geotiff: refuse VRT non-nearest resample algorithms (#1751)

[brendancol]

Summary

• Closes geotiff: VRT ResampleAlg is parsed but ignored (silent nearest-neighbour fallback) #1751. ComplexSource parses <ResampleAlg> from VRT XML (_vrt.py:356-360) but read_vrt always calls _resample_nearest at the placement step (_vrt.py:798-804), so a VRT declaring Bilinear / Cubic / CubicSpline / Lanczos / Average / Mode silently received nearest-sampled pixels mislabelled as the requested algorithm -- quietly wrong analytics.
• Raise NotImplementedError at the resample call site when src.resample_alg is anything other than nearest. The case-insensitive accept list is '', None, Nearest, NearestNeighbour, NearestNeighbor, NEAR.
• The check sits inside the needs_resample branch rather than at parse time so a ComplexSource declaring Bilinear with matching SrcRect/DstRect sizes is still accepted -- no resample kernel runs, so no mislabelled output.
• Implementing the higher-quality kernels is out of scope for this PR; the goal is to stop the silent wrong-numbers behaviour. Error message names the requested algorithm and points at issue geotiff: VRT ResampleAlg is parsed but ignored (silent nearest-neighbour fallback) #1751 so a user can re-export with Nearest or wait for follow-up work.

Test plan

• New test_vrt_resample_alg_1751.py (16 cases): each of Bilinear/Cubic/CubicSpline/Lanczos/Average/Mode raises with the algorithm name and 1751 in the message; case-insensitive bilinear also raises; Nearest / NearestNeighbour / NearestNeighbor / NEAR / nearest / NEAREST / empty / absent <ResampleAlg> all round-trip; Bilinear at matching SrcRect/DstRect sizes does NOT raise (pins down the resample-site placement).
• Confirmed all 16 tests fail on main without the source change and pass with it.
• pytest xrspatial/geotiff/tests/ -q (1992 pass, 3 skipped; the GPU test_predictor2_big_endian_gpu_1517 and test_no_georef_windowed_coords_1710 failures are pre-existing environment issues, identical on main).

[Copilot]

Pull request overview

Prevents read_vrt from silently returning nearest-neighbour sampled pixels when a VRT ComplexSource declares a non-nearest <ResampleAlg> and resampling is actually required, addressing issue #1751.

Changes:

• Add a resample-algorithm support check that raises NotImplementedError for non-nearest algorithms when SrcRect/DstRect sizes differ.
• Define and use a normalized allowlist of “nearest” algorithm variants (Nearest*, NEAR, empty/absent).
• Add regression tests covering unsupported algorithms, case-insensitivity, nearest variants, missing <ResampleAlg>, and the “no resample needed” case.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
xrspatial/geotiff/_vrt.py	Refuses non-nearest <ResampleAlg> at the resample call site to avoid silently wrong output.
xrspatial/geotiff/tests/test_vrt_resample_alg_1751.py	Adds regression coverage for supported/unsupported <ResampleAlg> values and the conditional behavior.

Comments suppressed due to low confidence (1)

xrspatial/geotiff/_vrt.py:364

• This comment says the resample site 'refuses the read for any other declared algorithm', but in the implementation the refusal only happens when a resample is actually needed (SrcRect/DstRect sizes differ). Please tweak the wording to reflect the conditional behavior so the docs match the code path.

                # ``<ResampleAlg>`` records the requested resampler for
                # the placement step.  ``read_vrt`` only implements
                # nearest-neighbour today, so the resample site refuses
                # the read for any other declared algorithm rather than
                # returning silently-mislabelled pixels.  See issues
                # #1694 and #1751.
                resample_alg = _text(src_elem, 'ResampleAlg')

---

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

[brendancol]

Clarified both comments in 4bf143f. They now state that the NotImplementedError fires only when SrcRect and DstRect sizes differ (resampling actually required), and that a non-nearest alg with matching rect sizes is a no-op that passes through.