search

holoviz-topics / examples

Visualization-focused examples of using HoloViz for specific topics
https://examples.holoviz.org
Creative Commons Attribution 4.0 International
82 stars 24 forks source link

Best practice for pinning min version in anaconda-project.yml #275

Closed sandhujasmine closed 1 year ago

sandhujasmine commented 1 year ago

Benefit Adding a lower bound on package versions in the anaconda-project.yml makes it easy to see min requirement at the time of lock file generation without actually looking at the lock file. This is convenient and nice for sure.

Cost We already have a lock file with this info even if it's not convenient for a human to read/parse.

Is this nice to have for all packages in anaconda-project.yml? If so, should we consider writing a script to automatically get the min versions for packages that are unpinned from reading the lock file? I did this in a quick way and while the min versions look correct, I'm not sure the solver will resolve to the same lock file.

Curious to know if:

Thanks!

jbednar commented 1 year ago

It's great to have this info when a notebook is first contributed to this repo. And if we had a feature in conda to "install like it's April 2021" or similar (which I think I requested years ago), it would be great to have this info recorded retroactively for the existing notebooks. It's a bit confusing when it's a notebook from 2016 with a solve done in 2023, since those are not the minimum versions this notebook should work with. So I'm not sure whether it's helpful or confusing to add this info for existing older notebooks.

maximlt commented 1 year ago

Having the information on the minimum set of pins would be useful in many ways I'm sure but:

  1. Reporting back the pins in the anaconda-project.yml file is cumbersome, to say the least
  2. I'm not sure we can enforce it, i.e. force all the projects to have a minimum pin to all their direct dependencies, and if we don't enforce it I don't think it's going to happen (mostly because of 1.)
  3. I wouldn't invest too much time in scripting that as I hope that some time hopefully not too far ahead we'll migrate from anaconda-project to conda-project
sandhujasmine commented 1 year ago

Yep - makes sense and agree that it is cumbersome to do this. Take-away: No need to pin in anaconda-project.yml when the lock file is being updated; unless there is a strict requirement like on jupyter_client because 8.0 currently has issues, see holoviz/holoviews#5699

sandhujasmine commented 1 year ago

Sorry - accidentally closed it. But I think this is resolved as above and can be closed.