fort14_io: reader collapses present-but-empty and absent boundary sections (empty ≡ absent) — expose presence or ratify the contract

[t-step]

Summary

• Reader collapses present-but-empty and absent

  ↪ Both fort.14 read paths reduce "the NOPE/NBOU boundary section was physically in the file but declared zero segments" and "the file had no boundary section at all" to the same empty result, so a downstream consumer cannot tell a genuinely boundaryless mesh from one whose boundary topology was stripped in its lineage.

  ▸ Call sites

    a. `src/chilmesh/CHILmesh.py:2722` catches `IndexError` on a truncated file and leaves `boundary_segments` empty.
  
        ↪ The comment reads *"Boundary section absent (legacy mesh) — leave segments empty"*, but a present `0`/`0` NOPE-NBOU block reaches the same `boundary_segments == []` state, and nothing records which happened.
  
    b. `src/chilmesh/fort14_io.py` `read_fort14_raw()` returns empty `open_boundaries` / `flow_boundaries` with no presence signal on `Fort14Raw`.
  
        ↪ The docstring states *"A legacy mesh with no boundary block is accepted"*, so both absent and present-empty yield empty lists and there is no `boundaries_present`-style field to disambiguate.
  

Why it matters

• Empty vs absent is semantically load-bearing

  ▸ Downstream symptom

    ↪ Valence #13 documented that `WNAT_Test.14` / `WNAT_Onur.14` carry `0` declared boundaries because the section was discarded in their ADmesh-regenerated lineage, indistinguishable via the parser from a legitimately boundaryless mesh, which lost Bermuda plus 112 other land segments on round-trip.
  

  ▸ Current workaround is a heuristic

    ↪ That issue proposed a `topologically_complete` classifier (detect an island-shaped elevation cluster with no boundary ring) precisely because the parser gives no authoritative present-vs-absent signal to key off.
  

  ▸ Writer half is already resolved

    ↪ The mirror concern on the write side was fixed upstream in v1.3.0 (commit `8ed8758`, cited by Valence #216): `write_fort14` now always emits the trailing NOPE/NBOU block (`0/0/0/0` when empty), so only the *reader* still conflates the two states.
  

Proposed resolution (pick one)

• Two viable contracts

  I. Expose presence

    ↪ Add a `boundaries_present: bool` (or equivalent) to `Fort14Raw` and mirror it on `CHILmesh.boundary_segments` provenance, set `True` only when a NOPE/NBOU block was physically parsed, so present-empty and absent are distinguishable.
  

  II. Ratify empty equals absent

    ↪ Explicitly document that the reader treats an empty and an absent boundary section as identical by contract, closing the ambiguity as a deliberate decision rather than an accident of `IndexError` handling.
  

References

• Related Valence issues

  a. domattioli/Valence#212 — CHILmesh-adoption epic; fort.14 read delegation (P2/chore: introspect-v2 migration incomplete on development — 28 deprecated docs/introspections records + write-target contradiction (needs operator) #214) targets read_fort14_raw, enumerates reader gaps (id columns, header count, winding, mixed arity, NBOU) but not a presence flag.

  b. domattioli/Valence#216 — writer mirror, resolved upstream in v1.3.0.

  c. domattioli/Valence#13 — downstream symptom and the topologically_complete classifier workaround.

---

Filed via Claude Code at operator request.

[claude]

• Shipped — option I (expose presence) (11e2a49, on development → rolling PR Rolling development → main #258)

  ▸ What landed

    a. new `Fort14Raw.boundaries_present: bool` — `True` only when a NOPE/NBOU block is physically parsed.
  
    b. mirrored `CHILmesh.boundaries_present` instance attr (default `False`, set in `read_from_fort14`).
  
    c. stays `False` on an absent section AND on `parse_boundaries=False` (skipped, not inspected).
  

  ▸ Why option I over II

    ↪ Exposing presence is the strictly-additive, non-lossy choice: it lets Valence #13's `topologically_complete` classifier key off an authoritative signal instead of a heuristic, whereas ratifying "empty ≡ absent" (option II) would permanently discard the distinction the downstream lineage needs.
  

  ▸ Verified

    ↪ new raw + CHILmesh-level tests (present-empty→True, absent→False, parse-skip→False); fort14 suite 16 passed, boundary-adjacent regression 47 passed / 0 failed.
  

  ▸ Writer half was already resolved upstream (v1.3.0, 8ed8758); this closes the reader gap. Closing as completed.

[model: claude-fable-5, repo: CHILmesh, session: 01UsjeLCBpMSbXMffrdkBiQe]

---

Generated by Claude Code