refactor(python): InterchangeDataFrame.version should be a ClassVar (not a property)

[twoertwein]

No description provided.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 81.33%. Comparing base (27de0d8) to head (f927b46).
Report is 8 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff            @@
##             main   #16312    +/-   ##
========================================
  Coverage   81.33%   81.33%            
========================================
  Files        1403     1403            
  Lines      183293   183463   +170     
  Branches     2924     2929     +5     
========================================
+ Hits       149085   149226   +141     
- Misses      33705    33734    +29     
  Partials      503      503            

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[stinodego]

On the Protocol it needs to be a @property since it's not instantiated. On PolarsDataFrame it can just be a regular property.

In fact, in the official reference it's a regular property, so we should follow that as closely as possible:
https://data-apis.org/dataframe-protocol/latest/API.html

If there is something not working correctly, please make an issue to discuss and I may reopen this.

[twoertwein]

On the Protocol it needs to be a @property since it's not instantiated.

I'm sorry, I don't understand how property vs ClassVar is affected by it being instantiated or not. The ClassVar approach could be used in the protocol and based on your reference to https://data-apis.org/dataframe-protocol/latest/API.html it should be! I updated the PR with that change.

The issue is that replacing a property with a ClassVar (or the other way round), is not the same from a typing perspective. A ClassVar can be accessed on the class itself, a property can't. A ClassVar can be set, a property without a setter can't.

Mypy does not enforce this consistency unless the parent class explicitly declares the attribute as a ClassVar (pyright flags it always).

edit: since github does not render new commits for closed PRs, here is how the ClassVar version would look like:
twoertwein@6443593#diff-8f55094cafe51841709ae47ec54b72d97f52ce5ee386a6534256ad22d89abd2bR192

[stinodego]

You're right, ClassVar is the way to go here. Thanks for the PR and the explanation!