Reject float-dtype labels with clear error in render_labels

src/spatialdata_plot/pl/render.py

@@ -1621,6 +1621,13 @@ def _render_labels(
             is_label=True,
         )
 
+    if np.issubdtype(label.dtype, np.floating):
+        raise ValueError(
+            f"Label element '{element}' has dtype {label.dtype}. Label arrays must use an "
+            f"integer dtype (e.g. int32 or uint16). Cast before plotting, e.g.:\n"
+            f"    sdata['{element}'] = sdata['{element}'].astype('int32')"
+        )
+
     # rasterize spatial image if necessary to speed up performance
     if rasterize:
         label = _rasterize_if_necessary(

tests/pl/test_render_labels.py

@@ -473,3 +473,20 @@ def track(x):
     plt.close(fig)
 
     assert called, "transfunc was not called for continuous labels data"
+
+
+@pytest.mark.parametrize("dtype", [np.float16, np.float32, np.float64])
+def test_render_labels_rejects_float_dtype(dtype):
+    # Regression test for #606: float-dtype labels must raise a clear
+    # ValueError naming the element and dtype, not a cryptic skimage TypeError.
+    arr = np.zeros((20, 20), dtype=dtype)
+    arr[3:8, 3:8] = 1
+    arr[12:17, 12:17] = 2
+    sdata = SpatialData(labels={"lbl": Labels2DModel.parse(arr, dims=["y", "x"])})
+
+    fig, ax = plt.subplots()
+    try:
+        with pytest.raises(ValueError, match=r"Label element 'lbl'.*integer dtype"):
+            sdata.pl.render_labels("lbl").pl.show(ax=ax)
+    finally:
+        plt.close(fig)