After a collaborator tried to install Trippy (from Pypi, not from GitHub) and was having issues, I looked into this. The following reproduces the procedure and error that my collaborator got:
conda create -n trippyconda37 python=3.7 numpy stsci matplotlib astropy sextractor emcee numba pytest
conda activate trippyconda37
pip install trippy
python -c "import numpy, scipy, matplotlib, astropy, emcee, numba"
python -c "import trippy; from trippy import bgFinder, psfStarChooser, scamp, MCMCfit"
cd
git clone https://github.com/fraserw/trippy.git
cd trippy/trippy/tests
python tests.py
Which results in the following error:
fail.txt
The same happened if I tried with Python 3.6 rather than 3.7. I believe that what the error message means is that a numpy float is used as an integer index, which is of course a no-no now that it's apparently no longer obvious how to turn a float into an integer. When I remove the pip installed trippy and instead soft-link to the version in the git directory, it all works. Turns out pip is still installing version 1.1.1 whereas here on GitHub we are at 1.2. I assume that the problem with 1.1.1 is only relevant for some versions of Numpy or Python, but rather than looking into that, perhaps its easier to just push the version that does work to Pypi?
After a collaborator tried to install Trippy (from Pypi, not from GitHub) and was having issues, I looked into this. The following reproduces the procedure and error that my collaborator got:
Which results in the following error:
fail.txt
The same happened if I tried with Python 3.6 rather than 3.7. I believe that what the error message means is that a numpy float is used as an integer index, which is of course a no-no now that it's apparently no longer obvious how to turn a float into an integer. When I remove the pip installed trippy and instead soft-link to the version in the git directory, it all works. Turns out pip is still installing version 1.1.1 whereas here on GitHub we are at 1.2. I assume that the problem with 1.1.1 is only relevant for some versions of Numpy or Python, but rather than looking into that, perhaps its easier to just push the version that does work to Pypi?