Avoid mutating Poly score arrays

patsy/contrasts.py

@@ -268,7 +268,7 @@ def _code_either(self, intercept, levels):
         # Strategy: just make a matrix whose columns are naive linear,
         # quadratic, etc., functions of the raw scores, and then use 'qr' to
         # orthogonalize each column against those to its left.
-        scores -= scores.mean()
+        scores = scores - scores.mean()
         raw_poly = scores.reshape((-1, 1)) ** np.arange(n).reshape((1, -1))
         q, r = np.linalg.qr(raw_poly)
         q *= np.sign(np.diag(r))

patsy/test_poly_input.py

@@ -0,0 +1,50 @@
+# This file is part of Patsy
+# See file LICENSE.txt for license information.
+
+import numpy as np
+import pytest
+
+from patsy import Poly, dmatrix, build_design_matrices
+
+
+@pytest.mark.parametrize("intercept", [True, False])
+@pytest.mark.parametrize("kind", ["float", "readonly", "view", "integer", "list"])
+def test_poly_scores_input_unchanged(intercept, kind):
+    owner = np.array([0.0, 91.0, 10.0, 92.0, 12.0, 93.0])
+    if kind == "view":
+        scores = owner[::2]
+    elif kind == "integer":
+        scores = np.array([0, 10, 12])
+    elif kind == "list":
+        scores = [0.0, 10.0, 12.0]
+    else:
+        scores = np.array([0.0, 10.0, 12.0])
+    if kind == "readonly":
+        scores.setflags(write=False)
+    before = np.asarray(scores).copy()
+    owner_before = owner.copy()
+    method = "code_with_intercept" if intercept else "code_without_intercept"
+    expected = getattr(Poly(before.tolist()), method)(["a", "b", "c"])
+    actual = getattr(Poly(scores), method)(["a", "b", "c"])
+    np.testing.assert_allclose(actual.matrix, expected.matrix)
+    assert actual.column_suffixes == expected.column_suffixes
+    np.testing.assert_array_equal(scores, before)
+    np.testing.assert_array_equal(owner, owner_before)
+
+
+@pytest.mark.parametrize("intercept", [True, False])
+@pytest.mark.parametrize("readonly", [True, False])
+def test_poly_scores_formula_input_unchanged(intercept, readonly):
+    scores = np.array([0.0, 10.0, 12.0])
+    before = scores.copy()
+    if readonly:
+        scores.setflags(write=False)
+    formula = "C(a, Poly(scores))" if intercept else "0 + C(a, Poly(scores))"
+    a = ["a", "b", "c", "a", "b", "c"]
+    actual = dmatrix(formula, {"a": a, "scores": scores})
+    expected = dmatrix(formula, {"a": a, "scores": before.tolist()})
+    np.testing.assert_allclose(actual, expected)
+    assert actual.design_info.column_names == expected.design_info.column_names
+    prediction = build_design_matrices([actual.design_info], {"a": ["c", "a", "b"]})[0]
+    np.testing.assert_allclose(prediction, actual[[2, 0, 1]])
+    np.testing.assert_array_equal(scores, before)