search

fyalcin / surfflow

High-throughput workflows to calculate surface energies of solids.
MIT License
7 stars 6 forks source link

Some questions on the dependencies used by this package. #1

Closed hongyi-zhao closed 5 months ago

hongyi-zhao commented 5 months ago

Regarding the numerous dependencies required by this package, I have the following questions:

  1. You use very strict version specifications, aka, for all the dependencies, == is used instead of >= to select versions. Why?
  2. Some packages already have newer implementations, for example, why not use the recent atomate2 implementation instead of the older atomate?

Regards, Zhao

fyalcin commented 5 months ago

Hey @hongyi-zhao,

To be honest, I'm still educating myself about best practices when it comes to setting dependencies. As I understand it, if one expects their package to be used as a library (expecting it to be installed in an already mature environment), then the dependencies should be as abstract as possible, with lower and/or upper bounds as you suggested.

This package, however, is meant to be standalone, which is why I also generated the monstrosity of a requirements.txt you see using pip-tools. I will update the README soon with a note about the best reproducible way to install this package, which is simply git clone followed by pip install -r requirements.txt, in a fresh environment. To be fair, I'm still playing around with things, and whenever I create a new environment to install, I just pip install -e . after I clone the repo, which usually grabs the most recent versions of the dependencies in pyproject.toml. I should, at some point, define some bounds for the dependencies in file, though.

If you have any other questions, please don't hesitate.

hongyi-zhao commented 5 months ago

My core philosophy is:

  1. If your project depends on many packages, how can you ensure that you're using the latest and most stable features of these packages? If you use >= for specifying dependencies, pip will choose the latest available version that meets or exceeds the specified version.
  2. If there are better alternatives available among these dependencies, efforts should be made to ensure compatibility with these implementations. For example, can your package work compatibly with atomate2?
fyalcin commented 5 months ago
  1. If your project depends on many packages, how can you ensure that you're using the latest and most stable features of these packages? If you use >= for specifying dependencies, pip will choose the latest available version that meets or exceeds the specified version.

That would be quite risky, as some of the dependencies (pymatgen being one) sometimes introduce breaking changes. I still follow changelogs and if I see a change important enough to bump up dependency versions, I will do so. Other than that, I'm content with knowing that the package works as intended with the packages specified in requirements.txt.

  1. If there are better alternatives available among these dependencies, efforts should be made to ensure compatibility with these implementations. For example, can your package work compatibly with atomate2?

This is something I've considered and decided against in the interest of time and other work because I only ran into atomate2 well into the development cycle. At some point, I will spend some time with atomate2 to gauge how much effort it would be to convert.

fyalcin commented 5 months ago

I guess I addressed your questions/concerns and can close this now.