If you've looked at the parquet code base you know there are many, many
classes and methods marked for deprecation, usually explicitly in the 2.0
release.  These deprecations have accumulated over the past 10 years.

I have played with using Copilot to mass remove and remediate the
deprecated code.  It's definitely a powerful tool for doing that
especially if you have access to the premium LLMs.  The downside of doing
it in bulk is that I would often end up with a stubborn unit test that was
very hard to make fit and so many changes had been made it was hard to
understand which change was creating the difficulty and if something
went wrong.  The final PR is also usually quite large and hard for a human
to follow.

Removing one method/class at a time keeps the changes simpler and
understandable, especially for human review.  It will just take a lot
longer and create a much larger number of PRs.

I don't think any of this is terribly surprising, but I wanted to highlight
that removing all of the deprecations will be a decent amount of work.
That should be considered if there are any goals for the timeline of a 2.0
release

Cheers, Aaron

-- 
Aaron Niskode-Dossett, Data Engineering -- Etsy

Reply via email to