CERN / TIGRE

TIGRE: Tomographic Iterative GPU-based Reconstruction Toolbox
BSD 3-Clause "New" or "Revised" License
529 stars 180 forks source link

Use "pip install" for setup #480

Closed tsadakane closed 9 months ago

tsadakane commented 10 months ago

See #479

tsadakane commented 10 months ago

As far as I know, as for now, there are two options to achieve the entitled goal.

  1. Remove pyproject.toml
  2. Add the two lines below to pyproject.toml:
    [build-system]
    requires = ["setuptools", "wheel", "Cython", "numpy"]
AnderBiguri commented 10 months ago

@tsadakane I am definitely not an expert here, but I think maybe 2) is the way to go? Happy if you think otherwise.

tsadakane commented 10 months ago

@AnderBiguri , neither am I...

I tested four patters of the initial conditions before running pip install .

Installed packages by conda before pytigre installation
test01 jupyter numpy matplotlib opencv pillow pandas scipy imageio cython pylint (my "standard" environment)
test02 numpy
test03 Cython
test04 none
after installation, the following packages are installed by pip Additionally installed packages by pytigre installation (increased items of 'pip list')
test01 pytigre
test02 Cython, scipy, pytigre, Pillow, packaging, python-dateutil, pyparsing, matplotlib, six, kiwisolver, contourpy, cycler
test03 numpy, scipy, pytigre, Pillow, packaging, python-dateutil, pyparsing, matplotlib, six, kiwisolver, contourpy, cycler
test04 Cython, numpy, scipy, pytigre, Pillow, packaging, python-dateutil, pyparsing, matplotlib, six, kiwisolver, contourpy, cycler

In all cases, example.py could run successfully.

I think that test04 can be a simulation of the "pip only" environment.

AnderBiguri commented 10 months ago

Fantastic. Should we merge this?