fix(structure): reject atom-to-self bonds in BondList and chem_comp_bond writer

[haoyu-haoyu]

Fixes #812.

A `BondList` should encode chemistry, so a bond whose two atom indices refer to the same atom is meaningless. Today both the constructor and `add_bond()` accept self-bonds silently.

Changes

• `BondList.add_bond` raises `ValueError` when the resolved positive indices are equal, including the case where one index is given positive and the other negative (e.g. `4, -1` with `atom_count=5` — both resolve to atom `4`).
• `BondList.init` performs the same check on the bulk-add path after the per-row sort, so building a `BondList` from a numpy array that contains a self-bond row raises instead of silently storing it.
• The companion concern from Bond can be created of an atom to itself #812 — that `pdbx.set_structure()` could emit a self-bond into `chem_comp_bond` when two distinct atoms in an `AtomArray` share the same `(res_name, atom_name)` annotations — is also addressed: `_set_intra_residue_bonds()` now filters bond rows where the two `atom_name` annotations are equal before deduplication, and returns `None` if no rows remain.

Tests

• `test_invalid_creation` extended with two self-bond cases (positive index pair, mixed positive/negative).
• New `test_add_self_bond_rejected` covers `add_bond()` for both index forms and verifies the list is unchanged after each rejected add.

[padix-key]

Hi, thanks for fixing the bug! Regarding the fix for bonds.pyx, this is already fixed in the BondList rewrite in the rust branch (#871), which I plan to merge for the next Biotite release. If the Rust port is not included in the next release I probably would give #829 priority here, as it was created earlier.

[padix-key]

If we have a self-bond I think it would be better to raise a BadStructureError.

[codspeed-hq]

Merging this PR will not alter performance

✅ 98 untouched benchmarks

---

_{Comparing haoyu-haoyu:fix/bondlist-no-self-bond (95eb8e4) with main (2e5ef85)}

[haoyu-haoyu]

Thanks @padix-key — agreed, BadStructureError is the right signal here. The function already raises BadStructureError for the adjacent "empty res_name" / "empty atom_name" preconditions, so raising one more for ambiguous annotations keeps the behavior consistent.

Plan for the follow-up push:

• Replace the silent not_self_bond filter in _set_intra_residue_bonds with an up-front check that raises BadStructureError("Structure contains bonded atoms sharing the same (res_name, atom_name) annotation, which cannot be written to chem_comp_bond without producing a self-bond").
• Drop the now-redundant if len(bond_array) == 0: return None guard that existed only to catch the post-filter empty case — the earlier len(bond_array) == 0 check on the unfiltered bond_array still handles the genuine "no intra bonds" path.
• Add a focused test in tests/structure/io/pdbx/ that builds an AtomArray with two distinct atoms sharing (res_name, atom_name) plus a bond between them, and asserts set_structure() raises BadStructureError.

Will push and ping when ready.

[haoyu-haoyu]

Pushed aeea708:

• _set_intra_residue_bonds() now raises BadStructureError up-front when any bond row has atom_id_1 == atom_id_2 after annotation lookup, instead of filtering the row out. Error message: "Structure contains bonded atoms sharing the same (res_name, atom_name) annotations, which cannot be written to chem_comp_bond without producing a self-bond." — mirrors the style of the existing empty-name preconditions.
• Dropped the post-filter if len(bond_array) == 0: return None guard (the pre-filter guard at the top of the function still covers the genuine "no intra bonds" path).
• Added tests/structure/io/test_pdbx.py::test_set_structure_self_bond_raises — 2-atom AtomArray with both atoms annotated (ALA, CA) plus a bond between them, asserting pdbx.set_structure() raises.

Ready for another look @padix-key.