feat: Add big endian fixtures with predictor 2 and predictor 3

[james-willis]

Adds two big endian (MM) fixtures, as requested in developmentseed/deck.gl-raster#687.

Fixture	Exercises
uint16_1band_lzw_block128_predictor2_big_endian	Samples must be swapped to native order before undoing horizontal differencing, which adds neighbouring samples as integers.
float32_1band_deflate_block64_predictor3_big_endian	Predictor 3 byte planes are most significant byte first whatever the file's byte order, so these samples must not be swapped.

Both are 128×128 with 64×64 tiles and one overview, and pass rio cogeo validate.

How they're made

GDAL's COG driver has no ENDIANNESS option, so a new rewrite_big_endian helper in write_utils.py copies a little endian COG through the GTiff driver with COPY_SRC_OVERVIEWS=YES and ENDIANNESS=BIG, which keeps the COG layout. The uint16 fixture is rewritten from uint16_1band_lzw_block128_predictor2; the float32 one from a np.linspace COG built the same way as float32_1band_lerc_block32.

Tiling, compression and predictor are copied from the source, which relies on #74 (write_cog now writes the predictor into the COG).

Tests

tests/test_big_endian.py checks, for each fixture, that the file starts with MM, that every IFD carries the expected predictor (via tifffile), and that every level matches its little endian source as read by rasterio. pyproject.toml gains pythonpath = ["."] so the test can import the generators. pixi run check and pixi run test pass locally, and regenerating is byte-for-byte stable.

Downstream check

With deck.gl-raster's fixtures submodule pointed at this branch and both fixtures added to integration-rasterio.test.ts:

• With fix(geotiff): decode big endian tiles in the platform's byte order deck.gl-raster#687: all 19 integration cases pass.
• On main: the uint16 predictor 2 fixture fails tile pixel data matches, and the predictor 3 fixture passes, as expected.

[Copilot]

Copilot review overview

🟢 Approval recommended

The fixtures, generation path, metadata, and regression coverage consistently satisfy the stated endian and predictor requirements.

Review effort: Balanced
Findings: None

What changed in this PR

Adds big-endian predictor fixtures to validate downstream GeoTIFF decoding behavior.

Changes:

• Adds Predictor 2 and Predictor 3 big-endian COG fixtures.
• Adds reusable endian-rewriting logic and fixture validation tests.
• Configures pytest imports for generator modules.

File	Description
pyproject.toml	Adds the repository root to pytest’s Python path.
rasterio_generated/​write_utils.py	Adds the big-endian COG rewrite helper.
tests/​test_big_endian.py	Validates byte order, predictors, and pixels.
rasterio_generated/​fixtures/​uint16_1band_lzw_block128_predictor2_big_endian.py	Generates the Predictor 2 fixture.
rasterio_generated/​fixtures/​uint16_1band_lzw_block128_predictor2_big_endian.tif	Provides the generated uint16 fixture.
rasterio_generated/​fixtures/​uint16_1band_lzw_block128_predictor2_big_endian_info.md	Records uint16 fixture metadata.
rasterio_generated/​fixtures/​float32_1band_deflate_block64_predictor3_big_endian.py	Generates the Predictor 3 fixture.
rasterio_generated/​fixtures/​float32_1band_deflate_block64_predictor3_big_endian.tif	Provides the generated float32 fixture.
rasterio_generated/​fixtures/​float32_1band_deflate_block64_predictor3_big_endian_info.md	Records float32 fixture metadata.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.