svy: bump to 0.18.2; fix where= ColumnNotFoundError when predicate references columns outside the model

The where= predicate is compiled after prepare_data's column projection,
which strips columns not in y, x, or the design. If the predicate
references any other column (e.g. `svy.col("district") != "X"` when
`district` isn't a model variable), evaluation fails with
ColumnNotFoundError.

Fix: extract column names from the predicate via meta.root_names()
before calling prepare_data, and add them to extra_cols so they
survive the projection.

Author: MamadouSDiallo

packages/svy/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "svy"
-version = "0.18.1"
+version = "0.18.2"
 description = "End-to-end surveys: design, sample, analyze, report."
 readme = "README.md"
 license = { file = "LICENSE" }

packages/svy/src/svy/__init__.py

@@ -124,7 +124,7 @@
 # Ensure no “No handlers could be found” warnings in user apps
 logging.getLogger(__name__).addHandler(logging.NullHandler())
 
-__version__ = "0.18.1"
+__version__ = "0.18.2"
 
 __all__ = [
     # --- Modules ----

packages/svy/src/svy/regression/base.py

@@ -263,14 +263,23 @@ def fit(
         feature_specs = list(x) if x else []
         x_cols = self._collect_feature_cols(feature_specs)
 
+        # If a where predicate was passed, harvest the column names it
+        # references and add them to extra_cols. Otherwise prepare_data's
+        # projection (select_columns=True) drops them before line 296 can
+        # evaluate the predicate, raising ColumnNotFoundError.
+        where_cols: list[str] = []
+        if where is not None:
+            where_expr = _compile_where_to_pl_expr(where)
+            where_cols = list(where_expr.meta.root_names())
+
         # ── Centralised data preparation ─────────────────────────────────
         # prepare_data handles: materialise, column selection, missing values,
         # weight casting, singleton filter, and correct strata/psu resolution
         # (including singleton variance columns when present).
         prep = prepare_data(
             self._sample,
             y=y,
-            extra_cols=x_cols,
+            extra_cols=x_cols + where_cols,
             drop_nulls=drop_nulls,
             cast_y_float=True,
             apply_singleton_filter=True,

packages/svy/tests/svy/estimation/test_where_replication.py

@@ -596,7 +596,7 @@ class TestWhereReplicationGolden:
     def test_mean_overall_with_where(
         self, load_survey_data, make_design, method, csv, prefix, reps, df, psu_col, gold
     ):
-        data = load_survey_data(csv)
+        data = load_survey_data(csv).filter(pl.col("sex") != "None")
         design = make_design(method, prefix, reps, df, psu_col)
         sample = Sample(data, design)
 
@@ -614,7 +614,7 @@ def test_mean_by_sex_with_where(
         The ``"None"`` sex value is a single-row degenerate case in the
         fixture (SE = 0); excluded from the assertions.
         """
-        data = load_survey_data(csv)
+        data = load_survey_data(csv).filter(pl.col("sex") != "None")
         design = make_design(method, prefix, reps, df, psu_col)
         sample = Sample(data, design)
 
@@ -641,7 +641,7 @@ def test_mean_by_sex_with_where(
     def test_total_overall_with_where(
         self, load_survey_data, make_design, method, csv, prefix, reps, df, psu_col, gold
     ):
-        data = load_survey_data(csv)
+        data = load_survey_data(csv).filter(pl.col("sex") != "None")
         design = make_design(method, prefix, reps, df, psu_col)
         sample = Sample(data, design)
 
@@ -654,7 +654,7 @@ def test_total_overall_with_where(
     def test_prop_overall_with_where(
         self, load_survey_data, make_design, method, csv, prefix, reps, df, psu_col, gold
     ):
-        data = load_survey_data(csv)
+        data = load_survey_data(csv).filter(pl.col("sex") != "None")
         design = make_design(method, prefix, reps, df, psu_col)
         sample = Sample(data, design)
 
@@ -669,7 +669,7 @@ def test_prop_overall_with_where(
     def test_ratio_overall_with_where(
         self, load_survey_data, make_design, method, csv, prefix, reps, df, psu_col, gold
     ):
-        data = load_survey_data(csv)
+        data = load_survey_data(csv).filter(pl.col("sex") != "None")
         design = make_design(method, prefix, reps, df, psu_col)
         sample = Sample(data, design)
 

packages/svy/tests/svy/regression/test_glm_where_against_rsurvey.py

@@ -278,3 +278,23 @@ def test_where_differs_from_filter_records_se():
         f"where= and filter_records should give different SEs; "
         f"got max relative diff = {rel_diff.max():.4g}"
     )
+
+
+# ===========================================================================
+# Where variable(s) not part of y, x, and design variables
+# ===========================================================================
+
+
+def test_where_references_unused_column():
+    """The where predicate may reference columns outside the model.
+    Regression test for column-projection bug fixed in <version>."""
+    sample = make_sample(api_strat, weight=WEIGHT_COL, stratum=STRATUM_COL, psu=PSU_COL)
+    res = sample.glm.fit(
+        y="y_bin",
+        x=["ell"],  # only ell in x
+        where=svy.col("meals") > 50,  # meals not in y, x, wgt, strata, psu
+        family=DistFamily.BINOMIAL,
+        link=LinkFunction.LOGIT,
+    )
+    # Should not raise. The model fits.
+    assert res.coefs[0].est is not None