Remove stale external-overview skips on dask/fsspec/GPU golden-corpus backends

xrspatial/geotiff/tests/golden_corpus/test_dask_gpu.py

@@ -82,11 +82,12 @@
 # xrspatial reader. The base-IFD parity check still runs; the
 # overview-level loop driven by ``candidate_factory`` is skipped for
 # these ids. See the eager module for the rationale.
-_OVERVIEW_READER_GAPS: dict[str, str] = {
-    "overview_external_ovr_uint16": (
-        "External .ovr sidecar reader is not implemented in xrspatial."
-    ),
-}
+#
+# Empty now that the dask+GPU reader resolves the sibling ``.tif.ovr``
+# pyramid through the shared ``_sidecar`` discovery the eager numpy
+# backend uses, so the overview-level comparison runs for
+# ``overview_external_ovr_uint16`` on this backend too.
+_OVERVIEW_READER_GAPS: dict[str, str] = {}
 
 _INTENTIONAL_SKIPS: dict[str, str] = {
     "nodata_miniswhite_uint8": (

xrspatial/geotiff/tests/golden_corpus/test_dask_numpy.py

@@ -78,11 +78,12 @@
 # xrspatial reader. The base-IFD parity check still runs; the
 # overview-level loop driven by ``candidate_factory`` is skipped for
 # these ids. See the eager module for the rationale.
-_OVERVIEW_READER_GAPS: dict[str, str] = {
-    "overview_external_ovr_uint16": (
-        "External .ovr sidecar reader is not implemented in xrspatial."
-    ),
-}
+#
+# Empty now that the dask reader resolves the sibling ``.tif.ovr``
+# pyramid through the shared ``_sidecar`` discovery the eager numpy
+# backend uses, so the overview-level comparison runs for
+# ``overview_external_ovr_uint16`` on this backend too.
+_OVERVIEW_READER_GAPS: dict[str, str] = {}
 
 _INTENTIONAL_SKIPS: dict[str, str] = {
     "nodata_miniswhite_uint8": (

xrspatial/geotiff/tests/golden_corpus/test_fsspec.py

@@ -84,13 +84,7 @@
 # xrspatial reader. The base-IFD parity check still runs; the
 # overview-level loop driven by ``candidate_factory`` is skipped for
 # these ids. See the eager module for the rationale.
-_OVERVIEW_READER_GAPS: dict[str, str] = {
-    "overview_external_ovr_uint16": (
-        "External .ovr sidecar reader is not implemented in xrspatial. "
-        "The base IFD still parity-checks; the overview levels live in "
-        "a sibling .tif.ovr that the reader does not open."
-    ),
-}
+_OVERVIEW_READER_GAPS: dict[str, str] = {}
 
 _INTENTIONAL_SKIPS: dict[str, str] = {
     "nodata_miniswhite_uint8": (
@@ -152,13 +146,23 @@ def _serve_via_memory(payload: bytes, fixture_id: str) -> str:
     Returns the ``memory:///corpus/<fixture_id>.tif`` URL the reader
     should open. The three-slash form is required: ``memory://`` writes
     must use a path beginning with ``/`` or fsspec rejects them.
+
+    When the fixture ships a sibling ``.tif.ovr`` sidecar on disk, push
+    it under the matching ``.tif.ovr`` URL so the reader's fsspec
+    sidecar discovery (``_probe_fsspec`` -> ``load_sidecar``) can resolve
+    the external overview levels. Without it the memory filesystem holds
+    only the base ``.tif`` and ``overview_level>=1`` fails as out of
+    range, which is a test-harness gap rather than a reader limitation.
     """
     fs = fsspec.filesystem("memory")
     url = f"memory:///corpus/{fixture_id}.tif"
     # ``pipe`` is the idiomatic single-shot byte writer for fsspec
     # filesystems; ``test_cloud_read_byte_limit_1928.py`` uses the
     # same pattern.
     fs.pipe(f"/corpus/{fixture_id}.tif", payload)
+    sidecar_path = FIXTURES_DIR / f"{fixture_id}.tif.ovr"
+    if sidecar_path.exists():
+        fs.pipe(f"/corpus/{fixture_id}.tif.ovr", sidecar_path.read_bytes())
     return url
 
 

xrspatial/geotiff/tests/golden_corpus/test_gpu.py

@@ -84,11 +84,12 @@
 # xrspatial reader. The base-IFD parity check still runs; the
 # overview-level loop driven by ``candidate_factory`` is skipped for
 # these ids. See the eager module for the rationale.
-_OVERVIEW_READER_GAPS: dict[str, str] = {
-    "overview_external_ovr_uint16": (
-        "External .ovr sidecar reader is not implemented in xrspatial."
-    ),
-}
+#
+# Empty now that the GPU reader resolves the sibling ``.tif.ovr``
+# pyramid through the shared ``_sidecar`` discovery the eager numpy
+# backend uses, so the overview-level comparison runs for
+# ``overview_external_ovr_uint16`` on this backend too.
+_OVERVIEW_READER_GAPS: dict[str, str] = {}
 
 # Fixtures whose codec is not implemented on the GPU read path and
 # legitimately need to fall back to CPU. The test routes these through