geotiff: dask streaming write coverage for 1x1/1xN/Nx1 + NaN/Inf

[brendancol]

Summary

Test coverage gap sweep pass 14 (2026-05-15). Closes Cat 3 HIGH plus
Cat 2 HIGH/MEDIUM and Cat 4 MEDIUM gaps on the dask streaming write
path (to_geotiff with a dask-backed DataArray, #1084).

Before this PR test_streaming_write.py covered a 100x100 raster with
an embedded NaN block plus a 2x2 test_small_raster, but had nothing
for:

• 1x1, 1xN, or Nx1 dask streaming writes. The tile-row segmenter
  (Stream tile writes per dask chunk segment to bound peak memory in to_geotiff #1485) and the chunk-merge math on a 1-pixel-tall or 1-pixel-wide
  raster are reachable only via this path.
• All-NaN dask write where every chunk the writer streams is entirely
  NaN and must mask to the nodata sentinel.
• +Inf / -Inf round-trip on the dask streaming path (Inf is a valid
  IEEE-754 float and must round-trip bit-exactly; only NaN is treated
  as nodata).
• predictor=3 (TIFF float predictor) through the dask streaming path.
  The existing test_predictor case in test_streaming_write.py
  covers predictor=True (=2) only.

Pass 5's test_degenerate_shapes_backends_2026_05_11.py covered the
GPU writer's degenerate shapes; a regression in the dask streaming
writer would still ship undetected.

The new file
xrspatial/geotiff/tests/test_dask_streaming_write_degenerate_2026_05_15.py
adds 16 tests:

• 1x1 chunk-matches-shape + nodata-attr round-trip + uint16
• 1xN single chunk + chunks-split-columns + wide-segmented-by-buffer
  (streaming_buffer_bytes=1 forces the Stream tile writes per dask chunk segment to bound peak memory in to_geotiff #1485 segmenter)
• Nx1 single chunk + chunks-split-rows
• All-NaN with finite sentinel + all-NaN without sentinel
• Mixed NaN / +Inf / -Inf preserving Inf bit-exact + sentinel masking
  NaN only
• All-+Inf and all--Inf
• predictor=3 round-trip on float32 + float64 plus int-dtype
  ValueError

Test plan

• pytest xrspatial/geotiff/tests/test_dask_streaming_write_degenerate_2026_05_15.py -x -q -- 16 passed
• CSV state row updated (pass 14 note appended)

[Copilot]

Pull request overview

Adds dask streaming GeoTIFF write regression coverage for degenerate raster shapes, NaN/Inf handling, and predictor-related cases, plus updates the sweep coverage state.

Changes:

• Adds a new geotiff test module for 1x1, 1xN, Nx1, all-NaN, Inf, and predictor=3 dask-backed writes.
• Updates .claude/sweep-test-coverage-state.csv with the pass 14 coverage note.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
xrspatial/geotiff/tests/test_dask_streaming_write_degenerate_2026_05_15.py	Adds dask streaming write regression tests for degenerate shapes, NaN/Inf, and predictor=3 cases.
.claude/sweep-test-coverage-state.csv	Records the latest geotiff coverage sweep status and notes.

Comments suppressed due to low confidence (3)

xrspatial/geotiff/tests/test_dask_streaming_write_degenerate_2026_05_15.py:194

• These read-back checks do not prove that only NaN was masked to the nodata sentinel during the streaming write. Because open_geotiff returns NaN both for raw NaN pixels and for sentinel pixels after masking, the test would still pass if the writer stopped replacing NaNs with -9999.0; add a raw on-disk/sentinel assertion for the NaN positions while keeping the Inf round-trip checks.

        to_geotiff(da, path)
        result = open_geotiff(path)
        # NaN positions round-trip to NaN.
        assert np.isnan(result.values[0, 1])
        assert np.isnan(result.values[2, 2])

xrspatial/geotiff/tests/test_dask_streaming_write_degenerate_2026_05_15.py:251

• The added tests only assert value round-trip, which would still pass if predictor=3 were ignored because deflate/zstd compression remains lossless without a predictor. Existing predictor tests assert the TIFF Predictor tag directly; this streaming coverage should similarly verify that tag (or another predictor-specific side effect) to catch a dropped predictor thread-through.

        to_geotiff(da, path, compression='deflate', predictor=3,
                   tile_size=16)

xrspatial/geotiff/tests/test_dask_streaming_write_degenerate_2026_05_15.py:261

• This float64 case has the same blind spot as the float32 predictor test: a lossless write without applying or tagging Predictor=3 would round-trip exactly, so the assertion does not exercise the predictor-specific streaming branch. Please verify the Predictor tag or another predictor-specific observable here as well.

        to_geotiff(da, path, compression='deflate', predictor=3,
                   tile_size=16)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.