Drop layer/interface from the sectionate call (unbreaks #43 against sectionate 0.4.0rc2)

[hdrake]

Downstream half of hdrake/sectionate#7, which is now merged and released as sectionate 0.4.0rc2 (PyPI, 2026-08-04). Stacked on #43 (modernize-for-v0.7.0), where this call site lives, and targeted at it.

This is not a forward-looking change any more — it repairs #43 against the sectionate its own pyproject.toml resolves to. sectionate >= 0.4.0rc1 picks up 0.4.0rc2, which no longer has layer/interface parameters at all, so _convergence_along_section raises TypeError: convergent_transport() got an unexpected keyword argument 'layer'. #43's last green CI run (2026-08-02) installed 0.4.0rc1, published two days before rc2; re-run today it fails.

What changed upstream

convergent_transport dropped both arguments because neither was carrying information the call did not already have:

• the layer coordinate is a dimension of the transports themselves, so it reaches the output whether or not anyone names it;
• the interface coordinate cannot ride along (it is one point longer), but the grid names it: whichever axis registers the transports' layer dimension at its "center" position.

That axis is grid.axes["Z"] — precisely where this function read the two names from before handing them straight back:

zc = grid.axes["Z"].coords["center"]
zi = grid.axes["Z"].coords["outer"]
...
    layer=zc,
    interface=zi,

So the whole change is deleting them. Five lines, one function.

Why this is a no-op beyond unbreaking the call

• The layer coordinate arrives either way — it is a dimension of the utr/vtr this call passes, independent of which grid sectionate is handed.
• The interface coordinate never survived this call site. Two independent reasons, either alone sufficient: this call passes hgrid = horizontal_grid(grid), which has no vertical axis for sectionate to read one off; and even on a grid that does, the interface coordinate sits on its own dimension, so .rename(...)["conv_mass_transport"] drops it (extracting a DataArray keeps only coordinates whose dims are a subset of its own). The interface= name was the only thing putting it on the returned Dataset, and nothing ever read it from there.
• Nothing downstream wants it. transform_to_lambda takes both its target and its target_data from grid._ds.

Verification

Checked, not argued — #43 brings a test suite that covers this path, so this is testable in a way it was not on main:

• 57 passed, 0 skipped, 0 failed locally against sectionate's merged head, whose transports.py is byte-identical to the released 0.4.0rc2 wheel (verified by diff). That includes the three tests that run against the real CM4p25 file, and test_tiled_boundary_transport_obeys_divergence_theorem / test_fold_transport_methods_agree, which check the along-section transport numerically rather than just calling it.
• Reverting just this hunk against that same sectionate fails 4 tests with the TypeError above — so the tests genuinely exercise this call, and this is the hunk that fixes it.
• CI on this PR is green (3.11/3.12/3.13 + docs), installing sectionate 0.4.0rc2 from PyPI.
• On the sectionate side the equivalence is pinned directly: on a grid with a vertical axis, convergent_transport(grid, i_c, j_c) returns output xr.testing.assert_identical to the same call passing layer=grid.axes["Z"].coords["center"], interface=grid.axes["Z"].coords["outer"].

Not done: the two example notebooks were not re-executed. ECCO_AABW_watermass_budget.ipynb is the one that touches this path (along_section=True), and its data is a 2.7 GB / 12-file Zenodo record I did not pull; MOM6_water_mass_budgets.ipynb does not use along_section at all. Worth a run of the ECCO notebook before merging, though the suite already covers the multi-tile along-section path.

Note

There is an update-sectionate-API branch (1e23195 Updated sectionate API after breaking v0.4.0 changes) that predates #43 and covers the pre-modernization layout. If it also addresses this, close whichever of the two is redundant.

_{Drafted with AI assistance (Claude Code). I have read the diff and run the tests myself.}

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB