search

haesleinhuepf / devbio-napari

BSD 3-Clause "New" or "Revised" License
43 stars 13 forks source link

current build incorrectly installing aicsimageio version and dependencies #14

Closed TimMonko closed 2 years ago

TimMonko commented 2 years ago

Hi @haesleinhuepf , Back again. CZIreader from AICSImageIO does not work in current conda build. conda create -name new-env -c conda-forge devbio-napari. I have found the reason is due to 1) lacking install of xmlschema module and 2) some other weirdness... older forges result in aicsimageio 4.7.0 and aicspylibczi 3.0.5, consistent with the current versions installed by conda forge. Installing the missing xmlschema module does not fix the issue, so not sure what tangles are back there.

Solution: conda install aicsimageio solves this by installing aicsimageio 4.9.1, aicspylibczi 3.0.5 and additionally installs xmlschema 1.4.1. Alternatively I had success with conda create --name new-env -c conda-forge aicsimageio devbio-napari

Is the way to solve this just by putting aicsimageio>=4.9.1 in the requirements.txt?

haesleinhuepf commented 2 years ago

Hi Tim @TimMonko ,

we cannot add aicsimageio due to licensing incompatibly. Sad I know.

See #5

CZIreader from AICSImageIO does not work in current conda build. conda create -name new-env -c conda-forge devbio-napari.

Can you describe the problem a bit in detail? And consider telling the aicsimageio folks about it. Your description sound like their installation is broken ...

TimMonko commented 2 years ago

@haesleinhuepf thanks for the reply.

To describe the problem in more detail... I think it was my doing ;) At some point I must have accidentally pip installed aicspylibczi in my base python directory. As such, conda list exposed the conda installed packages... but masked them if there was a package in the base python envs folder. I discovered this by comparing conda list and pip list.

Now doing it correctly (and reinstalling everything, this time with miniconda) conda list correctly shows packages install with pip and properly displays the source (pip, conda, etc). Previously with the packages in the base directory, conda list displayed the source as conda-forge, but would load the older version masked package.

It finally makes sense why building devbio-napari on different machines caused many different results -- depended if I bricked the base directory back when I was learning python or not.

Cheers :)

haesleinhuepf commented 2 years ago

Thanks for the explanation @TimMonko ! So your problem seems solved? If not feel free to reopen this issue and let me know how I can help.

Cheers! 😃