list of elements however they like before writing them out
write_elements(elements, output="example.opl")

[jacobwhall]

I have implemented use-case 1 in the repo now. It appears to work™

v0.8.0 will likely include an experimental use-case 1, and I'll punt use-case 2 for future releases.