fix: Make unique work for nested types and improve performance

[Oliver Borchert (borchero)]

Motivation

Follow-up to #325, see #325 (comment).

Changes

• Move unique to the column definitions as it is column- rather than schema-business
• Remove the unique_columns classmethod with the same argument
• Use pl.col("...").is_unique() by default instead of wrapping in a struct for improved performance
• Add tests for nested uniqueness and uniqueness of complex types

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (7126c1e) to head (518dab2).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##              main      #333   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           56        56           
  Lines         3408      3406    -2     
=========================================
- Hits          3408      3406    -2     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

This PR refactors how unique constraints are expressed and validated by moving uniqueness checks into per-column validation rules, enabling unique to work for nested column types (e.g., List/Array inner types) and improving performance for scalar columns.

Changes:

• Implement unique as part of Column.validation_rules() (and remove schema-level unique_columns() plumbing).
• Use expr.is_unique() for most columns, with a Polars workaround for List/Array (pl.struct(expr).is_unique()).
• Add/adjust tests for unique on List/Array columns and their inner types; remove the test for the removed unique_columns() method.

Reviewed changes

Copilot reviewed 15 out of 15 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
tests/schema/test_validate.py	Removes test coverage for deleted unique_columns() API.
tests/column_types/test_list.py	Adds unique and inner_unique tests for List.
tests/column_types/test_array.py	Adds unique and inner_unique tests for Array.
dataframely/columns/_base.py	Adds unique rule to base column validation rules.
dataframely/_base_schema.py	Removes schema-level unique column rule construction and unique_columns() method.
dataframely/columns/list.py	Implements unique for list columns using struct wrapper workaround.
dataframely/columns/array.py	Implements unique for array columns using struct wrapper workaround.
dataframely/columns/struct.py	Updates unique parameter documentation.
dataframely/columns/string.py	Updates unique parameter documentation.
dataframely/columns/integer.py	Updates unique parameter documentation.
dataframely/columns/float.py	Updates unique parameter documentation.
dataframely/columns/enum.py	Updates unique parameter documentation.
dataframely/columns/decimal.py	Updates unique parameter documentation.
dataframely/columns/datetime.py	Updates unique parameter documentation.
dataframely/columns/categorical.py	Updates unique parameter documentation.

[gab23r]

Indeed much cleaner than my implementation !
Just for fun. Wapping in_unique in pl.struct is actually faster on micro benchmark 😄

import dataframely as dy
import polars as pl


class Schema(dy.Schema):
    a = dy.Int64()
df = Schema.sample(1000000)

%timeit _ = df.select(pl.struct("a").is_unique())
# 8.08 ms ± 495 μs per loop (mean ± std. dev. of 7 runs, 100 loops each)

%timeit _ = df.select(pl.col("a").is_unique())
# 28.6 ms ± 1.04 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)

This is because pl.struct("a").is_unique() dispatch to df.select("a").unnest().is_unique() which is exectuted in parallel. Which is not the case for Expr.is_unique.

[Oliver Borchert (borchero)]

Thanks for the benchmark gab23r, this is really surprising 👀 but good to know that this behavior changed since I last benchmarked this >1y ago 😅