Closed gassmoeller closed 1 year ago
Ok, so it turns out it was not pandoc after all. During my testing I had set the channel-priority
setting of conda to strict
, which means once conda finds a package in one channel (e.g. conda-forge) then it doesnt look for other channels (like defaults) anymore. This removes a lot of possibilities for different package combinations and avoids the problems resolving the environment.
If the testers pass and we agree this is a good solution this is ready to merge.
After setting channel_priority: strict
it's taking about 5 min for me on Linux with conda 23.3.1 on Python 3.10. Better than hours for sure. Is that what we expect?
I think this is similar to the time it took before it broke, so yes, I think this it correct. We could probably do faster with mamba or using mamba's solver in conda, but part of the test is to see if the instructions that we give in the documentation actually work, so I want to keep the two consistent.
Our current conda environment seems to fail to resolve on linux machines. It works if I remove pandoc from the environment file. Conda will still be pulled in implicitly, but with a version that seems to satisfy requirements more easily. I can still build the documentation locally, let's see what the online tester says.