fix(xferfcn): normalize real zpk coefficients

marko1olo · codex · marko1olo · commit 483688e65a29 · 2026-06-07T12:21:17.000+04:00
Normalize floating coefficient arrays after zpk2tf so complex dtype zero/pole inputs that produce real polynomial coefficients do not leave float32 or tiny imaginary residue in TransferFunction internals. Continue rejecting actual complex transfer function coefficients.

Co-authored-by: OpenAI Codex &lt;codex@openai.com&gt;
diff --git a/control/tests/xferfcn_test.py b/control/tests/xferfcn_test.py
@@ -52,6 +52,9 @@ def test_constructor_bad_input_type(self):
 
         with pytest.raises(TypeError, match="unsupported data type"):
             ct.tf([1j], [1, 2, 3])
+        for dtype in [np.complex64, np.complex128]:
+            with pytest.raises(TypeError, match="unsupported data type"):
+                ct.tf(np.array([1 + 1j], dtype=dtype), [1, 2, 3])
 
         # good input
         TransferFunction([[[0, 1], [2, 3]],
@@ -1550,6 +1553,19 @@ def test_zpk(zeros, poles, gain, args, kwargs):
     if kwargs.get('name'):
         assert sys.name == kwargs.get('name')
 
+
+@pytest.mark.parametrize("dtype", [np.complex64, np.complex128])
+def test_zpk_complex_dtype_real_coefficients(dtype):
+    zeros = np.array([1 + 1j, 1 - 1j], dtype=dtype)
+    poles = np.array([-1 + 1j, -1 - 1j], dtype=dtype)
+
+    sys = ct.zpk(zeros, poles, gain=1, dt=0)
+
+    assert sys.num_array[0, 0].dtype == float
+    assert sys.den_array[0, 0].dtype == float
+    assert "float32" not in str(sys)
+
+
 @pytest.mark.parametrize("create, args, kwargs, convert", [
     (StateSpace, ([-1], [1], [1], [0]), {}, ss2tf),
     (StateSpace, ([-1], [1], [1], [0]), {}, ss),
diff --git a/control/xferfcn.py b/control/xferfcn.py
@@ -1960,7 +1960,7 @@ def _clean_part(data, name="<unknown>"):
     if isinstance(data, np.ndarray) and data.ndim == 2 and \
        data.dtype == object and isinstance(data[0, 0], np.ndarray):
         # Data is already in the right format
-        return data
+        out = data
     elif isinstance(data, ndarray) and data.ndim == 3 and \
           isinstance(data[0, 0, 0], valid_types):
         out = np.empty(data.shape[0:2], dtype=np.ndarray)
@@ -1995,15 +1995,26 @@ def _clean_part(data, name="<unknown>"):
             "The numerator and denominator inputs must be scalars or vectors "
             "(for\nSISO), or lists of lists of vectors (for SISO or MIMO).")
 
-    # Check for coefficients that are ints and convert to floats
+    # Check for real numeric coefficients and normalize floating arrays
     for i in range(out.shape[0]):
         for j in range(out.shape[1]):
-            for k in range(len(out[i, j])):
-                if isinstance(out[i, j][k], (int, np.integer)):
-                    out[i, j][k] = float(out[i, j][k])
-                elif isinstance(out[i, j][k], unsupported_types):
+            coefficients = np.asarray(out[i, j])
+            convert_to_float = np.issubdtype(coefficients.dtype, np.floating)
+            if np.iscomplexobj(coefficients):
+                real_coefficients = coefficients.real
+                zero_tol = 1000 * np.finfo(float).eps * max(
+                    1, np.max(np.abs(real_coefficients)))
+                if np.any(np.abs(coefficients.imag) > zero_tol):
                     raise TypeError(
-                        f"unsupported data type: {type(out[i, j][k])}")
+                        f"unsupported data type: {type(coefficients.flat[0])}")
+                coefficients = real_coefficients
+                convert_to_float = True
+            for k in range(len(coefficients)):
+                if isinstance(coefficients[k], unsupported_types):
+                    raise TypeError(
+                        f"unsupported data type: {type(coefficients[k])}")
+            out[i, j] = np.asarray(
+                coefficients, dtype=float) if convert_to_float else coefficients
     return out
 
 
