fix: honor explicit zero valence instead of hallucinating hydrogens

[fbaensch-beilstein]

Summary

Fixes #105. An explicit zero valence — VAL=-1 in V3000, normalized to the V2000 valence-field value 15 — was only special-cased in get_num_H when the atom had no bonds. A zero-valence atom carrying a bond fell into the literal-valence branch and received num_H = 15 - chem_bonds_valence implicit hydrogens.

Concretely, ethane with VAL=-1 on both carbons produced:

InChI=1S/C2H28/c1-2/h1-2H14

(14 "hallucinated" hydrogens per terminal carbon = 15 - 1).

Fix

In get_num_H (INCHI_BASE/src/util.c), treat valence 15 as zero valence unconditionally — no implicit H regardless of whether bonds were drawn. The isolated-atom behavior is unchanged; only the bonded case is corrected:

InChI=1S/C2/c1-2

Test

New INCHI-1-TEST/tests/test_executable/test_github_105.py, parametrized over both V3000 (VAL=-1) and V2000 (valence field 15), asserting the zero-valence ethane gives InChI=1S/C2/c1-2. Written test-first: confirmed RED on the old binary (C2H28), GREEN after the fix.

Verification

• New test: passes for both formats.
• Full CLI suite (test_executable): no unexpected failures.
• C++ unit tests (ctest): 15/15 pass.
• test_meta: pass.
• Library multithreading: no segfault; regression: 4 expected, 0 unexpected.
• Invariance: the 3 non-expected failures (Arsine, phosphabicyclo, mcule 1609394746) were confirmed pre-existing by rebuilding libinchi from baseline and reproducing them identically — not introduced here.
• No output drift: baseline vs fixed libinchi produce byte-identical InChIs across all 4190 records of both CI regression datasets. The new branch only fires for an explicit 15/VAL=-1 valence field, which real PubChem/mcule structures never carry.

Note

This implements the "honor the zero valence" resolution from the issue discussion. The stricter "reject contradictory input (bond + zero valence) as an error" option is intentionally not taken here — it would be a larger, behavior-changing diff better handled separately.

[github-actions]

Unit Test Coverage Report

Coverage Regression Summary

Metric	PR (%)	Base (%)	Difference (pp)
Lines	18.6	18.6	+0.00
Branches	12.3	12.3	+0.00

Find details on the base coverage at https://iupac-inchi.github.io/InChI/coverage/index.html

Find details on this PR's coverage by downloading coverage-reports-29074940986 and opening html/index.html

[Copilot]

Pull request overview

Fixes incorrect implicit-hydrogen assignment when an explicit zero valence is provided in molfile input (V3000 VAL=-1, normalized to V2000 valence-field 15). This aligns InChI output with the “honor zero valence” behavior discussed in issue #105 and prevents large, invalid “hallucinated hydrogen” counts for bonded atoms.

Changes:

• Update get_num_H to treat input valence 15 as zero valence unconditionally (no implicit H added even if bonds are present).
• Add a regression test covering both V3000 (VAL=-1) and V2000 (valence field 15) encodings using a minimal ethane example.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
INCHI-1-TEST/tests/test_executable/test_github_105.py	Adds regression coverage for explicit zero-valence inputs to ensure no implicit H are added.
INCHI-1-SRC/INCHI_BASE/src/util.c	Fixes implicit-H calculation to honor explicit zero valence (15) regardless of drawn bonds.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[JanCBrammer]

I can confirm that the test is red pre-fix and green post-fix (ran test locally). Good to merge as far as I'm concerned.