[Aarab Malik (aarab-malik)]

Motivation

Float(allow_nan=True) still rejects NaN once a bound is set. Polars ranks NaN above every number, so NaN <= max is false and the max / max_exclusive rules fail even though NaN is allowed. NaN >= min happens to pass today, but that depends on the same ordering, so min and min_exclusive have the same bug waiting. allow_nan should be the only switch for NaN.

I also see it in sampling. test_sample_unchecked_nan draws NaNs for Float(min=0, max=10, allow_nan=True), but Schema.sample() filters those rows back out through the max rule, so the checked sampler returns no NaN at all.

Closes #416.

Changes

• In _BaseFloat.validation_rules, when allow_nan is set, I OR expr.is_nan() into each bound rule (min, min_exclusive, max, max_exclusive). Finite values outside the bound are still rejected. NaN is still rejected when allow_nan is false, by the nan rule.
• Added regression tests for inclusive and exclusive min and max bounds, with allow_nan=True and allow_nan=False.
• Checked the other paths that reuse these rules. Schema sampling of a bounded float column keeps NaNs when allow_nan is set, and a list of such floats accepts an inner NaN. Unchecked sampling was already consistent and still is.

[Copilot]

🟢 Approval recommended

There are no active review issues for the change allowing NaN in bounded float columns when explicitly enabled.

No open issues 🎉

What changed in this PR

Fixes #416 by allowing NaN values in bounded float columns when allow_nan=True, keeping validation and checked sampling consistent.

Changes:

• Makes all four float bound rules accept NaN when explicitly allowed.
• Adds parameterized regression tests for bounds with NaN allowed and disallowed.
• Covers NaN preservation in checked sampling and validation of nested float lists.

File	Description
tests/​column_types/​test_float.py	Adds regression coverage for bounded NaN validation, sampling, and lists.
dataframely/​columns/​float.py	Allows NaN through bound rules when allow_nan=True.

_{Review effort: Balanced}

[Oliver Borchert (borchero)]

Nice, thank you!

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (58a2402) to head (e672cbf).

Additional details and impacted files

@@            Coverage Diff            @@
##              main      #423   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           46        46           
  Lines         2584      2587    +3     
=========================================
+ Hits          2584      2587    +3     

☔ View full report in Codecov by Harness.
📢 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.