[echoix]

Description

Follow-up to OSGeo#7902 (grass.pygrass: Close RasterRow via context manager in region tests, already merged to main). Fixing that resource leak turned the fatal, whole-file-crashing FERROR: Input window changed while maps are open for read on Windows CI into a distinct, real bug: the small region that test_resampling_1, test_resampling_2, and test_resampling_to_numpy each set up isn't actually restricting what RasterRow/raster2numpy read back. Windows CI showed, for example, rast[0].tolist() coming back with roughly 100000 elements instead of the expected 10, and a numpy _ArrayMemoryError trying to allocate an 8,150,000-element array where an 8-element one was expected.

I built GRASS locally (CMake) to check whether this is a genuine Windows-only build difference or something reproducible here:

• test_raster_region.py run in isolation: passes (3/3).
• The entire 359-file gunittest suite run in sequence (matching how CI runs it, in case an earlier test file leaves region-related state behind that only this ordering exposes): still passes, and isn't among the files that failed in that full run.

Since it doesn't reproduce here even under matching full-suite conditions, this looks like a genuine platform difference in the compiled GRASS libraries (Windows CI here builds with MinGW via OSGeo4W) rather than a Python-level state leak, and isn't something I can responsibly guess a fix for without Windows access. Marked @xfail_windows on the three affected test methods instead, per the project's existing convention for exactly this situation (see python/grass/gunittest/utils.py), with a comment describing the symptom and noting it needs Windows access to debug further.

Motivation and context

OSGeo#7902 already turned the fatal crash into three individually visible failures. Without xfail, those would show as ordinary CI red without acknowledging they're a known, unreproduced, Windows-only issue distinct from the leak OSGeo#7902 fixed.

How has this been tested?

Built GRASS locally and ran the test both in isolation and as part of the full local gunittest suite (see above); confirmed it still passes on Linux with @xfail_windows added (a no-op there).

Types of changes

• Bug fix (non-breaking change which fixes an issue)

Checklist

• PR title provides summary of the changes and starts with one of the
  pre-defined prefixes
• My code follows the code style
  of this project.
• My change requires a change to the documentation.
• I have updated the documentation accordingly.
• I have added tests to cover my changes.

---

I used Claude Code to help investigate and draft this fix, under my review throughout. OSGeo#7902 already fixed the crash that was masking this bug; this PR doesn't fix the bug itself, only marks it xfail on Windows using the project's own existing mechanism for an unreproduced platform-specific failure, rather than claiming a fix I can't verify.