Follow up items for the week of July 4, 2026

[espg]

🤖 from Claude

Sweep of closed PRs from the last 10 days (June 24 – July 4, 2026): #59, #70, #72, #76, #84, #85, #86, #87, #89, #91, #94, #95. Items from #59/#70/#72/#76/#84 were already surfaced by the prior sweep (#88); this issue covers new gaps from #85/#86/#87/#89/#91/#94/#95, plus a re-verification of #88's 10 carry-forward items against current main. Six of #88's items are now resolved and excluded: Morton __repr__ (#76), split_base_cells ordering + Python-wrapper decision (#84), mort2norm FORCE_PYTHON remnant (#49), CI versioning automation (#91/#95, tracked to completion via now-closed #92), and the stale benchmark scripts (removed entirely). Four carry-forwards are still open and are folded into the numbered list below. Issue #90 (on_minor_arc/node_straddles, from #87) and #96 (points= geo2mort, filed today by espg) are separately tracked and excluded here.

Items

(1) res2display() order cap still wrong above order 20 — carry-forward from #88 — PR #76

mortie/tools.py::res2display() still loops for res in range(20), so it silently returns nothing for any order above 19 even though the packed-u64 kernel supports order 29. Unchanged by any of the 7 new PRs.

(2) arro3-core CI leg for the pyarrow-free consumer was designed but never wired up — PR #94 (Refs #92/#93)

PR #94 built the Arrow C Data Interface specifically so zagg's pyarrow-free Lambda worker can consume morton_index (TestArro3Interop, currently skipif-gated and local-only — arro3-core is deliberately kept out of the test extra). A concrete arro3-no-pyarrow CI job was proposed with exact YAML in a comment on issue #92 (#92#issuecomment-4859163096), explicitly because the routine lacks the workflow OAuth scope needed to push .github/workflows/ changes itself. Issue #92 was subsequently closed by PR #95 — but #95 only fixed the changelog/version-carryover and codecov.yml; the CI leg was never added. The one real pyarrow-absent consumer this feature exists for currently has zero CI coverage.

(3) // floor-division Morton coarsening operator still not implemented — carry-forward from #88 — PR #72

No __floordiv__ exists anywhere in mortie/morton_index.py, mortie/prefix_trie.py, or src_rust. Still deferred, still no tracking issue of its own.

(4) Earthdata-gated notebooks not publicly Binder-runnable — carry-forward from #88 — PR #76

No PR in this window touched any notebook; the three velocity/grounding-line notebooks requiring authenticated earthaccess downloads are unchanged. CLAUDE.md §4 still calls for Binder-runnable notebooks with anonymously-reachable data.

(5) Dissolve winding-sign ambiguity near/over a hemisphere is documented but unguarded — PR #89

mortie/geometry.py::_dissolved_rings_py (and the Rust mirror src_rust/src/dissolve.rs::classify_and_split) classifies exterior-vs-hole rings via if sum(areas) < 0.0: reverse. Spherical signed area is only defined mod 4π, so the test is unreliable once a dissolved cover approaches/exceeds a hemisphere. PR #89's review (discussion_r3491006825) suggested either a guard/raise or a comment; only the comment ("this assumes the cover stays well under a hemisphere") was added — there's no runtime guard and no test near a hemisphere-sized cover.

(6) Clippy useless_conversion warnings still pending, now 31 (was 27) — carry-forward from #88 — PR #57

New #[pyfunction] bindings added in #86/#94 grew the pre-existing false-positive count from 27 to 31. Still no #[allow(clippy::useless_conversion)] anywhere in src_rust, and still flagged as "pre-existing, not introduced here" in each new PR's self-review rather than decided one way or the other.

(7) from_geometry(..., normalize=False) is silently a no-op for LineString/MultiLineString input — PR #89

The linear branch of mortie/geometry.py::from_geometry (~line 262) only rejects moc/tolerance/max_cells as polygon-only; normalize passes straight through with no effect and no error, since linestring_coverage has no such parameter. Raised in review (discussion_r3487712390); not fixed.

(8) Two non-blocking design questions from PR #85 remain unanswered — PR #85

(a) Whether moc_to_order's size guard should raise ValueError (current) or MemoryError; (b) whether the hard-raise threshold should stay pinned to the existing 1<<20 warn threshold or use a separate, higher line. Both were posed as "Questions for review" in the PR body with no reply visible in the thread and no later follow-up.

(9) PR #91's mktemp+mv vs. sed -i portability question — no explicit sign-off visible — PR #91

PR #91 asked for explicit confirmation of a deliberate deviation from sed -i (portability across GNU/BSD sed); the PR was merged with no visible reply. Effectively moot since it shipped, but noting it since the question was never explicitly closed out.

---

Priority / severity table

Priority	Item	Severity	Source PR
1	res2display() order cap wrong above order 20 — silent wrong output for orders 20–29	High	#76 (carry-forward)
2	arro3-core CI leg missing — the one real pyarrow-free consumer path has zero CI coverage	High	#94
3	// Morton coarsening operator missing — deferred API gap, still untracked	Medium	#72 (carry-forward)
4	Earthdata-gated notebooks not Binder-runnable — CLAUDE.md §4 compliance gap	Medium	#76 (carry-forward)
5	Dissolve winding-sign ambiguity near a hemisphere — unguarded correctness risk on large polar covers	Medium	#89
6	Clippy useless_conversion warnings, now 31 — decision still pending	Low	#57 (carry-forward)
7	from_geometry(normalize=False) silently ignored for linear input	Low	#89
8	Two unanswered design questions in moc_to_order's size guard	Low	#85
9	PR #91 portability question — no explicit sign-off, effectively moot	Low	#91

[espg]

🤖 from Claude

Triage of the nine items — grouped by what's already moving, what's a labeled-and-go fix, what needs a design call, and what's moot. Opening/labeling issues is side-effecting, so I've flagged the label I'd apply for each and am leaving the actual carve-up to your go-ahead.

Already tracked (no action needed here)

• (1) res2display order cap — now its own issue res2display() order cap still wrong above order 20 — carry-forward from #88 — PR #76 #99, which I planned this run (plan comment): extend the loop to order 29 + a km/m/cm unit ladder. waiting on your read of the three formatting questions there.
• (2) arro3-core CI leg — tracked as arro3-core CI leg for the pyarrow-free consumer was designed but never wired up #98. The blocker is real and not code: the routine lacks the workflow OAuth scope to push .github/workflows/, and the exact YAML is already drafted on #92's comment. This needs you (or a local agent with workflow scope) to commit the workflow file — I can't. Recommend either you paste the drafted job in, or confirm you want it left for a scoped local run.

Labeled-and-go (small, unambiguous — just need the label to activate)

• (7) from_geometry(normalize=False) silently ignored for LineString/MultiLineString. Two clean options: (a) raise ValueError("normalize is not supported for linear geometries") in the linear branch (mortie/geometry.py ~262), symmetric with how moc/tolerance/max_cells are already rejected there — my recommendation, since a silently-ignored kwarg is a latent footgun; or (b) document it as accepted-but-ignored. Tiny either way + a test. → small-fix if you want (a).
• (6) Clippy useless_conversion, now 31. Decision, not research: (a) add #[allow(clippy::useless_conversion)] at the crate root with a one-line comment explaining the pyo3/numpy ? idiom makes the conversion load-bearing (kills the noise, one line) — recommended; or (b) leave it and keep flagging "pre-existing" in every PR self-review. This has bounced across #34 §D cleanup + fmt/clippy sweep #57/Follow up items for the week of June 23, 2026 #73/Follow up items for the week of June 27, 2026 #88 unresolved; (a) closes it for good. → small-fix.
• (3) // floor-division coarsening operator. MortonIndex // n = coarsen-to-order, the deferred sibling of clip2order. Straightforward to bind (the Rust coarsen kernel already exists), but it's really the first bite of the deferred NEP-42 operator grammar you flagged on Version 1.0 Sweep (Follow up to new morton datatype) #48 (///%/+/* = coarsen/contains/parent/…). Fork: (a) ship just __floordiv__ now on the morton_index dtype as a standalone implement; or (b) hold it and design the whole operator grammar in one discuss so we don't freeze a partial grammar into 1.x. Given the 1.x freeze contract, I lean (b) — pinning one operator's semantics piecemeal is the kind of thing the freeze makes expensive to revise. Your call on which.

Needs a design call (won't touch blind)

• (5) Dissolve winding-sign ambiguity near/over a hemisphere. _dissolved_rings_py / classify_and_split use if sum(areas) < 0: reverse, and spherical signed area is only defined mod 4π, so the exterior/hole classification is unreliable once a dissolved cover approaches a hemisphere. Options: (a) add a runtime guard that raises when |sum(areas)| is within some margin of 2π (fail loud instead of silently mis-classifying) + a near-hemisphere test; (b) switch to a winding/containment test that doesn't rely on the sign of a mod-4π quantity (more work, removes the ambiguity entirely); (c) leave the comment-only status quo. Recommend (a) now (cheap, converts a silent wrong answer into a clear error) and file (b) as a follow-up if polar-scale covers become a real workload. Risk with (c): a large Antarctic/Arctic dissolve can silently invert exterior/hole. → discuss to pick, or small-fix if you're happy with (a) directly.
• (8) The two moc_to_order size-guard questions from small fixes 2026-06-27: densify size guard (#80) + morton_polygon determinism tests (#83) #85. (a) ValueError (current) vs MemoryError for the size guard — I'd keep ValueError (it's a bad-argument condition the caller controls, not an actual allocation failure; MemoryError implies the allocation was attempted); (b) hard-raise threshold pinned to the existing 1<<20 warn line vs a separate higher line — I'd use a separate, higher hard line (e.g. warn at 1<<20, raise at a higher explicit ceiling) so the warn stays a soft nudge and the raise is a real wall. Both are one-pass answers from you; then it's a trivial change.

Moot / no action

• (9) Tag-driven Cargo.toml version sync #91 mktemp+mv vs sed -i portability — shipped and working; the deviation was deliberate (GNU/BSD sed -i incompatibility). Noting for the record only; nothing to do unless you want the rationale captured in a comment on the merged PR.

What I need from you (one pass)

1. Which to activate now by label: I'd suggest small-fix on (6) and (7) (both unambiguous, bundleable into one claude/small-fixes-<date> PR), and your pick between (3a)/(3b) and (5a)/(5b)/(5c).
2. (8) — your read on the two moc_to_order questions (my recs above).
3. (2) — confirm you'll add the drafted arro3 CI YAML yourself (workflow-scope gated), or leave arro3-core CI leg for the pyarrow-free consumer was designed but never wired up #98 parked.

Marking waiting. No code/branch from me on this discuss issue — the carve-into-issues and labeling are yours.

---

Generated by Claude Code