feat(rust): Pub-licize Expr DSL Function enums

[Shoeboxam]

Changes the visibility of parts of the Expr DSL that I'd like to add support for in OpenDP (see current docs) to pub.

[ritchie46]

They are not meant to be public. They are internal state of the DSL. Could you explain why you need that public?

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 79.92%. Comparing base (de1d9d5) to head (f4511de).
Report is 8 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #20421      +/-   ##
==========================================
+ Coverage   79.85%   79.92%   +0.07%     
==========================================
  Files        1596     1596              
  Lines      228642   228642              
  Branches     2615     2615              
==========================================
+ Hits       182571   182733     +162     
+ Misses      45473    45311     -162     
  Partials      598      598              

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

[Shoeboxam]

The use-case is a software project (OpenDP) that proves that a given Polars DSLPlan satisfies differential privacy. For an example of its use, the Switzerland Federal Statistical Office used the tool to release high-resolution poverty statistics.

To accomplish this, I've written a number of Polars plugins, as well as a static analyzer for DSLPlans. The analyzer tracks key properties about the data via a recursive parser over each expression/step in the DSLPlan. Over the last year I've been expanding the kinds of expressions and plans that the analyzer can reason about.

For example, if an expression in the plan contains an ArrayFunction::Explode, then in differential privacy jargon the analyzer needs to account for a k-stable transformation (where k is the static array length). Whereas if the DSLPlan contains an ArrayFunction::Min, then the analyzer needs to account for a 1-stable transformation. Unfortunately, since the array enum is private, I can't add logic to the analyzer for array functions, so the analyzer rejects all Polars plans with array functions.

For more context on the project:

• repository
• recent blog post
• getting-started documentation

[Shoeboxam]

Based on my use-case, is this something you would be open to adding to the public API? Thanks.

[ritchie46]

Right, though I do want to warn that we don't guarantee stability on this yet. I first want to get the DSL out into a seperate crate and version that properly.

Anyhow can you rebase?

[ritchie46]

@Shoeboxam can you rebase?

[Shoeboxam]

Right, though I do want to warn that we don't guarantee stability on this yet. I first want to get the DSL out into a seperate crate and version that properly.

Don't worry, I don't expect API stability. I've been making quarterly releases on my side that brings OpenDP in-line with the latest version of Polars that has DSL compatibility between the Rust and Python crate.

Occasionally the Python and Rust DSLs can drift apart due to patch releases with DSL changes, making a monthly Rust release non-viable, in which case I downgrade until finding a matching pair.

Anyhow can you rebase?

Done! Thanks for the response.

[ritchie46]

Yeah, I hope we can get to a typed DSL soon.