Add a bump-min-versions Claude skill

.claude/skills/upgrade-min-versions.md

@@ -0,0 +1,126 @@
+# Upgrade Minimum Dependency Versions
+
+This skill upgrades xarray's minimum dependency versions to match the policy defined in `ci/policy.yaml`.
+
+## When to Use
+
+Run this skill when:
+- Preparing a new release that bumps minimum versions
+- The policy file has been updated and manifests need to catch up
+- `pixi run policy-min-versions` shows `<` for any packages
+
+## Steps
+
+### 1. Check Current Policy Status
+
+```bash
+pixi run policy-min-versions
+```
+
+This shows a table with:
+- `=` means the version matches policy
+- `<` means the version needs to be upgraded
+- `>` means the version exceeds policy (usually fine)
+
+### 2. Update Version Pins in pixi.toml
+
+Update versions in these sections of `pixi.toml`:
+
+- `[feature.minimal.dependencies]` - for numpy, pandas
+- `[feature.minimum-scipy.dependencies]` - for scipy
+- `[feature.min-versions.dependencies]` - for all other pinned dependencies
+- `[package.run-dependencies]` - for packaging
+
+Example changes:
+```toml
+# [feature.minimal.dependencies]
+numpy = "2.0.*"  # was "1.26.*"
+
+# [feature.minimum-scipy.dependencies]
+scipy = "1.15.*"  # was "1.13.*"
+
+# [feature.min-versions.dependencies]
+zarr = "3.0.*"  # was "2.18.*"
+dask-core = "2025.1.*"  # was "2024.6.*"
+```
+
+### 3. Update pyproject.toml
+
+Update the corresponding versions in `pyproject.toml` optional dependencies:
+- `dependencies` - numpy, packaging, pandas
+- `[project.optional-dependencies]` - accel, io, viz sections
+
+### 4. Verify Lock File
+
+```bash
+pixi lock
+```
+
+This must pass. If it fails, there may be dependency conflicts to resolve.
+
+### 5. Verify Policy Compliance
+
+```bash
+pixi run policy-min-versions
+```
+
+All packages should now show `=`.
+
+### 6. Clean Up Obsolete Test Decorators
+
+In `xarray/tests/__init__.py`, remove any `has_*` / `requires_*` decorators for versions that are now guaranteed by the new minimums. For example:
+- If numpy >= 2.0 is now required, remove `has_numpy_2` / `requires_numpy_2`
+- If dask >= 2025.1 is now required, remove `has_dask_ge_2024_*` / `has_dask_ge_2025_1_0` / `has_dask_expr` / `requires_dask_expr`
+- If zarr >= 3.0 is now required, remove `has_zarr_v3` / `requires_zarr_v3`
+
+Then search for all usages of the deleted decorators and fix those files:
+
+```bash
+# Search for usages of deleted decorators
+rg "has_scipy_ge_|has_dask_ge_|has_zarr_v3[^_]|has_pandas_ge_|has_numpy_2|has_dask_expr|requires_dask_expr" xarray/tests/
+```
+
+For each file found:
+- Remove imports of deleted decorators
+- Remove `@requires_*` decorators that are always true
+- Remove `if has_*:` conditionals (keep only the true branch)
+- Remove `if not has_*:` conditionals (delete the code block)
+- Delete tests marked with `@pytest.mark.skipif(has_*, ...)` (always skipped)
+
+### 7. Update whats-new.rst
+
+Add a table to `doc/whats-new.rst` under "Breaking Changes" documenting the version changes (in alphabetical order):
+
+```rst
+Breaking Changes
+~~~~~~~~~~~~~~~~
+
+- The minimum versions of some dependencies were changed:
+
+  ===================== =========  =======
+   Package                    Old      New
+  ===================== =========  =======
+    boto3                  1.34     1.36
+    cartopy                0.23     0.24
+    dask                 2024.6   2025.1
+    distributed          2024.6   2025.1
+    h5netcdf                1.4      1.5
+    iris                    3.9     3.11
+    lxml                    5.1      5.3
+    matplotlib              3.8     3.10
+    numpy                   1.26      2.0
+    packaging              24.1     24.2
+    rasterio                1.3      1.4
+    scipy                   1.13     1.15
+    toolz                  0.12      1.0
+    zarr                   2.18      3.0
+  ===================== =========  =======
+```
+
+## Common Issues
+
+### Lock file conflicts
+If `pixi lock` fails, check for incompatible version combinations. Some packages (like h5py/hdf5) are noted in comments as prone to conflicts.
+
+### Test failures after cleanup
+After removing version guards, some tests may fail if they relied on version-specific behavior. Update the test logic to use only the new minimum version's behavior.