fix: full scanpy 1.13 compatibility layer (#760)

Author: timtreis

src/spatialdata_plot/pl/_color.py

@@ -28,7 +28,6 @@
 from numpy.random import default_rng
 from pandas.api.types import CategoricalDtype, is_bool_dtype, is_numeric_dtype, is_string_dtype
 from pandas.core.arrays.categorical import Categorical
-from scanpy.plotting._utils import add_colors_for_categorical_sample_annotation
 from skimage.color import label2rgb
 from skimage.morphology import erosion, footprint_rectangle
 from skimage.util import map_array
@@ -42,7 +41,12 @@
 )
 
 from spatialdata_plot._logging import logger
-from spatialdata_plot.pl._scanpy_palettes import default_20, default_28, default_102
+from spatialdata_plot.pl._scanpy_compat import (
+    add_colors_for_categorical_sample_annotation,
+    default_20,
+    default_28,
+    default_102,
+)
 from spatialdata_plot.pl.render_params import (
     CmapParams,
     Color,

src/spatialdata_plot/pl/_palette.py

@@ -21,7 +21,7 @@
 from matplotlib.colors import ListedColormap, to_hex, to_rgb
 from matplotlib.pyplot import colormaps as mpl_colormaps
 
-from spatialdata_plot.pl._scanpy_palettes import default_20, default_28, default_102
+from spatialdata_plot.pl._scanpy_compat import default_20, default_28, default_102
 
 if TYPE_CHECKING:
     import spatialdata as sd

src/spatialdata_plot/pl/_scanpy_compat.py

@@ -0,0 +1,52 @@
+"""Version-tolerant access to scanpy internals used by spatialdata-plot.
+
+scanpy 1.13 relocated the default palettes and several private plotting helpers from
+``scanpy.plotting.{palettes,_tools,_utils}`` to ``scanpy.plotting.legacy.*``, and dropped the
+``settings._vector_friendly`` flag. The values and behaviour are unchanged, so we import from
+whichever path the installed scanpy exposes and re-export from a single place. This keeps
+spatialdata-plot working on scanpy both < and >= 1.13 and confines the reliance on scanpy
+internals to one module.
+
+The fallbacks catch ``ModuleNotFoundError`` (not the broader ``ImportError``) so that a legacy
+module which exists but fails to import for an unrelated reason surfaces instead of being masked
+by a silent fall-back to the old path.
+"""
+
+from scanpy import settings as _sc_settings
+
+try:  # scanpy >= 1.13
+    from scanpy.plotting.legacy.palettes import default_20, default_28, default_102
+except ModuleNotFoundError:  # scanpy < 1.13
+    from scanpy.plotting.palettes import default_20, default_28, default_102
+
+try:  # scanpy >= 1.13
+    from scanpy.plotting.legacy._tools.scatterplots import _add_categorical_legend
+except ModuleNotFoundError:  # scanpy < 1.13
+    from scanpy.plotting._tools.scatterplots import _add_categorical_legend
+
+try:  # scanpy >= 1.13
+    from scanpy.plotting.legacy._utils import add_colors_for_categorical_sample_annotation
+except ModuleNotFoundError:  # scanpy < 1.13
+    from scanpy.plotting._utils import add_colors_for_categorical_sample_annotation
+
+
+def vector_friendly() -> bool:
+    """Scanpy's rasterize-for-vector-output flag, read dynamically.
+
+    Controls whether scatter/image artists are rasterized (so vector output stays small). scanpy
+    1.13 removed the ``settings._vector_friendly`` attribute, so on scanpy >= 1.13 this always
+    returns ``False`` (scanpy's unconfigured default): users who had enabled it via
+    ``sc.set_figure_params(vector_friendly=True)`` lose rasterization there — unavoidable, as the
+    flag no longer exists. On older scanpy the configured value is still honoured.
+    """
+    return bool(getattr(_sc_settings, "_vector_friendly", False))
+
+
+__all__ = [
+    "_add_categorical_legend",
+    "add_colors_for_categorical_sample_annotation",
+    "default_20",
+    "default_28",
+    "default_102",
+    "vector_friendly",
+]

src/spatialdata_plot/pl/_scanpy_palettes.py

@@ -20,8 +20,6 @@
 from matplotlib import patheffects
 from matplotlib.cm import ScalarMappable
 from matplotlib.colors import BoundaryNorm, Colormap, ListedColormap, Normalize, to_rgba_array
-from scanpy._settings import settings as sc_settings
-from scanpy.plotting._tools.scatterplots import _add_categorical_legend
 from spatialdata import get_extent, get_values
 from spatialdata.models import PointsModel, ShapesModel, get_table_keys
 from spatialdata.transformations import set_transformation
@@ -60,6 +58,7 @@
     _scale_geometries,
     _validate_polygons,
 )
+from spatialdata_plot.pl._scanpy_compat import _add_categorical_legend, vector_friendly
 from spatialdata_plot.pl._validate import (
     _check_obs_var_shadow,
 )
@@ -933,7 +932,7 @@ def _draw_centroids(xy: np.ndarray, radius: float | None = None) -> None:
                 s=render_params.scale,
                 c=np.array(["white"]),  # hack, will be invisible bc fill_alpha=0
                 render_params=render_params,
-                rasterized=sc_settings._vector_friendly,
+                rasterized=vector_friendly(),
                 cmap=None,
                 fill_alpha=0.0,
                 outline_alpha=render_params.outline_alpha[0],
@@ -949,7 +948,7 @@ def _draw_centroids(xy: np.ndarray, radius: float | None = None) -> None:
                 s=render_params.scale,
                 c=np.array(["white"]),  # hack, will be invisible bc fill_alpha=0
                 render_params=render_params,
-                rasterized=sc_settings._vector_friendly,
+                rasterized=vector_friendly(),
                 cmap=None,
                 fill_alpha=0.0,
                 outline_alpha=render_params.outline_alpha[0],
@@ -966,7 +965,7 @@ def _draw_centroids(xy: np.ndarray, radius: float | None = None) -> None:
                 s=render_params.scale,
                 c=np.array(["white"]),  # hack, will be invisible bc fill_alpha=0
                 render_params=render_params,
-                rasterized=sc_settings._vector_friendly,
+                rasterized=vector_friendly(),
                 cmap=None,
                 fill_alpha=0.0,
                 outline_alpha=render_params.outline_alpha[1],
@@ -984,7 +983,7 @@ def _draw_centroids(xy: np.ndarray, radius: float | None = None) -> None:
             c=color_spec.to_rgba(render_params.cmap_params),
             prebuilt_paths=prebuilt_paths,
             render_params=render_params,
-            rasterized=sc_settings._vector_friendly,
+            rasterized=vector_friendly(),
             cmap=render_params.cmap_params.cmap,
             fill_alpha=render_params.fill_alpha,
             outline_alpha=0.0,
@@ -1072,7 +1071,7 @@ def _scatter_points(
         # `size` at high dpi) only that ring survives, rendering markers as hollow outlines.
         # linewidths=0 keeps them solid fills whose radius honours `size`.
         linewidths=0,
-        rasterized=sc_settings._vector_friendly,
+        rasterized=vector_friendly(),
         alpha=alpha,
         transform=trans_data,
         zorder=zorder,

src/spatialdata_plot/pl/render.py

@@ -31,7 +31,6 @@
 from pandas.api.types import CategoricalDtype, is_numeric_dtype
 from pandas.core.arrays.categorical import Categorical
 from scanpy import settings
-from scanpy.plotting._tools.scatterplots import _add_categorical_legend
 from spatialdata import (
     SpatialData,
     get_element_annotators,
@@ -56,7 +55,7 @@
 from xarray import DataArray, DataTree
 
 from spatialdata_plot._logging import logger
-from spatialdata_plot.pl._scanpy_palettes import default_102
+from spatialdata_plot.pl._scanpy_compat import _add_categorical_legend, default_102
 from spatialdata_plot.pl.render_params import (
     Color,
     ColorbarSpec,

src/spatialdata_plot/pl/utils.py

@@ -197,7 +197,9 @@ def test_default_returns_dict(self, clustered_sdata: SpatialData):
         assert all(v.startswith("#") for v in result.values())
 
     def test_default_matches_scanpy_order(self, clustered_sdata: SpatialData):
-        from scanpy.plotting.palettes import default_20
+        # Import via the compat shim so this resolves across scanpy versions (1.13 moved the
+        # palettes to scanpy.plotting.legacy.palettes); it's the same source make_palette uses.
+        from spatialdata_plot.pl._scanpy_compat import default_20
 
         result = make_palette_from_data(clustered_sdata, "cells", "cell_type")
         for i, cat in enumerate(sorted(result.keys())):