The installation process does not work in many configurations and has a major issue of deps conflict resolution, e.g. for NightOsmReg on Windows : GDAL installation issue, on Linux pyrosm / pyrobuff issue. Also, versions should be bumped to minor updates when available to benefit from bugfixes (e.g. GDAL 3.11.5).
When using conda, the docs should explain how to install most dependencies using the conda-forge registry. It should also provide examples on how to install without conda (pure pip / venv, or using uv which is a really good alternative).
Currently, the README show how to create a conda env with a pinned python version, and libgdal.
Then pip will install every other dependencies and fail to resolve conflicts, and install a PyPI version of GDAL that will come with its own libgdal bundle. Same for rasterio that when installed with pip comes with its own GDAL lib. Installing all using conda should avoid later import errors and segfaults.
I'll open a PR to address current issue by updating pyproject dependencies and recommended install commands, but a deeper package structure refactoring would simplify things for both users and devs IMHO (one package per program), while still being hosted in a single repo.
The installation process does not work in many configurations and has a major issue of deps conflict resolution, e.g. for NightOsmReg on Windows : GDAL installation issue, on Linux pyrosm / pyrobuff issue. Also, versions should be bumped to minor updates when available to benefit from bugfixes (e.g. GDAL 3.11.5).
When using conda, the docs should explain how to install most dependencies using the conda-forge registry. It should also provide examples on how to install without conda (pure pip / venv, or using uv which is a really good alternative).
Currently, the README show how to create a conda env with a pinned python version, and libgdal.
Then pip will install every other dependencies and fail to resolve conflicts, and install a PyPI version of GDAL that will come with its own libgdal bundle. Same for rasterio that when installed with pip comes with its own GDAL lib. Installing all using conda should avoid later import errors and segfaults.
I'll open a PR to address current issue by updating pyproject dependencies and recommended install commands, but a deeper package structure refactoring would simplify things for both users and devs IMHO (one package per program), while still being hosted in a single repo.