Cover degenerate shape + NaN/Inf reads on GPU and dask geotiff backends#1630
Merged
brendancol merged 2 commits intoMay 11, 2026
Merged
Conversation
Backend coverage gap for 1x1, 1xN, Nx1 rasters and for all-NaN, Inf, and NaN-sentinel inputs on non-eager backends. The eager numpy path covers these in test_edge_cases; the GPU, dask+numpy, and dask+cupy paths went unexercised. Adds 23 tests pinning the contract. Cat 3 HIGH (geometric edges on non-eager backends): - 1x1 reads on dask+numpy, GPU, dask+cupy - 1xN single-row reads on the same three backends - Nx1 single-column reads on the same three backends - 1x1 / 1xN / Nx1 writes through write_geotiff_gpu Cat 2 MEDIUM (NaN / Inf / nodata edge cases): - All-NaN raster reads on GPU + dask+cupy - Inf / -Inf reads on every non-eager backend - NaN sentinel mask on the float dask read path, including a sentinel block that straddles a chunk boundary
Contributor
There was a problem hiding this comment.
Pull request overview
Adds targeted GeoTIFF backend regression coverage for degenerate raster shapes (1x1 / 1xN / Nx1) and special float values (NaN/Inf) across the non-eager read/write backends (GPU, dask+numpy, dask+cupy), closing previously unexercised edge cases without changing any production code.
Changes:
- Add a new test module exercising degenerate-shape reads across dask+numpy, GPU, and dask+cupy, plus degenerate-shape GPU-writer round-trips.
- Add backend coverage for all-NaN and +/-Inf reads, and for finite nodata sentinels being masked to NaN on the dask read path.
- Update the sweep tracking state CSV to record this audit pass and covered categories.
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.
| File | Description |
|---|---|
xrspatial/geotiff/tests/test_degenerate_shapes_backends_2026_05_11.py |
Adds 23 regression tests covering degenerate shapes and NaN/Inf behavior across non-eager GeoTIFF backends (incl. GPU writer). |
.claude/sweep-test-coverage-state.csv |
Updates audit tracking notes/state for the geotiff module sweep pass. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+15
to
+16
| * 1x1 and 1xN writes through ``write_geotiff_gpu`` (Cat 3 HIGH for | ||
| the GPU writer's degenerate-shape path). |
Contributor
Author
There was a problem hiding this comment.
Fixed in 8147b35: module docstring bullet now reads "1x1, 1xN, and Nx1 writes through write_geotiff_gpu" to match the actual GPU-writer coverage in the file.
Comment on lines
+217
to
+222
| """``write_geotiff_gpu`` must accept 1-pixel and 1-row inputs. | ||
|
|
||
| The GPU writer's tile-encoding path uses an internal grid sizing | ||
| helper that fell back to host code for shapes smaller than the | ||
| default tile. The fallback exists but had no regression test that | ||
| would catch a future "fast-path only" refactor. |
Contributor
Author
There was a problem hiding this comment.
Fixed in 8147b35: TestGpuWriterDegenerateShapes docstring now says "must accept 1-pixel, 1-row, and 1-column inputs", and the section header above the class also lists Nx1.
Comment on lines
+352
to
+397
| class TestNanSentinelDaskRead: | ||
| """Float raster with ``nodata=NaN`` sentinel reads consistently | ||
| across backends. | ||
|
|
||
| The integer-sentinel equivalent is pinned by issue #1597. The | ||
| float path has no such per-chunk dtype divergence (the input is | ||
| already float), but the dask graph still has to forward the | ||
| sentinel substitution. A regression in the float branch of | ||
| ``_delayed_read_window`` would silently break this. | ||
| """ | ||
|
|
||
| @pytest.fixture | ||
| def nan_sentinel_path(self, tmp_path): | ||
| arr = np.arange(64, dtype=np.float32).reshape(8, 8) | ||
| arr[2:4, 2:4] = -9999.0 | ||
| arr[6, 0] = -9999.0 | ||
| p = tmp_path / "nan_sentinel_float.tif" | ||
| to_geotiff(arr, str(p), nodata=-9999.0) | ||
| return str(p), arr | ||
|
|
||
| def test_eager_path_baseline(self, nan_sentinel_path): | ||
| """Baseline: eager path replaces the sentinel with NaN.""" | ||
| path, _ = nan_sentinel_path | ||
| result = open_geotiff(path) | ||
| assert np.isnan(result.values[2, 2]) | ||
| assert np.isnan(result.values[6, 0]) | ||
| assert result.values[0, 0] == 0.0 # non-sentinel survives | ||
|
|
||
| def test_dask_numpy_matches_eager(self, nan_sentinel_path): | ||
| """dask compute reproduces the eager mask exactly.""" | ||
| path, _ = nan_sentinel_path | ||
| eager = open_geotiff(path) | ||
| dk = open_geotiff(path, chunks=4).compute() | ||
| np.testing.assert_array_equal(np.isnan(dk.values), np.isnan(eager.values)) | ||
| finite = ~np.isnan(eager.values) | ||
| np.testing.assert_array_equal(dk.values[finite], eager.values[finite]) | ||
|
|
||
| def test_dask_numpy_chunks_smaller_than_sentinel_block(self, nan_sentinel_path): | ||
| """Sentinels split across two chunks still mask correctly. | ||
|
|
||
| The 2x2 sentinel block at rows 2-3 cols 2-3 lands in a single | ||
| chunk for chunks=4 (rows 0-3) but straddles a chunk boundary | ||
| for chunks=2 (rows 2-3 split between chunks 1 and 2). This | ||
| exercises the per-block sentinel comparison. | ||
| """ | ||
| path, _ = nan_sentinel_path |
Contributor
Author
There was a problem hiding this comment.
Fixed in 8147b35: class docstring now reads "Float raster with a finite nodata sentinel (-9999.0) is masked to NaN consistently across backends on read", matching what the fixture actually writes.
- Module docstring: list Nx1 GPU-writer coverage that was already tested - Class docstring (DegenerateGPUWriter): mention 1-column inputs - NaN-sentinel docstring: clarify it's a finite sentinel masked to NaN Tests unchanged; docstrings only.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
and NaN-sentinel inputs end-to-end. The GPU, dask+numpy, and
dask+cupy backends went unexercised for the same inputs.
test_degenerate_shapes_backends_2026_05_11.pypinning the contract across every non-eager backend.
Cat 3 HIGH (geometric edges):
write_geotiff_gpuCat 2 MEDIUM (NaN / Inf / nodata):
sentinel block straddling a chunk boundary
No source changes; tests only. Test-coverage gap sweep 2026-05-11
pass 5 against the geotiff module.
Test plan
pytest xrspatial/geotiff/tests/test_degenerate_shapes_backends_2026_05_11.py(23 pass on a GPU host)test_edge_cases.py,test_attrs_parity_1548.py,test_dask_int_nodata_chunks_1597.py,test_gpu_nodata_1542.py,test_dask_cupy_combined.py).