Thoughts on install targets?

[barrettMCW]

Love the kit, the Macenko deconv and tissue masking implementations are awesome!
The only issue I have is how many dependencies I need to use them.
We use OMERO rather than DSA. Things like girder and large_image can make installation kind of a pain for no benefit in my use case.
Any thoughts on a non-DSA install target?
Working on a model I want to publish (which I'll of course cite you in <3) but it'd be nice to be able to skip the girder stuff so I can simply add your package with that target as a dependency instead of copying the code into my package.
I'd be happy to work with y'all on creating the target!
Thanks for your work y'all! It's a great kit!

[manthey]

Thanks, @barrettMCW . We should make it so that the install has only the minimum necessary, and things like girder are doing in an extras_require. Then pip install histomicstk would get the minimum required, and pip install histomicstk[girder] (or whatever) would get the girder dependencies. Besides refactoring setup.py (and, to keep with the times, perhaps we could move that to just a pyproject.toml), we just have to decide what the minimum divisions would be.

[barrettMCW]

@manthey
Exactly what I was thinking, thank you!
I only use part of the toolkit, mostly deconv and masking, which doesn't (necessarily) use the girder stuff.
I'd be happy to work with y'all on dividing the package requirements.
+1 on moving to pyproject vs setup!
Let me know how I can help! Thanks again!

[manthey]

As far as I can tell, there currently isn't a way to avoid having setup.py or some other python script; scikit-build won't find numpy headers without it (as near as I can tell), nor does skipping scikit-build and trying to use cython and setuptools directly. Maybe I'm just not finding the right documentation. If you know of a way to switch entirely away from setup.py, I'd love to do it, but currently I don't know how.

We can still move certain packages to appropriate extras_requires, of course.

[barrettMCW]

I worked out something! Builds on my end, but I'm not too familiar with cython so I might have missed something.
Migrated to scikit-build-core as per the guide, built on python3.9 on mac

#pyproject.toml

[build-system]
requires = ["scikit-build-core>=0.7", "setuptools-scm", "cython>=0.29", "numpy", "cmake>=3.18"]
build-backend = "scikit_build_core.build"

...

[tool.scikit-build]
wheel.packages = ["histomicstk"]
sdist.exclude = ["*.git*"]
cmake.build-type = "Release"

#./CMakeLists.txt

cmake_minimum_required(VERSION 3.15...3.26)

project(histomicstk LANGUAGES CXX)

find_package(Python3 REQUIRED COMPONENTS Interpreter Development NumPy)

add_subdirectory(histomicstk/features)
add_subdirectory(histomicstk/segmentation/label)
add_subdirectory(histomicstk/segmentation/nuclear)

#./histomicstk/segmentation/nuclear/CMakeLists.txt (same for the other two)

find_package(Python COMPONENTS Interpreter Development.Module NumPy REQUIRED)

add_custom_command(
  OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/_max_clustering_cython.cpp
  COMMENT
    "Making ${CMAKE_CURRENT_BINARY_DIR}/_max_clustering_cython.cpp from ${CMAKE_CURRENT_SOURCE_DIR}/_max_clustering_cython.pyx"
  COMMAND Python::Interpreter -m cython --cplus -o "${CMAKE_CURRENT_BINARY_DIR}/_max_clustering_cython.cpp"
          "${CMAKE_CURRENT_SOURCE_DIR}/_max_clustering_cython.pyx"
  DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/_max_clustering_cython.pyx"
  VERBATIM)

python_add_library(_max_clustering_cython MODULE
                   "${CMAKE_CURRENT_BINARY_DIR}/_max_clustering_cython.cpp"
                   WITH_SOABI)

target_link_libraries(_max_clustering_cython PRIVATE Python::Module Python::NumPy)

install(TARGETS _max_clustering_cython DESTINATION histomicstk/segmentation/nuclear)

[manthey]

Sorry for taking some time to get back to this. Thank you -- that gets much further than I did.

This works locally, but on CI I am seeing failures in the aarch64 builds (in the github actions) and something similar on the CircleCI actions. See the branch pyproject.toml and a typical workflow run like https://github.com/DigitalSlideArchive/HistomicsTK/actions/runs/15395226324.

The dependency split out doesn't have to wait on this (but I'd rather it did).

[barrettMCW]

No worries!

I figured out a way to fix it, feels kind of hacky but it works:

set(Python3_FIND_STRATEGY LOCATION)
set(Python3_FIND_REGISTRY NEVER)
set(Python3_FIND_FRAMEWORK NEVER)

find_package(Python3 COMPONENTS Interpreter Development NumPy)

if(NOT Python3_FOUND)
  find_package(PythonLibs 3)
  find_package(PythonInterp 3)
  find_package(NumPy)
endif()

I believe it has something to do with pypy's non-standard install location.
Working on figuring out how to fix this correctly but we should have a solution.

[manthey]

Thanks, @barrettMCW . I'll make another PR shortly that actually moves some things to optional installs.